In line with the market, TMT Investments (TMT:300¢), a venture capital company with a portfolio of more than 55 high-growth, internet-based companies, has continued to see divergence between its stronger and weaker performers.

Despite write-downs, TMT is a bargain way to own start-ups

With Europeans increasingly purchasing liquefied natural gas (LNG) from Norway, Qatar, and the United States, as well as natural gas from North African and Central Asian producers, Russia is no longer the key supplier. The composition of European oil ...

Europe's New Energy Map

As President Joe Biden’s National Security Strategy puts it, the administration views «modern industrial and innovation strategy» as the backbone of the future economy. It is an economic policy, a trade philosophy, and a political strategy focused on ...

Beyond Industrial Policy

ASTANA – The 1Fit mobile application for unlimited fitness membership, founded by a Kazakh developer, announced its launch in the United Kingdom (UK) on its LinkedIn page on Feb. 20. London residents will be able to try it out as early as this month.

Kazakh Startup 1Fit Announces Launch in UK

LONDON, Feb. 20, 2023 /PRNewswire/ -- 1Fit, developer of unlimited fitness membership app of the same name, announced its launch in the UK. Starting this February, Londoners will be the first to ...

1Fit, Central Asia's top all-sports unlimited fitness membership app, comes to the UK

LONDON, Feb. 20, 2023 /PRNewswire/ -- 1Fit, developer of unlimited fitness membership app of the same name, announced its launch in the UK. Starting this February, Londoners will be the first to try out the custom all-in-one membership to gyms, studios and ...

The global firewall as a service market size is anticipated to reach USD 12.27 billion by 2030, registering a CAGR of 22.3% from 2023 to 2030, according to a new report by Grand View Research, Inc. The surge in data breaches in the cloud

Firewall As A Service Market to be Worth $12.27 Billion By 2030: Grand View Research, Inc.

at 04:00 1Fit, Central Asia's top all-sports unlimited fitness membership app, comes to the UK 1Fit, developer of unlimited fitness membership app of the same name, announced its launch in the UK. Starting this February, Londoners will be the first to try ...

Shanghai Electric's EW8.X-230 Recognized One of the Top 10 Offshore Turbines of the Year 2022 by Wind Power Monthly

1FIT

1fit

This Data Science-based Approach Can Help Convert Users Into Paying Customers

Load Balancing Strategies for Applications: From Infrastructure to Code

44,000 startups. 4,553 cities. “4 billion votes” It was such a blessing to get to know so many of your ideas and innovations from every corner of the world in 2021. The startup community amazed all of us with your hard work and persistence despite all of the challenges that we faced globally throughout last year. Not gonna lie, the competition got our hearts racing like Eurovision. We won’t let you wait for much longer, below are long-awaited results! ✨✨✨ To see who is a winner in each respective city, visit and pick a city. The name at the top of the list is the winner. (Compromised awards are under review.) startups.hackernoon.com The will announce the winners per region in their respective cities. Startups Winners Announcement Series 🔥 Central Europe Central Europe has always caught our attention with the talented pool of people in tech community. But were they able to get through 2021 despite the pandemic? Let’s cheers to the most outstanding amongst 4,500+ startups who have been working hard this year: Basel, Switzerland An overwhelming vote for ~ in their mission of helping machinery companies drive more service revenues from their installed base MachiQ Software Berlin, Germany We have with a whooping 538 votes! Let’s support Plan A with its mission to help companies to measure, monitor and reduce their environmental footprint. Plan A Düsseldorf, Germany It doesn’t get any easier with Kicking off in helping people organize online schedules more easily. EasyWeek! Geneva, Switzerland Manage code quality? We have to give props to for their startup! SonarSource Hamburg, Germany Taking over Hamburg is , an acoustic device to reduce noises in vehicle cabins. No more annoying sounds on the road 🚗 ReCalm Köln, Germany Small business? Hoping to manage your marketing campaign? is winning the vote! rankingCoach Lausanne, Switzerland is a Swiss-based company focused on cryptocurrency management solutions built for financial institutions. Metaco Leipzig, Germany empowers companies to operate and utilize drones for the most effective industrial data applications. FlyNex Mainz, Germany is growing sustainable digital mobility startups! Next Mobility Labs Munich, Germany wins for top-line, high-security international Identity Management! ORY Stuttgart, Germany Remote desktop application? Ask anywhere. AnyDesk Vienna, Austria , an IT company enabling privacy-preserving big data applications in the digital age. Mostly AI Zug, Switzerland Way to go 1762 votes for the first provably live pure Proof-of-Stake (PoS) CBC-Casper blockchain designed for real-world applications. CasperLabs! Nyon Calling for robust data solutions for impact-driven organizations! Havuta LLC Zürich is revolutionizing small business lending to give easy access to capital to those who need it most: SMEs ✨ Teylor Heidelberg, Germany is an innovative HR startup providing a solution for fast and efficient recruiting of employees with simple job profiles. Talk'n'Job Stuttgart, Germany Congratulations for winning this title! AnyDesk Moving on to our buddies in... 🔥 Eastern Europe Eastern Europe is home to many familiar big names in our start-up world. Trace history for Miro, Telegram, and Grammarly and you know where they are from. Who is the next generation for Eastern European startups? Budapest, Hungary is a platform for mobile app development, enabling tens of thousands of developers to automate integration, testing, and deployment. Bitrise I__ __ stanbul, Turkey Leading here is - a brand new software intelligence platform- helps organizations deliver faster, high-quality products by connecting their DevOps tools. Oobeya Kraków, Poland is a human cloud platform that connects global enterprises with European software houses, offering tech talent as a service. Talent Alpha Kyiv, Ukraine 281 votes for , an AI powered app where people can swap faces in the videos, GIFs and images in just a few seconds. Reface Sofia, Bulgaria 387 votes for __ __378, leading Blockchain and DLT solutions company that specializes in helping enterprises and blockchain startups design, build and scale innovative solutions using Blockchain & DLT! LimeChain Warsaw, Poland is a next-generation blockchain document management software (DMS). DoxyChain Yerevan, Armenia Whoever clicked to vote for was sure a vote, because it ended up winning as a full-service software provider delivering Artificial Intelligence & Machine Learning solutions for businesses. SmartClick smart Almaty Just like the name suggested, provides one pass to dozens of gyms and studios. 1FIT Debrecen got all the coders ranked figured out! CodersRank Kharkiv is complete solution digital assets management! Pics.io Minsk, Belarus with 243 votes is the first blockchain-powered sports media platform. Scorum Moscow, Russian Federation 1 e-grocery company in Russia with 26 votes and #1 on our HackerNoon board for Minsk. SberMarket Saint Petersburg, Russian Federation leads this area as a personal finance application that rewards responsible spending and savings. CoinKeeper Kazan, Russian Federation Smelling a sweet victory? 3d App designers for flower bouquets is in the house! Flowito Western Europe 🔥 Brexit? Pandemic? It seems like startups in Western Europe still keep their spirits regardless. Startups in Western Europe that stood out in votes this year go to: Antwerp is on a streak! As a bug bounty platform, it ensures online security through a crowd-sourced platform for computer experts. intigriti Belfast A record-breaking 10,003 votes? is an independent coffee shop. Come grab a coffee! Bobs & Berts Birmingham What’s happening in Birmingham? We have , an innovative UK-based advanced lithium battery developer aiming to accelerate the global shift to cleaner. Aceleron Bordeaux 30 is an open-source web framework for building modern apps and websites with Vue.js. NuxtJS Boulogne-billancourt is a software company that provides an in-store analytics platform to optimize operations. DiGEiZ Brighton In for some jam? offers musical education services. BIMM Bristol Another tasty victory for Western Europe is enabling the strongest digital encryption on the internet. KETS Quantum Security Brussels A fashionable win for , a social fashion network. You scan barcodes in fashion shops and see every influencer wearing those outfits! Share your wishlist with your friends and earn cashback. SCANDALOOK Cardiff is a technology company developing end-to-end digital connectivity for the point-of-care diagnostics market. Bond Digital Health London A whooping number of 1085 vote__ __ offers hardware based cyber-security solution for authentication and encryption in a wired local network. Crypto Quantique Edinburgh Hopping up on the crypto movement, just launched last year but they are already rocking it at making crypto accessible to everyone. Zumo Ghent 5379 votes for ! A Belgian InsurTech startup that helps insurance providers all over the world to better connect with their digital customers. WeGroup Glasgow specializes in single-cell analysis transforming the life sciences. Biomage Paris Show some love in the city of love for , a sports platform that is connected to a mobile sports coaching application. AirFit Aberystwyth enables artificial intelligence, data science & digital innovation solutions for the agriculture, biotech & life sciences industries. Agxio Norwich Congratulations for the amazing win of 1124 votes! WATR Charleroi Congratulations for your amazing win! Hirering Gibraltar Give a round of applause to Fortknoxter! Waltham Abbey is ruling the area with its amazing 72 votes! RPAIR Antwerpen Rocking up Antwerpen area is the rising startup ! Willow Egham What’s happening in Egham? Proudly announcing winning the Noonies this year! ATA.ONE Leicester Leicester is calling the name of as the winner for the region! DeadHappy Macclesfield Who else but ? Congratulations on dominating the award! AeroCloud Northern Europe 🔥 Want to listen to some music? Spotify. Need some games to kill time? Candy Crush. What about a quick call to someone you haven’t seen in a long time? Skype. These are all startups that originated from Northern Europe! Noonies, let’s take a look at the upcoming generation of startups from Northern Europe that hopefully will see an even bigger and brighter future. Gothenburg, Sweden Shipping was never made for small shippers, so took on the mission to built an end-to-end logistics platform to empower small businesses to spend less time and money on shipping and more time on growing their businesses. Sendify The Netherlands provides a virtual platform for many kinds of conferences and events. Let's Get Digital Helsinki, Finland A curated collection of public basketball courts in the Helsinki Metropolitan Area called . Everyone is welcome to submit to their local court. Koripallopaikat Lund, Sweden develops an antibiotic-free, safe and effective treatment of vaginal fungal infections. Gedea Biotech Luxembourg, Luxembourg 🚀 is a job-matching platform that connects developers and tech experts with CTOs and team leads from the best tech companies across Europe. 👾 nexten.io Malmö, Sweden Feeling studious, Noonies? __ __e is a tool for Google Classroom, that enables formative matrix assessment towards course learning objectives directly. StudyBe Oslo, Norway is the platform supports a unified API across all connected device categories, removing significant friction for app developers. Domos Labs Reykjavík, Iceland __ __s is a mobile-first technologically advanced neo-bank. Indo service #S__ __ tockholm, Sweden is the biggest blockchain start-up in Sweden. Double winner of the most prestigious nomination in Nordics 'NyTeknik 33-listan', winner of the Deloitte Fast 50, and now Noonies Award! ChromaWay AB T__ __ allinn, Estonia Passing 1000 votes is a fully digital and decentralized ecosystem for all actors in supply chain. OBORTECH Vilnius, Lithuania is an easy-to-use money app for cards and accounts. Via Genome, users can open a business or a merchant account or a personal one with dedicated IBANs, access transfers, exchange currency, process payments, and manage the funds. Genome Aalborg Digura A legaltech offering tenants fast, digital and reliable legal assistance Aarhus Partisia Market Design provides secure, simple, efficient solutions for auctions and exchanges realizing the potential of the market. Helsingborg is a hi-tech company developing cardiac occlusion devices and other vascular implants. Occlutech Tallinna is an AI-powered system for marketing automation in social networks. POSTOPLAN 🔥 Southern Europe Southern Europe’s economy went through a hard time this year due to a huge reduction in tourism. But our startups are of course always headstrong, from Spain to Italy and all the way through Greece. Who are the heroes rising amidst changes of Southern Europe? Skopje is a versatile software development company that solves the business challenges of SMBs, big brands, and brave hearts with awesome ideas. CodeChem Sliema Sounds like just a relaxing exhale. is a holistic health and wellbeing app built to allow users work on the development of their physical, mental and spiritual health Few Tajonar is a company of experienced business & industry entrepreneurs offering solutions to industries based on cutting-edge technologies. das-Nano Tarragona is a technology-based company focusing on the production of integrated microsystems. IMICROQ Terni is an ICT-based integrated solution that is able to modulate, respond, assist and learn to control most home activities. Mind Tirana is a video content and e-commerce platform for the Balkans built on top of an Albanian language specialized search engine. Gjirafa Torino develops and commercializes algorithms based on Artificial Intelligence to increase the autonomy of small satellite missions. AIKO T__ __ orres Buy online groceries like you would buy offline with personalized picking as fast as in 4 hours with Vedras 360hyper! Trento is a music-based platform built around streaming platforms that allow users to discover new emerging artists and make new friends. Brots 2021, what a YEAR. Congratulations to all of the startups that made it to the top winning list for Europe! Stay tuned for Startup awards in 2022 and let’s cheers to a new year of more innovative ideas, bold innovations, and robust creations from you, Europe 🌷 More Information: If you’re a winner, please answer ! If you’re a reader, participant, just a curious individual, consider answering one of our interviews . this interview here Read the FAQs about or . Startups here here To see all of the winners you may visit , pick any city and view the name that appears at the top. this site If you’re interested in partnering with us, we invite you to look through some of these or reach out to our business department. unique sponsorship options Perhaps creating your very own tickles your fancy or maybe you have a whole new idea for how we can partner! writing contest Thank you to all of our sponsors! We truly appreciate your contribution to the tech world and thus the future! The Startups Awards are sponsored by: , , , , , , , , , , Blockster Brex FreeCodeCamp LaunchDarkly Coil Udacity Udemy Vercel Asia Pacific Transgender Network Bybit Everscale Network This post was written by An Le and edited by Natasha Nel. If you have any questions or concerns or would like to report a discrepancy please email ellen@hackernoon.com.

