Dev Resource

---

---

Public dataset

• Dataset source
  • Require login - https://public.roboflow.com/
• Conll-03
• MNIST
• CIFAR-10 - https://www.cs.toronto.edu/~kriz/cifar.html
  • 10 classes labeled images
• Coco dataset - https://cocodataset.org/#home
• LVIS - Large Vocabulary Instance Segmentation
  • https://www.lvisdataset.org/
  • ~2 million high-quality instance segmentation masks for over 1000 entry-level object categories in 164k images
• Flower dataset
  • TF - https://www.tensorflow.org/datasets/catalog/tf_flowers
• CARLA - self-driving car dataset
  • https://paperswithcode.com/dataset/carla

---

Package/Library 🔧

• List of all popular ML package for various fields
  • https://github.com/josephmisiti/awesome-machine-learning#python-natural-language-processing

Small package :

• deepchecks - https://deepchecks.com/

  • A very easy to use lib to do model & data checking, e.g. train/test set distribution check

    

  • Very useful seemingly

ML core framework :

• Scikit-learn (Based on SciPy)
  • https://scikit-learn.org/stable/install.html
  • The user guide is too detailed. Best usage if you have a target to search for.
    • https://scikit-learn.org/stable/user_guide.html

• Pycaret

  • https://pycaret.gitbook.io/docs/
  • wrapper around several machine learning libraries and frameworks, such as scikit-learn, XGBoost, LightGBM, CatBoost, spaCy, Optuna, Hyperopt, Ray, and a few more.

• PyTorch

  • Machine Learning lib
  • https://pytorch.org/get-started/locally/
• pytorch lightning (Based on Pytorch)
  • A high level API allow researcher to build PyTorch model more efficiently
  • i.e. More speedy in coding
  • https://www.pytorchlightning.ai/

• Fastai (Based on PyTorch)

  • Deep Learning lib
  • Allow the control from High, Middle, Low level
  • Domains: NLP, CV, Tabular, (?)collaborative filtering
  • NOT support Mac, so run it on Cloud/Colab

• Tensorflow (VS PyTorch)

• Keras (Based on Tensorflow)
• Autokeras (Based on Keras)
  • Deep Learning lib
  • Similar to scikit but more high level api
  • e.g. you can specify a task type, they help to choose the model algorithm

Domain-Specific :

• NLP
  • Scikit-crfsuite
    • CRF model classifier
  • MITIE
    • Train NER model
  • NCRF++
  • FastText - https://fasttext.cc/docs/en/supervised-tutorial.html
    • fastText is a library for efficient learning of word representations and sentence classification.****
• Computer vision
  • Framework
    • OpenMMLab - https://openmmlab.com/codebase
      • From China, quite a big framework
      • Wrap up a wide range of application.
    • Icevision - https://airctic.com/0.12.0/
    • YOLOv5 - https://docs.ultralytics.com/
      • Quite infamous due to not recognized by original YOLO author
      • try YOLOv4 instead
    • Detectron
    • Mediapipe
  • Data
    • augmentation lib - Albumentations

---

Tool / Software

• Annotation tool:
  • LabelImg
  • labelstudio - https://labelstud.io/
    • can label many types of data(img, audio, text…etc)
    • Extremely useful than labelimg (recommened by colleagues)
  • CVAT
  • makesense.ai
  • Labelbox
  • Roboflow (not free for business)

---

Resource List

• State-of-the-Art models list - https://paperswithcode.com/sota

Comments