troschiev / ds_portfolio

Избранные проекты по машинному обучению

Intro

Here I post projects with something interesting and useful.

Each project is a .ipynb file with:

• formally set task
• EDA
• working code
• text explanations
• visualized results
• conclusions and / or plans for improving the result

Large projects

Project	Methods	Stack
Full face recognition pipeline	Face detection (YOLO), face landmarks coordinates regression, face alignment, face embedding DNN, cos similarity analysis, telegram bot implementation	python, pytorch, albumentations, timm, aiogram, cv2
DCGAN for face generation	Hand-made DCGAN, custom Gauss noise layers, gradien penalty, regularization by labels (soft, noise, flip), specific metrics: Fréchet inception distance и Leave-one-out-1-NN classification, epoch animation	python, pytorch, matplotlib, albumentations, timm

Selected study projects

Project	Methods	Stack
Person Age by Photo Yandex Practicum	EffecientNet V2 pretrained, strong augmentation, result analysis	python, pytorch, matplotlib, albumentations, timm, альтернативный код в keras
Image Autoencoders DL school MIPT	CNN, latent vectors, latent space sampling, VAE, Conditional VAE, KL divergence loss	python, pytorch, sklearn, matplotlib
Toxic Comment Classification Yandex Practicum	Text preprocessing, TF-IDF, lemmatization, word2vec, fine-tuned BERT, weighted loss	python, pandas, numpy, pytorch, BERT, sklearn, optuna, matplotlib, seaborn, wordcloud, nltk, enchant, spacy, gensim
Semantic Segmentation of Skin Lesions DL school MIPT	CNN, hand-made Unet и SegNet (based on VGG16), image augmentations, custom loss functions	python, pytorch, sklearn, albumentations, matplotlib
Gold Recovery Efficiency Model Yandex Practicum	EDA, preprocessing, feature selection by phik correlation, feature dimensionality reduction, gradient boosting, hyperparameter tuning	python, pandas, seaborn, sklearn, pipeline, optuna, phik, xgboost
Simpsons Classification DL school MIPT	CNN, transfer learning, image augmentation, class imbalance, extended result visualization	python, pytorch, sklearn, matplotlib
Taxi Demand Time Series Yandex Practicum	Exponential smoothing, SARIMA, linear regression, gradient boosting, hybrid boosting, hyperparameter tuning, neural networks: dense, LSTM	python, pandas, numpy, seaborn, sklearn, pipeline, statsmodels, pytorch, lightgbm, optuna, prophet, phik, pmdarima
Used Car Price Prediction Yandex Practicum	EDA, deep preprocessing, feature selection, gradient boosting, hyperparameter tuning, fully connected neural network, residual analysis, postcode coordinates, feature importance (permutation)	python, pandas, numpy, seaborn, sklearn, pipeline, pytorch, lightgbm, xgboost, catboost, optuna, phik
Bank Customer Churn Yandex Practicum	Class imbalance, T-SNE visualization, gradient boosting, hyperparameter tuning, cross-validation ROC curves, feature engineering, wrapper class, stacking, feature importance (permutation)	python, pandas, numpy, seaborn, sklearn, pipeline, lightgbm, xgboost, optuna,
Titanic Analysis Kaggle	Advanced analysis, advanced visualization, gradient boosting, hyperparameter tuning	python, pandas, numpy, seaborn, plotly, sklearn, catboost, optuna,
Selecting Oil Well Location Yandex Practicum	Synthetic data, bootstrap, QQ-plot, confidence intervals	python, pandas, numpy, seaborn, sklearn, pipeline
Video Game Market Analysis Yandex Practicum	Deep data analysis, advanced visualization, hypothesis testing	python, pandas, numpy, seaborn, plotly, scipy
Tariff Recommendation for a Mobile Operator Client Yandex Practicum	Classic ML models: logistic regression with feature engineering, SVM, naive Bayes classifier, decision trees, random forest, gradient boosting, model stacking	python, pandas, sklearn, numpy, seaborn, xgboost

Contacts

E-mail: [email protected]

Telegram: https://t.me/sergey_doc