Meet the Winners of Startups in Europe

That's likely a question that you've never asked yourself, but as a manager, it's an important one. Career conversations are an important part of We’re going to define 5 levels that will make it easier for you to have a better career conversation. Through self-reflection or 360 feedback, it should be relatively easy to know where you stand. You can then set the next highest level as a tangible goal to improve. career development. Each level builds on top of the previous one. Let’s dive in. Level 0 Description: Manager never initiates any conversations with employees about their careers or development. If the employee brings up a career conversation the manager is not helpful. They’re either dismissive, unsupportive, ignorant, or don’t offer any insights. Grade: F You don’t want to be this manager! The good news is that it’s hard to be at this level. The most likely explanation is that you don’t believe in supporting your employees’ careers and development. You may believe that it’s not your job or that it’s not in the best interest of the organization. If so, I think you’re mistaken, though I would be curious to understand that point of view better. Here are some of the potential risks of not supporting them: Decreased engagement Decreased motivation Decreased innovation and creativity Lack of trust and goodwill Tips for improvement: Be prepared to answer the following questions from your employees: How do promotions work? What are the different career paths? What is my next role, and what does it take to get there? What should I be working on to progress in my career? Discuss this with your manager, HR, and peers. You want your answers to be accurate and consistent with the rest of the department. Level 1 Description: Manager doesn’t initiate conversations about careers and development. Manager is helpful when the employee initiates a career conversation. Grade: C This is better than nothing. At Level 1 you may think that you don’t have time to support their careers and that it’s their responsibility. You’re at least providing value to them. If you’re good at responding when they initiate a career conversation, you may even add a lot of value. You can do much better though. Here are some of the downsides of this level: It’s not equitable. The people that are comfortable and skilled at initiating career conversations will benefit more than those who aren’t. It’s hard to know what you don’t know. The employee can only bring up career topics and options that they’re familiar with. This could leave some valuable conversations out. The employee is making all the effort which is not as good at building trust and goodwill. If there’s little trust the employee is less likely to share their career goals. Tips for improvement: Schedule multiple times throughout the year to have career conversations. Prepare for each conversation. Make sure you understand career progressions in your department well. Anticipate common questions. Level 2 Description: Manager initiates some career conversations, usually at review time. Manager discusses the employee’s career options in the department (e.g. career ladder and/or different tracks). Manager discusses the employee’s next role, where they stand, and how to get there. Grade: B This is pretty good. You could consider this your target level or your minimum requirement. At Level 2 you're taking an active role in the career conversations. Your focus is helping employees progress their careers at your company. This is reasonable and will provide solid value to the employee. You can still do better! Here are some of the downsides of this level: The default progressions in your department may be limiting to some employees. Focusing on the next role may feel transactional and less inspiring. The focus is on the benefit to the company which is sub-optimal for building trust and goodwill. Tips for improvement: Expand the scope of your career conversations. Get curious about them as individuals and about their career goals: What are their personal goals? How much is career growth important to them? What kind of growth are they looking for right now? What are they passionate about? What do they want to learn? Expand the career options that you’re comfortable discussing beyond the next role in your department. Talk about career and growth frequently in your 1-1s. Experiment with a variety of different conversations and questions. Level 3 Description: Manager discusses the employee’s Career with a capital C. Manager takes a more holistic view of their career unconstrained by the roles available in the department, or even the company. Career conversations explore short-term and long-term goals. Conversations occur more frequently throughout the year. Employees trust the manager enough to shares their true career goals. Grade: A- This is a great target level for all managers. The big difference at this level is that for the first time we’re putting the goals of the individual on par with the goals of the company. We’ve made the transition to in the belief that the best way to serve the company is to serve the company the individual. We're proactively supporting the individual. This not only builds trust and goodwill; it also adds a lot more value to them! If they trust you, they are more likely to tell you what they really want. This gives you a much better chance to find them work they love. It’s not crazy to say that if you do this well you could have a profound effect on someone’s career, maybe even their life. servant leadership and Tips for improvement: At this level improvement becomes harder. Progressing through the earlier levels is mostly a question of taking the time to perform various actions. To progress beyond Level 3 you have to view career conversations as a and get serious about working on it. skill Read Radical Candor Three Powerful Conversations Managers Must Have to Develop Their People Three Questions Every Manager Struggles with Making Career Development Plans Different 1-1 question banks , , and here here here Practice Have career conversations often with different people and reflect. What went well, what could have gone better? Have mock conversations with peers or a coach. Work on your . listening skills Use mindfulness training to improve your . emotional intelligence Get feedback Ask your team members for feedback. What was helpful to them? What wasn’t? Use an and pay attention to the scores related to career support. upward feedback survey Get feedback from a coach, your manager, or HR. Get a coach or a mentor Level 4 Description: Manager inspires the employee. Manager motivates the employee. Manager expands the employee’s realm of career possibilities. Manager sees potential that the employee doesn’t see in themselves. Grade: A+ The Picasso of career conversations! This is Level 3 executed at an expert level. The other levels are attainable through effort and knowledge. This one is not. This level requires superb emotional intelligence, listening, and job crafting skills. You need to have a great feel for asking insightful questions. You have to know when and how to push people out of their comfort zones and when to reassure and build up their confidence. You have the humility to coach people to find their own path instead of imposing your vision. This level requires the skills and the dedication to apply them, one good career conversation at a time. This is really hard to do consistently. Which is exactly what you want for an expert level! Final thoughts These levels are not perfect They’re not going to be neat and precise, and they don’t capture all the subtleties of career conversations. You may find that you don’t fit nicely in any of them. You may reach a higher level with one employee versus another. As imperfect as they are, they can be helpful The benefit of these levels is using them for a quick self-assessment: . . What level are you at with each of your team members? What would you have to change to move up one level? If you realize that you’re not at the level you want to be, don’t be too hard on yourself. You have a lot on your plate and chances are you’ve had little to no training on career conversations. Being a better career conversationalist is a great way to add value to each of your team members. Each career conversation is an opportunity to build trust, increase engagement, and job satisfaction. If you can have them often and skillfully your team will be more likely to stay, perform at a high level, and be happy. Challenge yourself to get to the next level!

