https://github.com/amitness/toolbox

Curated list of libraries for a faster machine learning workflow

toolbox

Curated libraries for a faster workflow

Phase: Data

Data Annotation

• Image: makesense.ai
• Text: doccano, prodigy

Datasets

• Text: nlp-datasets, curse-words, badwords, LDNOOBW, english-words (A text file containing over 466k English words), 10K most common words
• Image: 1 million fake faces
• Dataset search engine: datasetlist, UCI Machine Learning Datasets

Importing Data

• Audio: pydub
• Video: pytube (download youTube vidoes)
• Image: py-image-dataset-generator (auto fetch images from web for certain search)
• News: news-please
• PDF: camelot, tabula-py
• Remote file: smart_open
• Crawling: pyppeteer (chrome automation), MechanicalSoup, libextract
• Google sheets: gspread
• Google drive: gdown
• Python API for datasets: pydataset
• Google maps location data: geo-heatmap

Data Augmentation

• Text: nlpaug
• Image: imgaug, albumentations, augmentor
• Audio: audiomentations, muda

Phase: Exploration

Data Preparation

• Missing values: missingno
• Split images into train/validation/test: split-folders
• Class Imbalance: imblearn
• Categorical encoding: category_encoders

Notebook Exploration

• View Jupyter notebooks through CLI: nbdime
• Parametrize notebooks: papermill
• Access notebooks programatically: nbformat
• Convert notebooks to other formats: nbconvert
• Extra utilities not present in frameworks: mlxtend
• Maps in notebooks: ipyleaflet

Phase: Feature Engineering

Feature Generation

• Automatic feature engineering: featuretools, autopandas, tsfresh (automatic feature engineering for time series)
• Custom distance metric learning: metric-learn
• List of holidays: python-holidays

Phase: Modeling

Model Selection

• Bruteforce through all scikit-learn model and parameters: auto-sklearn, tpot
• Autogenerate ML code: automl-gs, mindsdb
• Pretrained models: modeldepot, pytorch-hub, papers-with-code
• Find SOTA models: sotawhat
• Gradient Boosting: catboost, lightgbm (GPU-capable), thunderbm (GPU-capable)
• Hidden Markov Models: hmmlearn
• Genetic Programming: gplearn
• Active Learning: modal
• Support Vector Machines: thundersvm (GPU-capable)
• Rule based classifier: sklearn-expertsys
• Probabilistic modeling: pomegranate
• Graph Embedding and Community Detection: karateclub
• Anomaly detection: adtk

NLP

• Preprocessing: textacy
• Text Extraction from Image, Audio, PDF: textract
• Text generation: gp2client, textgenrnn, gpt-2-simple
• Text summarization: textrank, pytldr
• Spelling correction: JamSpell, pyhunspell, pyspellchecker, cython_hunspell, hunspell-dictionaries, autocorrect (can add more languages)
• Keyword extraction: rake, pke
• Multiply Choice Question Answering: mcQA
• Sequence to sequence models: headliner
• Transfer learning: finetune
• Translation: googletrans
• Embeddings: pymagnitude (manage vector embeddings easily), chakin (download pre-trained word vectors), sentence-transformers, InferSent, bert-as-service, sent2vec
• Multilingual support: polyglot, inltk (indic languages), indic_nlp
• NLU: snips-nlu
• Semantic parsing: quepy
• Inflections: inflect
• Contractions: pycontractions
• Coreference Resolution: neuralcoref
• Readability: homer
• Language Detection: language-check
• Topic Modeling: guidedlda, enstop
• Clustering: spherecluster (kmeans with cosine distance), kneed (automatically find number of clusters from elbow curve), kmodes
• Metrics: seqeval (NER, POS tagging)
• String match: jellyfish (perform string and phonetic comparison),flashtext (superfast extract and replace keywords), pythonverbalexpressions: (verbally describe regex), commonregex (readymade regex for email/phone etc)
• Sentiment: vaderSentiment (rule based)
• Text distances: textdistance, editdistance
• PID removal: scrubadub
• Profanity detection: profanity-check
• wordclouds: stylecloud

Speech Recognition

• Library: speech_recognition

RecSys

• Factorization machines (FM), and field-aware factorization machines (FFM): xlearn
• Scikit-learn like API: surprise

Computer Vision

• Image processing: scikit-image, imutils
• Segmentation Models in Keras: segmentation_models
• Face recognition: face_recognition, face-alignment (find facial landmarks)
• Face swapping: faceit
• Video summarization: videodigest
• Semantic search over videos: scoper
• OCR: keras-ocr, pytesseract
• Object detection: luminoth

Timeseries

• Predict Time Series: prophet
• Scikit-learn like API: sktime

Framework extensions

• Pytorch: Keras like summary for pytorch, skorch (wrap pytorch in scikit-learn compatible API)
• Einstein notation: einops
• Scikit-learn: scikit-lego

Phase: Monitoring

Model Training Monitoring

• Learning curve: lrcurve (plot realtime learning curve in Keras), livelossplot
• Notifications: knockknock (get notified by slack/email), jupyter-notify (notify when task is completed in jupyter)

Phase: Optimization

Hyperparameter Optimization

• Keras: keras-tuner
• Scikit-learn: sklearn-deap (evolutionary algorithm for hyperparameter search)
• General: hyperopt

Interpretability

• Visualize keras models: keras-vis
• Interpret models: eli5, lime, shap, alibi, tf-explain, treeinterpreter, pybreakdown, xai
• Interpret BERT: exbert
• Interpret word2vec: word2viz

Visualization

• Draw CNN figures: nn-svg
• Visualization for scikit-learn: yellowbrick, scikit-plot
• XKCD like charts: chart.xkcd
• Convert matplotlib charts to D3 charts: mpld3
• Generate graphs using markdown: mermaid
• Visualize topics models: pyldavis
• High dimensional visualization: umap

Phase: Production

Model Serialization

• Transpiling: sklearn-porter (transpile sklearn model to C, Java, JavaScript and others), m2cgen
• Pickling extended: cloudpickle, jsonpickle

Scalability

• Parallelize Pandas: pandarallel, swifter, modin

Bechmark

• Profile pytorch layers: torchprof

API

• Configuration Management: config, python-decouple
• Data Validation: schema, jsonschema, cerebrus, pydantic, marshmallow
• Enable CORS in Flask: flask-cors
• Caching: cachetools, cachew (cache to local sqlite)
• Authentication: pyjwt (JWT)
• Task Queue: rq

Dashboard

• Generate frontend with python: streamlit

Adversarial testing

• Generate images to fool model: foolbox
• Generate phrases to fool NLP models: triggers
• General: cleverhans

Python libraries

• Datetime compatible API for Bikram Sambat: nepali-date
• bloom filter: python-bloomfilter
• Run python libraries in sandbox: pipx
• Pretty print tables in CLI: tabulate
• Leaflet maps from python: folium
• Debugging: PySnooper