14 skills found
piyushpathak03 / Recommendation SystemsRecommendation Systems This is a workshop on using Machine Learning and Deep Learning Techniques to build Recommendation Systesm Theory: ML & DL Formulation, Prediction vs. Ranking, Similiarity, Biased vs. Unbiased Paradigms: Content-based, Collaborative filtering, Knowledge-based, Hybrid and Ensembles Data: Tabular, Images, Text (Sequences) Models: (Deep) Matrix Factorisation, Auto-Encoders, Wide & Deep, Rank-Learning, Sequence Modelling Methods: Explicit vs. implicit feedback, User-Item matrix, Embeddings, Convolution, Recurrent, Domain Signals: location, time, context, social, Process: Setup, Encode & Embed, Design, Train & Select, Serve & Scale, Measure, Test & Improve Tools: python-data-stack: numpy, pandas, scikit-learn, keras, spacy, implicit, lightfm Notes & Slides Basics: Deep Learning AI Conference 2019: WhiteBoard Notes | In-Class Notebooks Notebooks Movies - Movielens 01-Acquire 02-Augment 03-Refine 04-Transform 05-Evaluation 06-Model-Baseline 07-Feature-extractor 08-Model-Matrix-Factorization 09-Model-Matrix-Factorization-with-Bias 10-Model-MF-NNMF 11-Model-Deep-Matrix-Factorization 12-Model-Neural-Collaborative-Filtering 13-Model-Implicit-Matrix-Factorization 14-Features-Image 15-Features-NLP Ecommerce - YooChoose 01-Data-Preparation 02-Models News - Hackernews Product - Groceries Python Libraries Deep Recommender Libraries Tensorrec - Built on Tensorflow Spotlight - Built on PyTorch TFranking - Built on TensorFlow (Learning to Rank) Matrix Factorisation Based Libraries Implicit - Implicit Matrix Factorisation QMF - Implicit Matrix Factorisation Lightfm - For Hybrid Recommedations Surprise - Scikit-learn type api for traditional alogrithms Similarity Search Libraries Annoy - Approximate Nearest Neighbour NMSLib - kNN methods FAISS - Similarity search and clustering Learning Resources Reference Slides Deep Learning in RecSys by Balázs Hidasi Lessons from Industry RecSys by Xavier Amatriain Architecting Recommendation Systems by James Kirk Recommendation Systems Overview by Raimon and Basilico Benchmarks MovieLens Benchmarks for Traditional Setup Microsoft Tutorial on Recommendation System at KDD 2019 Algorithms & Approaches Collaborative Filtering for Implicit Feedback Datasets Bayesian Personalised Ranking for Implicit Data Logistic Matrix Factorisation Neural Network Matrix Factorisation Neural Collaborative Filtering Variational Autoencoders for Collaborative Filtering Evaluations Evaluating Recommendation Systems
PlanetCache / CacheJSONA JSON encoding / decoding utility for Intersystems Cache and Ensemble.
tonyleidong / OptimalFlowOptimalFlow is an omni-ensemble and scalable automated machine learning Python toolkit, which uses Pipeline Cluster Traversal Experiments(PCTE) and Selection-based Feature Preprocessor with Ensemble Encoding(SPEE), to help data scientists build optimal models, and automate supervised learning workflow with simpler coding.
mavalliani / Semantic Similarity Of SentencesMethods used: Cosine Similarity with Glove, Smooth Inverse Frequency, Word Movers Difference, Sentence Embedding Models (Infersent and Google Sentence Encoder), ESIM with pre-trained FastText embedding. Best performing method on Quora Question pair dataset was an Ensemble method with 0.27 log-loss.
amerob / Kaggle For Ml EngineersA systems-first handbook and Python framework for ML engineers. It provides architectural patterns for CV, NLP, and tabular data. Includes production-ready modules for OOF pipelines, safe target encoding, and ensemble stacking.
rishipython / One Eye Is All You Need Lightweight Ensembles For Gaze Estimation With Single EncodersNo description available
noahdasanaike / EnsemblelinkEnsembleLink: zero-shot record linkage using ensemble retrieval and cross-encoder reranking
SamSamhuns / Autodecoder Ensemble Net For 3D ObjectsAn Encoder-less Decoder Ensemble Deep Neural Network for extraction of for 3D shape descriptors
airKlizz / Passage OrderingCode, models and data for the paper *Ordering Sentences and Paragraphs with Pre-trained Encoder-Decoder Transformers and Pointer Ensembles* under review at DocEng2021.