Have You Hacked Level 4 of Career Conversations Yet?

Predicting the Attrition of Valuable Employees….. In an IT firm, there are many Employee Architectures available. Some IT firms or at particular departments or certain levels follow the structure, in which there is a “star” organisation around a “chief” position designated to the Engineer who best understands the system requirements. chief programmer While, some follow an ( ) structure, where all the Engineers are at the same level designated for different jobs like Front-End Design, Back-End Coding, Software Testing etc. But, this architecture is not followed by very big or Multi-National Software Giants. But all in all, this is a very successful and working environment-friendly architecture. egoless democratic Egoless (Democratic) Architecture 3rd Type of architecture is the structure, which is the combination of the above 2 types. This is the mostly followed architecture and very common among software giants. mixed Mixed Controlled Architecture Likewise, International Business Machine Corporation (IBM) probably follows either or structures. So, for the HR Department, an important responsibility is to measure the attrition of the Employees at specific time-gaps. The factors on which the Employee Attrition depends upon are: egoless mixed Age of the Employee Monthly Income Overtime Monthly Rate Distance from Home Years at Company and so on… IBM also made their Employee Information publicly available, with the problem statement: “ Predict the Attrition of the Employees i.e., will there be attrition of the employees or not, given the Employee Details i.e., the factors responsible for attrition” The Employee Dataset is made available at Kaggle: _Predict attrition of your valuable employees_www.kaggle.com IBM HR Analytics Employee Attrition & Performance A possible solution to solve this problem is by applying Machine Learning i.e., by imparting Machine Intelligence which involves development of a Predictive Model by training it, using the data available and validating it for Model Performance Analysis…. Given below is a step-by-step procedure of Machine Learning Model Development using Python and Scikit-Learn Machine Learning Toolbox: Model Development: # import numpy as npimport pandas as pdimport matplotlib.pyplot as plt%matplotlib inlineimport pylab as plfrom sklearn.metrics import roc_curve, auc importing all the libraries #**loading the dataset using Pandas**df = pd.read_csv('WA_Fn-UseC_-HR-Employee-Attrition.csv')df.head()# Output shown below Pandas Dataframe Output of the Dataset # df.shape == df.dropna().shape # Output shown below checking whether the dataset contains any missing values... Hence, there are no missing values present in the dataset. This is a Binary Classification Problem, so the Distribution of instances among the 2 classes, is visualized below: y_bar = np.array([df[df['Attrition'] 'Yes'].shape[0]])x_bar = ['No (0)', 'Yes (1)' 'No'].shape[0],df[df['Attrition'] # plt.bar(x, y)plt.xlabel('Labels/Classes')plt.ylabel('Number of Instances')plt.title('Distribution of Labels/Classes in the Dataset')# Output shown below Bar Visualization Bar Visualization of the Class Distribution # X = df.iloc[:,[0] + list(range(2,35))].valuesy = df.iloc[:,1].valuesfrom sklearn.preprocessing import LabelEncoder, OneHotEncoder Label Encoding for Categorical/Non-Numeric Data labelencoder_X_1 = LabelEncoder()X[:,1] = labelencoder_X_1.fit_transform(X[:,1])X[:,3] = labelencoder_X_1.fit_transform(X[:,3])X[:,6] = labelencoder_X_1.fit_transform(X[:,6])X[:,10] = labelencoder_X_1.fit_transform(X[:,10])X[:,14] = labelencoder_X_1.fit_transform(X[:,14])X[:,16] = labelencoder_X_1.fit_transform(X[:,16])X[:,20] = labelencoder_X_1.fit_transform(X[:,20])X[:,21] = labelencoder_X_1.fit_transform(X[:,21])y = labelencoder_X_1.fit_transform(y) #**Feature Selection using Random Forest Classifier's Feature**#**Importance Scores**from sklearn.ensemble import RandomForestClassifiermodel = RandomForestClassifier()model.fit(X,y) # Output shown below list_importances=list(model.feature_importances_)indices=sorted(range(len(list_importances)), key=lambda k:list_importances[k])feature_selected=[None]*34k=0for i in reversed(indices):if k<=33:feature_selected[k]=ik=k+1X_selected = X[:,feature_selected[:17]]l_features=feature_selectedi=0for x in feature_selected:l_features[i] = df.columns[x]i=i+1l_features = np.array(l_features) #**Extracting 17 most important features among 34 features**l_features[:17] #Output shown below #**Selecting the 17 most important features**df_features = pd.DataFrame(X_selected, columns=['Age','MonthlyIncome', 'OverTime','EmployeeNumber', 'MonthlyRate',,                       'DistanceFromHome', 'YearsAtCompany','TotalWorkingYears', 'DailyRate','HourlyRate', 'NumCompaniesWorked','JobInvolvement', 'PercentSalaryHike','StockOptionLevel','YearsWithCurrManager','EnvironmentSatisfaction','EducationField', 'Attrition']]df_selected.head() # Output shown below So, again label encoding has to be done for the selected categorical features: #**Label Encoding for selected Non-Numeric Features:**X = df_selected.iloc[:,list(range(0,17))].valuesy = df_selected.iloc[:,17].values X[:,2] = labelencoder_X_1.fit_transform(X[:,2])X[:,16] = labelencoder_X_1.fit_transform(X[:,16])y = labelencoder_X_1.fit_transform(y) Now the Data Pre-Processing Steps are over. Let’s move on to Model Training:- from sklearn.model_selection import train_test_splitX_train, X_test, y_train, y_test = train_test_split(X, y,test_size=0.2, random_state=1753) #80-20 splitting where 80% Data is for Training the Model#and 20% Data is for Validation and Performance Analysis #**Using Logistic Regression Algorithm for Model Training**from sklearn.linear_model import LogisticRegressionclf= LogisticRegression(verbose = 3) #**Training the Model**clf_trained = clf.fit(X_train, y_train) #Output shown below This is the Library of Parameter Optimization Strategy used by Logistic Regression 2. Model Performance Analysis: => Training Accuracy clf_trained.score(X_train,y_train) # Output shown below Training Accuracy of 84.44% is achieved by the model => Validation Accuracy clf_trained.score(X_test,y_test) # Output shown below Validation Accuracy of 89.12% is achieved by the model => , and Precision Recall F1-Score # predictions=clf_trained.predict(X_test) getting the predictions... print(classification_report(y_test,predictions)) Classification Report of the model => Confusion Matrix # MODULE FOR CONFUSION MATRIX import matplotlib.pyplot as plt%matplotlib inlineimport numpy as npimport itertoolsdef plot_confusion_matrix(cm, classes,normalize=False,title='Confusion matrix',cmap=plt.cm.Blues):"""This function prints and plots the confusion matrix.Normalization can be applied by setting `normalize=True`."""if normalize:cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]print("Normalized confusion matrix")else:print('Confusion matrix, without normalization') print(cm) plt.imshow(cm, interpolation='nearest', cmap=cmap)plt.title(title)plt.colorbar()tick_marks = np.arange(len(classes))plt.xticks(tick_marks, classes, rotation=45)plt.yticks(tick_marks, classes) fmt = '.2f' if normalize else 'd'thresh = cm.max() / 2.for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):plt.text(j, i, format(cm[i, j], fmt),horizontalalignment="center",color="white" if cm[i, j] > thresh else "black") plt.tight\_layout()  
plt.ylabel('True label')  
plt.xlabel('Predicted label') #**Generating the Confusion Matrix**plt.figure() cm = np.array([[252, 1], [31, 10]]) plot_confusion_matrix(confusion_matrix(y_test,predictions),classes=[0,1], normalize=True, title='Normalized Confusion Matrix')# Output shown below Normalized Confusion Matrix => Receiver Operator Characteristic Curve: # y_roc = np.array(y_test)fpr, tpr, thresholds = roc_curve(y_roc, clf_trained.decision_function(X_test))roc_auc = auc(fpr, tpr)pl.clf()pl.plot(fpr, tpr, label='ROC curve (area = %0.2f)' % roc_auc)pl.plot([0, 1], [0, 1], 'k--')pl.xlim([0.0, 1.0])pl.ylim([0.0, 1.0])pl.xlabel('False Positive Rate')pl.ylabel('True Positive Rate')pl.legend(loc="lower right")pl.show() # Output shown below Plotting the ROC Curve Receiver Operator Characteristic Curve (ROC Curve) According to the Performance Analysis, it can be concluded that the Machine Learning Predictive Model has been successful in effectively classifying 89.12% unknown (Validation Set) examples correctly and has shown quite descent statistical figures for different performance metrics. Hence, in this way an Employee Attrition Predictive Model can be developed using Data Analysis and Machine Learning. This model has been deployed in a Web Application by me using php ( ) as back-end with the help of . The link to the Web-App is given below: PHP: Hypertext Preprocessor PHP-ML navocommerce.in IBM-HR-ANALYTICS For Personal Contacts regarding the article or the Web-App or discussions on Machine Learning or any department of Data Science, feel free to reach out to me on LinkedIn . _View Navoneel Chakrabarty's profile on LinkedIn, the world's largest professional community. Navoneel has 3 jobs listed…_www.linkedin.com Navoneel Chakrabarty - Founder - Road To Financial Data Science | LinkedIn

A Machine Learning Approach to IBM Employee Attrition and Performance

When I first came across , I liked how easy it was to get started and that it’s built with React. I write React code on a daily basis and why not use it to create presentations? All went well for a while until I realized that the file started to exceed a few hundred lines. Spectacle ./presentation/index.js There must be a way to modularize my slides! The code is available at tl;dr: https://github.com/mikenikles/presentations/tree/master/packages/blog-post-source Get Started With `spectacle-boilerplate` As recommended in the Spectacle Github repo, I started by cloning the repo and removed the folder: spectacle-boilerplate .git $ git clone git@github.com:FormidableLabs/spectacle-boilerplate.git blog-post-source$ cd blog-post-source$ rm -fr .git At this point, you can install dependencies with and start the presentation with . It will be available at npm install npm start http://localhost:3000. Create a Folder For Each Slide There are four slides in the boilerplate presentation. Let’s create a directory where we’ll move each slide in its own subfolder. ./presentation/slides $ cd ./presentation$ mkdir slides && cd slides$ mkdir 1 && mkdir 2 && mkdir 3 && mkdir 4 We also want a file in each folder. This is where each slide’s content is going to be. index.js $ touch ./1/index.js && touch ./2/index.js && touch ./3/index.js && touch ./4/index.js This leaves us with the following directory structure: : Commit https://github.com/mikenikles/presentations/commit/5d208f669c633da7d95424d48e62588081bd7d56 Populate Each Slide’s Content The spectacle-boilerplate repo already provides each slide’s content in . All we need to do is move each React component into its corresponding file. ./presentation/index.js <Slide /> ./presentation/slides/[slide-number]/index.js Let’s do that together for the first slide. Cut and paste that to ./presentation/slides/1/index.js We also have to add a few import statements to . Also, let’s make sure we export the code for this slide. The final file looks like this: ./presentation/slides/1/index.js import React from "react";import { Heading, Slide, Text } from "spectacle"; export default (<Slide transition={["zoom"]} bgColor="primary"><Heading size={1} fit caps lineHeight={1} textColor="secondary">Spectacle Boilerplate</Heading><Text margin="10px 0 0" textColor="tertiary" size={1} fit bold>open the presentation/index.js file to get started</Text></Slide>); Follow the same steps for the remaining slides. : Commit https://github.com/mikenikles/presentations/commit/a45f144247d0e4f0c39d922a9a23cf73c05c0a32 Load Slides Dynamically Lastly, we have to load each slide dynamically. This sounds trickier than it is. At a high-level, the following steps are required: Load all slides dynamically with . import() Provide the loaded slides to the component’s state. Presentation Populate the state with placeholder components while step 1 and 2 above are in progress. Provide a unique prop to each dynamically loaded slide. key 1. Load all slides dynamically with import() In , we define a list of all slides and their order. ./presentation/index.js const slidesImports = [import("./slides/1"),import("./slides/2"),import("./slides/3"),import("./slides/4")]; Each statement returns a promise. So is an array of Promises. We can leverage that and use the to wait until all slides have been imported. More on that shortly. import() slidesImports Promise.all() function 2. Provide the loaded slides to the component’s state. Presentation The component needs a where we provide the loaded slides once they’re available. We populate an empty array in the and replace it with the actual slides’ content in the lifecycle method. The new component now looks like this: Presentation state constructor() componentDidMount() Presentation export default class Presentation extends React.Component {constructor(props) {super(props); this.state = {  
  slides: \[\] // A placeholder for slides once they're loaded.  
}; } componentDidMount() {const importedSlides = [];Promise.all(slidesImports).then((slidesImportsResolved) => {slidesImportsResolved.forEach((slide) => {importedSlides.push(slide.default);});this.setState({ slides: importedSlides });});} render() {return (<Deck transition={["zoom", "slide"]} transitionDuration={500} theme={theme}></Deck>);}} We’re almost done. Next up, let’s update the function and actually render all slides. render() render() {const { slides } = this.state;return (<Deck transition={["zoom", "slide"]} transitionDuration={500} theme={theme}>{slides.map((slide) => {return slide;})}</Deck>);} When we look at , we see a blank screen and the following error in the browser console: http://localhost:3000/ Browser console error based on the current code A closer look at on line 415 reveals the error is caused by the following line of code: manager.js children: _react.Children.toArray(child.props.children), Based on the error message, we know that is . That’s an easy fix. child undefined 3. Populate the state with placeholder components while step 1 and 2 above are in progress When the component’s function is called for the first time, is set to an empty array. Spectacle doesn’t like that, so let’s provide some placeholder slides until our real slides are imported and added to the state. Presentation render() this.state.slides We could provide an empty slide until is available, along the lines of: this.state.slides render() {const { slides } = this.state;return (<Deck transition={["zoom", "slide"]} transitionDuration={500} theme={theme}>{slides.length ? slides.map((slide) => {return slide;}) : <Slide />}</Deck>);} That actually works fine when we load the first slide at . However, try to navigate to the second slide and reload the page at . . http://localhost:3000/ http://localhost:3000/#/1 Error What this teaches us is that Spectacle needs to know the exact number of slides a presentation requires upfront upon first calling the function of the component. render() Presentation Easy, let’s make it happen by changing the function from: constructor() constructor(props) {super(props); this.state = {slides: []};} to: constructor(props) {super(props); this.state = {slides: Array(slidesImports.length).fill(<Slide key="loading" />)};} We basically populate the property of the state with the exact number of slides that we’ll import. In the code snippet above, we render an empty component, but we could just as well design a nice slide that displays a “Loading…” spinner. slides <Slide /> Now head back to and enjoy the second slide of your presentation without the nasty error we saw earlier. http://localhost:3000/#/1 Wait a minute, I spoke too soon… Each <Slide /> component requires a unique “key” prop Oh yeah, that’s right. in the error message explains why that prop is important. The link key 4. Provide a unique prop to each dynamically loaded slide key We have two options to do that: Add a prop to each component within . key <Slide /> ./presentation/slides/[slide-number]/index.js Provide the prop dynamically in the component’s function. key Presentation render() Option 1 sounds simple, but we’ll go for option 2 because it makes it easier to rearrange slides later. If the individual slide files are unaware of their position within the presentation, we can simply rename the slide’s folder from to to move a slide from the first to the third position in the presentation. index.js 1 3 The function’s updated component now looks like this: render() <Deck /> <Deck transition={["zoom", "slide"]} transitionDuration={500} theme={theme}>{slides.map((slide, index) => {return React.cloneElement(slide, {key: index});})}</Deck> : Commit https://github.com/mikenikles/presentations/commit/2c8630086548405e7d7ac2394d087fcfe504b06c Conclusion With this approach, I can now easily modularize my Spectacle presentations. In fact, I can take this to a whole new level… I could create a NPM module with a collection of commonly used slides, such as an “About Me” slide I use for every presentation. Whenever I want to use that in a presentation, I could simply add it as a dependency to my file and import it into my presentation at the correct index in my deck. package.json If you have any questions, don’t hesitate to reach out!