ReverendBayes / Telecom Churn PredictorPredicts which telecom customers are likely to churn with 95% accuracy using real-world data features from usage, billing, and support data. Implements Sturges-based binning, one-hot encoding, stratified 80/20 train-test split, and a two-level ensemble pipeline with soft voting. Achieves 94.60% accuracy, 0.8968 AUC, 0.8675 precision, 0.7423 recall.
Reza-Rezvan / AdvanceSpliceA robust method integrating N-gram encoding and ensemble modeling for enhanced splice site prediction accuracy.
parth2608 / Prediction Of Stock Market Security Price Using Chart Patterns And Indicator AnalysisIn this project, we have attempted to create and compare different machine learning models that would accurately predict stock prices. For this, we have built five machine learning models for time series and chart pattern analysis. The LSTM, RNN, GRU, and Ensemble models are based directly on historical data of the stock prices, whereas the chart pattern analysis is based on encoder-decoder architecture and DCNN.
Johnnywang1899 / Credit Risk Analysisimbalanced-learn library, supervised learning, Scikit-learn machine learning library for Python (sklearn), supervised learning in linear models (linear regression & logistic regression), dataset split into training & testing datasets, Accuracy/Precision/sensitivity(recall)/F1 score, confusion matrix, SVM(Support Vector Machine – support vector, hyperplane), data preprocessing: - labelling (encoding – convert all text columns to numeric label), - Data Scale & Normalization (Standard scaler – mean = 0, variance = 1), Decision Trees, Ensemble Learning – Random Forest (weak/moderate/strong learner), Bootstrap Aggregation, Boosting [Adaptive boosting (AdaBoost), Gradient Boosting, for Class imbalance (solution 1: Oversampling (Random oversampling, synthetic minority oversampling technique (SMOTE)), solution 2: Undersampling (Random undersampling, Cluster Centroid undersampling, solution 3: SMOTEENN)
Makundi / Machine Learning IPT ParrotAIWeek1 Report Here is a quick summary of what I have achieved to learn in my first week of training under ParrotAi. Introduction to Machine learning , I have achieved to know a good intro into Machine Learning which include the history of ML ,the types of ML such supervised, unsupervised, Reinforcement learning. And also answers to questions such why machine learning? , challenges facing machine learning which include insufficient data, irrelevant on data, overfitting, underfitting and there solutions in general. Supervised Machine algorithms, here I learnt the theory and intuition behind the common used supervised ML including the KNN, Linear Regressions, Logistic, Regression, and Ensemble algorithm the Random forest. Also not only the intuition but their implementation in python using the sklearn library and parameter tuning them to achieve a best model with stunning accuracy(here meaning the way to regularize the model to avoid overfitting and underfitting).And also the intuition on where to use/apply the algorithms basing on the problem I.e classification or regression. Also which model performs better on what and poor on what based on circumstances. Data preprocessing and representation here I learnt on the importance of preprocessing the data, also the techniques involved such scaling(include Standard Scaling, RobustScaling and MinMaxScaler) ,handling the missing data either by ignoring(technical termed as dropping) the data which is not recommended since one could loose important patterns on the data and by fitting the mean or median of the data points on the missing places. On data representation involved on how we can represent categorical features so as they can be used in the algorithm, the method learnt here is One-Hot Encoding technique and its implementation in python using both Pandas and Sklearn Libraries. Model evaluation and improvement. In this section I grasped the concept of how you can evaluate your model if its performing good or bad and the ways you could improve it. As the train_test_split technique seems to be imbalance hence the cross-validation technique which included the K-fold , Stratified K-fold and other strategies such LeaveOneOut which will help on the improvement of your model by splitting data in a convenience manner to help in training of model, thus making it generalize well on unseen data. I learnt also on the GridSearch technique which included the best method in which one can choose the best parameters for the model to improve the performance such as the simple grid search and the GridSearch with cross-validation technique, all this I was able to implement them in code using the Sklearn library in python. Lastly the week challenge task given to us was tremendous since I got to apply what I learned in theory to solve a real problem.It was good to apply the workflow of a machine learning task starting from understanding the problem, getting to know the data, data preprocessing , visualising the data to get more insights, model selection, training the model to applying the model to make prediction In general I was able to grasp and learn much in this week from basic foundation of Machine Learning to the implementations of the algorithms in code. The great achievement so far is the intuition behind the algorithm especially supervised ones. Though yet is much to be covered but the accomplishment I have attained so far its a good start to say to this journey on Machine learning. My expectation on the coming week is on having a solid foundation on deep learning.