Presentations With Spectacle — How I Modularize My Deck

It is indeed a total blessing to wake up this Sunday morning to the sound of the wind blowing through the trees. Lastight I watched the God Bless Bitcoin full documentary for the 3rd time. I fell asleep in the middle of the movie, while diving deep into Bitcoin's proof-of-work system. I woke up and stumbled upon a Bible verse from Proverbs 23:23 on my screen. Here's what hit me like a bolt of lightning: in the Passion Translation, which tells us to "Embrace the truth and hold it close. Don't let go of wisdom, instruction, and life-giving understanding. The Hebrew word for "buy" in this context means "acquire" – suggesting that real truth, genuine wisdom, requires effort. Sound familiar? It should, because this is exactly what Bitcoin's proof-of-work system is all about. Think about it. When Satoshi Nakamoto gave us Bitcoin, he/she/they didn't just create another digital currency. They built a system where you can't just buy your way into consensus – you have to prove your commitment through actual work. It's like they took this ancient biblical principle and coded it into the very DNA of Bitcoin. I've written before about how proof-of-work is reshaping everything from presidential politics to governance in places like the Kingdom Tonga. But seeing it through this biblical lens adds a whole new dimension. When we talk about miners expending computational power to validate transactions, they're not just securing a network – they're participating in a ritual as old as wisdom itself: proving their commitment to truth through effort. This hits especially hard when you look at what's happening in the political sphere. When 17% of U.S. adults hold crypto, they're not just investing in a new asset class. They're putting skin in the game, proving their dedication to financial self-determination through tangible effort. Every satoshi they hold represents proof of their commitment to this new truth. Remember how Satoshi disappeared without converting his/her/their early-mined coins? That selfless act makes even more sense now. They understood what the ancient wisdom was telling us: truth isn't something to be hoarded or sold – it's something to be earned and shared. And here's where it gets really interesting for the future of governance. In my work with Tonga's decentralized governance model, we're essentially creating a system where citizens become "truth miners" in their own community. Every time they participate in shaping their local policies, they're not just earning "govTokens" – they're proving their stake in the community's shared truth through actual effort. The parallels are striking. Just as Bitcoin's network becomes stronger with each block mined, each piece of proof-of-work added, a community becomes stronger when its members actively participate in creating and validating its shared truths. You can't buy this kind of legitimacy – you have to earn it. In a world where we're increasingly questioning what's real and what's not, where trust in traditional institutions is eroding, maybe we need this ancient wisdom more than ever. The idea that truth must be earned through effort, not bought with money, feels revolutionary. But it's not new – it's as old as wisdom itself. Bitcoin's proof-of-work system isn't just clever cryptography. It's a digital implementation of a timeless principle: that truth requires effort, that wisdom must be earned, that real value comes from genuine commitment. In a way, every Bitcoin miner is participating in a tradition that goes back thousands of years – proving their dedication to truth through tangible work. As we build this new digital future, maybe we should pay more attention to this ancient wisdom. After all, whether we're talking about spiritual truth or digital consensus, the principle remains the same: you can't buy your way to legitimacy. You have to prove it, one block at a time, one effort at a time. The future of human organization might be digital, but the principles that make it work? Those are timeless. It is indeed a total blessing to wake up this Sunday morning to the sound of the wind blowing through the trees. Lastight I watched the God Bless Bitcoin full documentary for the 3rd time. I fell asleep in the middle of the movie, while diving deep into Bitcoin's proof-of-work system. I woke up and stumbled upon a Bible verse from Proverbs 23:23 on my screen. God Bless Bitcoin Here's what hit me like a bolt of lightning: in the Passion Translation, which tells us to "Embrace the truth and hold it close. Don't let go of wisdom, instruction, and life-giving understanding. The Hebrew word for "buy" in this context means "acquire" – suggesting that real truth, genuine wisdom, requires effort. Sound familiar? It should, because this is exactly what Bitcoin's proof-of-work system is all about. Think about it. When Satoshi Nakamoto gave us Bitcoin, he/she/they didn't just create another digital currency. They built a system where you can't just buy your way into consensus – you have to prove your commitment through actual work. It's like they took this ancient biblical principle and coded it into the very DNA of Bitcoin. I've written before about how proof-of-work is reshaping everything from presidential politics to governance in places like the Kingdom Tonga. But seeing it through this biblical lens adds a whole new dimension. When we talk about miners expending computational power to validate transactions, they're not just securing a network – they're participating in a ritual as old as wisdom itself: proving their commitment to truth through effort. This hits especially hard when you look at what's happening in the political sphere. When 17% of U.S. adults hold crypto, they're not just investing in a new asset class. They're putting skin in the game, proving their dedication to financial self-determination through tangible effort. Every satoshi they hold represents proof of their commitment to this new truth. Remember how Satoshi disappeared without converting his/her/their early-mined coins? That selfless act makes even more sense now. They understood what the ancient wisdom was telling us: truth isn't something to be hoarded or sold – it's something to be earned and shared. And here's where it gets really interesting for the future of governance. In my work with Tonga's decentralized governance model, we're essentially creating a system where citizens become "truth miners" in their own community. Every time they participate in shaping their local policies, they're not just earning "govTokens" – they're proving their stake in the community's shared truth through actual effort. The parallels are striking. Just as Bitcoin's network becomes stronger with each block mined, each piece of proof-of-work added, a community becomes stronger when its members actively participate in creating and validating its shared truths. You can't buy this kind of legitimacy – you have to earn it. In a world where we're increasingly questioning what's real and what's not, where trust in traditional institutions is eroding, maybe we need this ancient wisdom more than ever. The idea that truth must be earned through effort, not bought with money, feels revolutionary. But it's not new – it's as old as wisdom itself. Bitcoin's proof-of-work system isn't just clever cryptography. It's a digital implementation of a timeless principle: that truth requires effort, that wisdom must be earned, that real value comes from genuine commitment. In a way, every Bitcoin miner is participating in a tradition that goes back thousands of years – proving their dedication to truth through tangible work. As we build this new digital future, maybe we should pay more attention to this ancient wisdom. After all, whether we're talking about spiritual truth or digital consensus, the principle remains the same: you can't buy your way to legitimacy. You have to prove it, one block at a time, one effort at a time. The future of human organization might be digital, but the principles that make it work? Those are timeless.

God, Bitcoin, and Proof-of-Work: How Ancient Wisdom Powers the Blockchain

አምላክ፣ ቢትኮይን እና የስራ ማረጋገጫ፡ የጥንት ጥበብ እንዴት ብሎክቼይንን እንደሚያጎናጽፈው

Tanrı, Bitcoin və İşin sübutu: Qədim Müdriklik Blockchain-i necə gücləndirir

Dios, Bitcoin y la prueba de trabajo: cómo la sabiduría ancestral impulsa la cadena de bloques

Deus, Bitcoin e proba de traballo: como a sabedoría antiga potencia a cadea de bloques

神、ビットコイン、そしてプルーフ・オブ・ワーク：古代の知恵がブロックチェーンに活力を与える

ღმერთი, ბიტკოინი და სამუშაოს მტკიცებულება: როგორ აძლიერებს უძველესი სიბრძნე ბლოკჩეინს

Бог, биткоин и доказ за работа: како античката мудрост го напојува блокчејнот

خدای، بټکوین، او د کار ثبوت: څنګه پخوانی حکمت بلاکچین ته ځواک ورکوي

Deus, Bitcoin e Prova de Trabalho: Como a Sabedoria Antiga Alimenta o Blockchain

Zoti, Bitcoin dhe prova e punës: Si e fuqizon mençuria e lashtë Blockchain

พระเจ้า Bitcoin และการพิสูจน์การทำงาน: ภูมิปัญญาโบราณขับเคลื่อนบล็อคเชนอย่างไร

Hudaý, Bitcoin we işiň subutnamasy: Gadymy paýhas blokcheini nädip güýçlendirýär

Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 1.1 A Polyethylene-based metamaterial for acoustic control 1.1 A Polyethylene-based metamaterial for acoustic control 2 Relaxed micromorphic modelling of finite-size metamaterials 2 Relaxed micromorphic modelling of finite-size metamaterials 2.1 Tetragonal Symmetry / Shape of elastic tensors (in Voigt notation) 2.1 Tetragonal Symmetry / Shape of elastic tensors (in Voigt notation) 3 Dispersion curves 3 Dispersion curves 4 New considerations on the relaxed micromorphic parameters 4 New considerations on the relaxed micromorphic parameters 4.1 Consistency of the relaxed micromorphic model with respect to a change in the unit cell’s bulk material properties 4.1 Consistency of the relaxed micromorphic model with respect to a change in the unit cell’s bulk material properties 4.2 Consistency of the relaxed micromorphic model with respect to a change in the unit cell’s size 4.2 Consistency of the relaxed micromorphic model with respect to a change in the unit cell’s size 4.3 Relaxed micromorphic cut-offs 4.3 Relaxed micromorphic cut-offs 5 Fitting of the relaxed micromorphic parameters: the particular case of vanishing curvature (without Curl P and Curl P˙) 5 Fitting of the relaxed micromorphic parameters: the particular case of vanishing curvature (without Curl P and Curl P˙) 5.1 Asymptotes 5.1 Asymptotes 5.2 Fitting 5.2 Fitting 5.3 Discussion 5.3 Discussion 6 Fitting of the relaxed micromorphic parameters with curvature (with Curl P) 6 Fitting of the relaxed micromorphic parameters with curvature (with Curl P) 6.1 Asymptotes and 6.2 Fitting 6.1 Asymptotes and 6.2 Fitting 6.3 Discussion 6.3 Discussion 7 Fitting of the relaxed micromorphic parameters with enhanced kinetic energy (with Curl P˙) and 7.1 Asymptotes 7 Fitting of the relaxed micromorphic parameters with enhanced kinetic energy (with Curl P˙) and 7.1 Asymptotes 7.2 Fitting 7.2 Fitting 7.3 Discussion 7.3 Discussion 8 Summary of the obtained results 8 Summary of the obtained results 9 Conclusion and perspectives, Acknowledgements, and References 9 Conclusion and perspectives, Acknowledgements, and References A Most general 4th order tensor belonging to the tetragonal symmetry class A Most general 4th order tensor belonging to the tetragonal symmetry class B Coefficients for the dispersion curves without Curl P B Coefficients for the dispersion curves without Curl P C Coefficients for the dispersion curves with P C Coefficients for the dispersion curves with P D Coefficients for the dispersion curves with P◦ D Coefficients for the dispersion curves with P◦ 8 Summary of the obtained results We want to emphasize that the main focus of this work is not the result of the fitting of the three different approaches per se, but the semi-analytical fitting algorithm itself and the underlying consistency of the relaxed micromorphic model with respect to the material properties and some of the geometrical characteristic of the metamaterial that it represents. Using the complex but analytically defined expressions of the asymptotes, we can find a numerical fit of all material parameters by only giving the numerical values computed with Comsol Multiphysics® and the apparent mass density ρ of the unit cell. Note that we only use the cut-offs k = 0 and asymptotes k → ∞ for calculating the material parameters while the shape of the curves for intermediate values of k comes automatically. The routine is completely written with Mathematica allowing us to use symbolic calculations. The essential part of the fitting procedure uses the inbuilt algorithm NMinimize (with the Method RandomSearch) to minimize the mean square error of the asymptotes between the relaxed micromorphic model and the numerical values of the finite element approach in Comsol Multiphysics®. Therefore, in general, if a local minimum is found, it is not guaranteed that it corresponds to a global optimum as well. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv Authors:
(1) Jendrik Voss, Institute for Structural Mechanics and Dynamics, Technical University Dortmund and a Corresponding Author (jendrik.voss@tu-dortmund.de);
(2) Gianluca Rizzi, Institute for Structural Mechanics and Dynamics, Technical University Dortmund;
(3) Patrizio Neff, Chair for Nonlinear Analysis and Modeling, Faculty of Mathematics, University of Duisburg-Essen;
(4) Angela Madeo, Institute for Structural Mechanics and Dynamics, Technical University Dortmund. Authors: Authors: (1) Jendrik Voss, Institute for Structural Mechanics and Dynamics, Technical University Dortmund and a Corresponding Author (jendrik.voss@tu-dortmund.de); (2) Gianluca Rizzi, Institute for Structural Mechanics and Dynamics, Technical University Dortmund; (3) Patrizio Neff, Chair for Nonlinear Analysis and Modeling, Faculty of Mathematics, University of Duisburg-Essen; (4) Angela Madeo, Institute for Structural Mechanics and Dynamics, Technical University Dortmund.

Semi-Analytical Algorithm for Metamaterial Parameter Fitting

Signals in Angular are quite a new concept, so not everyone is using them nor is familiar with them. Creating a signal is quite easy: Signals const count = signal(0); const count = signal(0); But what it really is? Well, it’s a construct that plays very well with reactive programming and Angular templates. Because so far Angular has had to do a lot of things to recognize what changes and when to update templates. Just to say zone.js is a really huge part of Angular, and it tries to tap into natural things like setTimeout functions to spy on stuff coders do to update things smoothly. setTimeout By the way, my name is Tom Smykowski, I’m an expert in building scalable, enterprise Angular applications, and in this series, I’ll be teaching you how to do it! By the way, my name is Tom Smykowski, I’m an expert in building scalable, enterprise Angular applications, and in this series, I’ll be teaching you how to do it! Tom Smykowski Tom Smykowski The OnPush method changed it because with this strategy, Angular reacts only to input changes and events. However, it’s not a perfect solution, and having to choose between ngOnChanges, get/setters for inputs, and Observables and Subjects isn’t really fun. All of these approaches come with more or less boilerplate that only obfuscates the feature-oriented components. OnPush ngOnChanges Observables Subjects What’s even more important when working with asynchronicity in Angular is that it’s really easy to introduce memory leaks. All you have to do is not unsubscribe from a Subject (or BehaviourSubject etc.) or event-based observables (like fromEvent) for them to live not-quite rent-free forever in the device memory. It’s considerably the biggest drawback of these approaches. memory leaks Subject BehaviourSubject fromEvent Not a lot is said about it, but Angular Signals are also a solution to this problem. Specifically, as we can read in Angular documentation: Signals When you create an effect, it is automatically destroyed when its enclosing context is destroyed. This means that effects created within components are destroyed when the component is destroyed. The same goes for effects within directives, services, etc. When you create an effect, it is automatically destroyed when its enclosing context is destroyed. This means that effects created within components are destroyed when the component is destroyed. The same goes for effects within directives, services, etc. So, we don’t have to worry about this. Context knows about signals, and destroys them properly. So if you have a signal in your service or component, you don’t have to worry about a memory leak. signal Moreover, and this is really impressive, is how signals handle Observables and RxJS. No doubt, signals, as of today, aren’t able to take over all the good RxJS gives, however, in some cases, we may not need RxJS when working with signals. But what it means is that we could still be at risk of creating memory leaks when integrating signals and RxJS. The good news is that the unique construction of signals addresses this problem: signals The toSignal function allows you to turn an Observable to signal: toSignal import { Component } from '@angular/core';
import { AsyncPipe } from '@angular/common';
import { interval } from 'rxjs';
import { toSignal } from '@angular/core/rxjs-interop';
@Component({
  template: `{{ counter() }}`,
})
export class Ticker {
  counterObservable = interval(1000);
  // Get a `Signal` representing the `counterObservable`'s value.
  counter = toSignal(this.counterObservable, {initialValue: 0});
} import { Component } from '@angular/core';
import { AsyncPipe } from '@angular/common';
import { interval } from 'rxjs';
import { toSignal } from '@angular/core/rxjs-interop';
@Component({
  template: `{{ counter() }}`,
})
export class Ticker {
  counterObservable = interval(1000);
  // Get a `Signal` representing the `counterObservable`'s value.
  counter = toSignal(this.counterObservable, {initialValue: 0});
} As we can see in this example provided by Angular documentation, counter becomes a signal tracking counterObservable. Because a standard observable doesn’t guarantee to emit a value, we set an initialValue, otherwise, it would be undefined. counterObservable initialValue undefined This is, however, not the important part of the story, because the important one is that toSignal takes care of unsubscribing from the observable: toSignal The subscription created by toSignal automatically unsubscribes from the given Observable when the component or service which calls toSignal is destroyed. The subscription created by toSignal toSignal Basically meaning that it acts as a memory leak sanitizer, making sure that observables we use in components and services won’t create memory leaks because unsubscription is taken care automatically of. It’s a considerable advantage of signals contributing to solving the major cause of lags of Angular apps. Since toSignal unsubscribes automatically when a service or component is destroyed, there’s no risk of human error. What’s great? Of course, there are cases where memory leaks can still happen, and for those cases, we still can clear resources manually with manualCleanup which is called when the signal would normally do house chores. manualCleanup From the perspective of building scalable Angular apps for this, and other reasons, I’ll write in the following articles, that signals are great for preserving resources, increasing the responsiveness of Angular apps, and saving resources, especially in apps that by themselves use a lot of these. signals The streamlined handling of reactivity Angular 16 introduced with gradual feature build-up (like signal-based inputs in Angular 19) makes them a great fit for scalable apps due to lower maintenance costs, lower risk of memory leaks, and better overall performance. I recommend to gradually rewrite your app to signals if you want to scale it smoothly. I’ve prepared a free checklist on how to make your Angular app scalable and ready for enterprise demand. If you’re interested, you can get this checklist for free here. I’ve prepared a free checklist on how to make your Angular app scalable and ready for enterprise demand. If you’re interested, you can get this checklist for free here. free here free here If you have any questions, ask them in the comment section!

Angular Signals Are A Game Changer: How They're A Great Fit For Scalable Apps

Authors:
(1) Andrew Draganov, Aarhus University and All authors contributed equally to this research;
(2) David Saulpic, Université Paris Cité & CNRS;
(3) Chris Schwiegelshohn, Aarhus University. Table of Links Abstract and 1 Introduction 2 Preliminaries and Related Work 2.1 On Sampling Strategies 2.2 Other Coreset Strategies 2.3 Coresets for Database Applications 2.4 Quadtree Embeddings 3 Fast-Coresets 4 Reducing the Impact of the Spread 4.1 Computing a crude upper-bound 4.2 From Approximate Solution to Reduced Spread 5 Fast Compression in Practice 5.1 Goal and Scope of the Empirical Analysis 5.2 Experimental Setup 5.3 Evaluating Sampling Strategies 5.4 Streaming Setting and 5.5 Takeaways 6 Conclusion 7 Acknowledgements 8 Proofs, Pseudo-Code, and Extensions and 8.1 Proof of Corollary 3.2 8.2 Reduction of k-means to k-median 8.3 Estimation of the Optimal Cost in a Tree 8.4 Extensions to Algorithm 1 References Abstract We study the theoretical and practical runtime limits of k-means and k-median clustering on large datasets. Since effectively all clustering methods are slower than the time it takes to read the dataset, the fastest approach is to quickly compress the data and perform the clustering on the compressed representation. Unfortunately, there is no universal best choice for compressing the number of points – while random sampling runs in sublinear time and coresets provide theoretical guarantees, the former does not enforce accuracy while the latter is too slow as the numbers of points and clusters grow. Indeed, it has been conjectured that any sensitivity-based coreset construction requires super-linear time in the dataset size. We examine this relationship by first showing that there does exist an algorithm that obtains coresets via sensitivity sampling in effectively linear time – within log-factors of the time it takes to read the data. Any approach that significantly improves on this must then resort to practical heuristics, leading us to consider the spectrum of sampling strategies across both real and artificial datasets in the static and streaming settings. Through this, we show the conditions in which coresets are necessary for preserving cluster validity as well as the settings in which faster, cruder sampling strategies are sufficient. As a result, we provide a comprehensive theoretical and practical blueprint for effective clustering regardless of data size. Our code is publicly available at source and has scripts to recreate the experiments. 1 Introduction The modern data analyst has no shortage of clustering algorithms to choose from but, given the ever-increasing size of relevant datasets, many are often too slow to be practically useful. This is particularly relevant for big-data pipelines, where clustering algorithms are commonly used for compression. The goal is to replace a very large dataset by a smaller, more manageable one for downstream tasks, with the hope it represents the original input well. Lloyd’s algorithm [49] was introduced for precisely this reason and minimizes the quantization error – the sum of square distance from each input point to its representative in the compressed dataset. Arguably the most popular clustering algorithm, Lloyd’s runs for multiple iterations until convergence with every iteration requiring O(ndk) time, where n is the number of points, d is the number of features and k is the number of clusters – or the size of the targeted compression. For such applications, the number of points can easily be hundreds of millions and, since the quality of compression increases with k, standard objectives can have k in the thousands [41, 4]. In such settings, any O(ndk) algorithm is prohibitively slow. Examples like these have prompted the rise of big data algorithms that provide both theoretical and practical runtime improvements. The perspectives of theoretical soundness and practical efficacy are, however, often at odds with one another. On the one hand, theoretical guarantees provide assurance that the algorithm will work regardless of whatever unlucky inputs it receives. On the other hand, it may be difficult to convince oneself to implement the theoretically optimal algorithm when there are cruder methods that are faster to get running and perform well in practice. Since datasets can be large in the number of points n and/or the number of features d, big-data methods must mitigate the effects of both. With respect to the feature space, the question is effectively closed as random projections are fast (running in effectively linear time), practical to implement [50], and provide tight guarantees on the embedding’s size and quality. The outlook is less clear when reducing the number of points n, and there are two separate paradigms that each provide distinct advantages. On the one hand, we have uniform sampling, which runs in sublinear time but may miss important subsets of the data and therefore can only guarantee accuracy under certain strong assumptions on the data [45]. On the other hand, the most accurate sampling strategies provide the strong coreset guarantee, wherein the cost of any solution on the compressed data is within an ε-factor of that solution’s cost on the original dataset [25]. Our contributions We study both paradigms (uniform sampling and strong coresets) with respect to a classic problem – compression for the k-means and k-median objectives. Whereas uniform sampling provides optimal speed but no worst-case accuracy guarantee, all available coreset constructions have a running time of at least Ω˜(nd + nk) when yielding tight bounds on the minimum number of samples required for accurate compression. It is easy to show that any algorithm that achieves a compression guarantee must read the entire dataset[1]. Thus a clear open question is what guarantees are achievable in linear or nearly-linear time. Indeed, currently available fast sampling algorithms for clustering [6, 5] cannot achieve strong coreset guarantees. Recently, [31] proposed a method for strong coresets that runs in time O˜(nd + nk) and conjectured this to be optimal for k-median and k-means. While this bound is effectively optimal for small values of k, there are many applications such as computer vision [34] or algorithmic fairness [18] where the number of clusters can be larger than the number of features by several orders of magnitude. In such settings, the question of time-optimal coresets remains open. Since the issue of determining a coreset of optimal size has recently been closed [25, 28, 44], this is arguably the main open problem in coreset research for center-based clustering. We resolve this by showing that there exists an easy-to-implement algorithm that constructs coresets in O˜(nd) time – only logarithmic factors away from the time it takes to read in the dataset. Nonetheless, this does not fully illuminate the landscape among sampling algorithms for clustering in practice. Although our algorithm achieves both an optimal runtime and an optimal compression, it is certainly possible that other, cruder methods may be just as viable for all practical purposes. We state this formally in the following question: When are optimal k-means and k-median coresets necessary and what is the practical tradeoff between coreset speed and accuracy? o answer this, we perform a thorough comparison across the full span of sampling algorithms that are faster than our proposed method. Through this we verify that, while these faster methods are sufficiently accurate on many real-world datasets, there exist data distributions that cause catastrophic failure for each of them. Indeed, these cases can only be avoided with a strong-coreset method. Thus, while many practical settings do not require the full coreset guarantee, one cannot cut corners if one wants to be confident in their compression. We verify that this extends to the streaming paradigm and applies to downstream clustering approaches. In summary, our contributions are as follows: • We show that one can obtain strong coresets for k-means and k-median in O˜(nd) time. This resolves a conjecture on the necessary runtime for k-means coresets [31] and is theoretically optimal up to log-factors. • Through a comprehensive analysis across datasets, tasks, and streaming/non-streaming paradigms, we verify that there is a necessary tradeoff between speed and accuracy among the linear- and sublinear-time sampling methods. This gives the clustering practitioner a blueprint on when to use each compression algorithm for effective clustering results in the fastest possible time. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Andrew Draganov, Aarhus University and All authors contributed equally to this research; (2) David Saulpic, Université Paris Cité & CNRS; (3) Chris Schwiegelshohn, Aarhus University. Authors: Authors: (1) Andrew Draganov, Aarhus University and All authors contributed equally to this research; (2) David Saulpic, Université Paris Cité & CNRS; (3) Chris Schwiegelshohn, Aarhus University. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Preliminaries and Related Work 2.1 On Sampling Strategies 2.1 On Sampling Strategies 2.2 Other Coreset Strategies 2.2 Other Coreset Strategies 2.3 Coresets for Database Applications 2.3 Coresets for Database Applications 2.4 Quadtree Embeddings 2.4 Quadtree Embeddings 3 Fast-Coresets 3 Fast-Coresets 4 Reducing the Impact of the Spread 4 Reducing the Impact of the Spread 4.1 Computing a crude upper-bound 4.1 Computing a crude upper-bound 4.2 From Approximate Solution to Reduced Spread 4.2 From Approximate Solution to Reduced Spread 5 Fast Compression in Practice 5 Fast Compression in Practice 5.1 Goal and Scope of the Empirical Analysis 5.1 Goal and Scope of the Empirical Analysis 5.2 Experimental Setup 5.2 Experimental Setup 5.3 Evaluating Sampling Strategies 5.3 Evaluating Sampling Strategies 5.4 Streaming Setting and 5.5 Takeaways 5.4 Streaming Setting and 5.5 Takeaways 6 Conclusion 6 Conclusion 7 Acknowledgements 7 Acknowledgements 8 Proofs, Pseudo-Code, and Extensions and 8.1 Proof of Corollary 3.2 8 Proofs, Pseudo-Code, and Extensions and 8.1 Proof of Corollary 3.2 8.2 Reduction of k-means to k-median 8.2 Reduction of k-means to k-median 8.3 Estimation of the Optimal Cost in a Tree 8.3 Estimation of the Optimal Cost in a Tree 8.4 Extensions to Algorithm 1 8.4 Extensions to Algorithm 1 References References Abstract We study the theoretical and practical runtime limits of k-means and k-median clustering on large datasets. Since effectively all clustering methods are slower than the time it takes to read the dataset, the fastest approach is to quickly compress the data and perform the clustering on the compressed representation. Unfortunately, there is no universal best choice for compressing the number of points – while random sampling runs in sublinear time and coresets provide theoretical guarantees, the former does not enforce accuracy while the latter is too slow as the numbers of points and clusters grow. Indeed, it has been conjectured that any sensitivity-based coreset construction requires super-linear time in the dataset size. We examine this relationship by first showing that there does exist an algorithm that obtains coresets via sensitivity sampling in effectively linear time – within log-factors of the time it takes to read the data. Any approach that significantly improves on this must then resort to practical heuristics, leading us to consider the spectrum of sampling strategies across both real and artificial datasets in the static and streaming settings. Through this, we show the conditions in which coresets are necessary for preserving cluster validity as well as the settings in which faster, cruder sampling strategies are sufficient. As a result, we provide a comprehensive theoretical and practical blueprint for effective clustering regardless of data size. Our code is publicly available at source and has scripts to recreate the experiments. 1 Introduction The modern data analyst has no shortage of clustering algorithms to choose from but, given the ever-increasing size of relevant datasets, many are often too slow to be practically useful. This is particularly relevant for big-data pipelines, where clustering algorithms are commonly used for compression. The goal is to replace a very large dataset by a smaller, more manageable one for downstream tasks, with the hope it represents the original input well. Lloyd’s algorithm [49] was introduced for precisely this reason and minimizes the quantization error – the sum of square distance from each input point to its representative in the compressed dataset. Arguably the most popular clustering algorithm, Lloyd’s runs for multiple iterations until convergence with every iteration requiring O(ndk) time, where n is the number of points, d is the number of features and k is the number of clusters – or the size of the targeted compression. For such applications, the number of points can easily be hundreds of millions and, since the quality of compression increases with k, standard objectives can have k in the thousands [41, 4]. In such settings, any O(ndk) algorithm is prohibitively slow. Examples like these have prompted the rise of big data algorithms that provide both theoretical and practical runtime improvements. The perspectives of theoretical soundness and practical efficacy are, however, often at odds with one another. On the one hand, theoretical guarantees provide assurance that the algorithm will work regardless of whatever unlucky inputs it receives. On the other hand, it may be difficult to convince oneself to implement the theoretically optimal algorithm when there are cruder methods that are faster to get running and perform well in practice. Since datasets can be large in the number of points n and/or the number of features d, big-data methods must mitigate the effects of both. With respect to the feature space, the question is effectively closed as random projections are fast (running in effectively linear time), practical to implement [50], and provide tight guarantees on the embedding’s size and quality. The outlook is less clear when reducing the number of points n, and there are two separate paradigms that each provide distinct advantages. On the one hand, we have uniform sampling, which runs in sublinear time but may miss important subsets of the data and therefore can only guarantee accuracy under certain strong assumptions on the data [45]. On the other hand, the most accurate sampling strategies provide the strong coreset guarantee, wherein the cost of any solution on the compressed data is within an ε-factor of that solution’s cost on the original dataset [25]. Our contributions We study both paradigms (uniform sampling and strong coresets) with respect to a classic problem – compression for the k-means and k-median objectives. Whereas uniform sampling provides optimal speed but no worst-case accuracy guarantee, all available coreset constructions have a running time of at least Ω˜(nd + nk) when yielding tight bounds on the minimum number of samples required for accurate compression. It is easy to show that any algorithm that achieves a compression guarantee must read the entire dataset[1]. Thus a clear open question is what guarantees are achievable in linear or nearly-linear time. Indeed, currently available fast sampling algorithms for clustering [6, 5] cannot achieve strong coreset guarantees. Recently, [31] proposed a method for strong coresets that runs in time O˜(nd + nk) and conjectured this to be optimal for k-median and k-means. While this bound is effectively optimal for small values of k, there are many applications such as computer vision [34] or algorithmic fairness [18] where the number of clusters can be larger than the number of features by several orders of magnitude. In such settings, the question of time-optimal coresets remains open. Since the issue of determining a coreset of optimal size has recently been closed [25, 28, 44], this is arguably the main open problem in coreset research for center-based clustering. We resolve this by showing that there exists an easy-to-implement algorithm that constructs coresets in O˜(nd) time – only logarithmic factors away from the time it takes to read in the dataset. Nonetheless, this does not fully illuminate the landscape among sampling algorithms for clustering in practice. Although our algorithm achieves both an optimal runtime and an optimal compression, it is certainly possible that other, cruder methods may be just as viable for all practical purposes. We state this formally in the following question: When are optimal k-means and k-median coresets necessary and what is the practical tradeoff between coreset speed and accuracy? o answer this, we perform a thorough comparison across the full span of sampling algorithms that are faster than our proposed method. Through this we verify that, while these faster methods are sufficiently accurate on many real-world datasets, there exist data distributions that cause catastrophic failure for each of them. Indeed, these cases can only be avoided with a strong-coreset method. Thus, while many practical settings do not require the full coreset guarantee, one cannot cut corners if one wants to be confident in their compression. We verify that this extends to the streaming paradigm and applies to downstream clustering approaches. In summary, our contributions are as follows: • We show that one can obtain strong coresets for k-means and k-median in O˜(nd) time. This resolves a conjecture on the necessary runtime for k-means coresets [31] and is theoretically optimal up to log-factors. • Through a comprehensive analysis across datasets, tasks, and streaming/non-streaming paradigms, we verify that there is a necessary tradeoff between speed and accuracy among the linear- and sublinear-time sampling methods. This gives the clustering practitioner a blueprint on when to use each compression algorithm for effective clustering results in the fastest possible time. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Stories, News & Mentions about 1FIT

Categories

Trending Topics

Stories, News & Mentions about 1FIT

1fit’s Stories on HackerNoon

1fit's News & Mentions

Categories

Trending Topics