- https://www.wandb.com/
- W&B is a key piece of our fast-paced, cutting-edge, large-scale research workflow: great flexibility, performance, and user experience.
- Experiment tracking for deep learning
- Instrument training scripts
- Stages covered: only from Model training to deploy (run multiple experiments, track and look at the data visualisations - only helps track, and use the info to plot graphs and make your decisions)
- Pros:
- Supports model execution and tracking of the execution
- Gathers good stats / metrics to make sound decisions after or during execution of experiments
- Has support for comparisons of experiments (and models) and performances
- Tracks the best model producing during experiments execution
- Supports multiple python frameworks (keras, tensorflow, pytorch)
- Getting started with tool is really easy and great way to integrate hooks into our experiment code
- Very good UX, docs and site layout, user-friendly
- Cons:
- Does not support or have tooling for stages before experiment execution (i.e. data cleaning, visualisation, data validation, feature extraction)
- Not sure if they provide infrastructure on which we can run our experiments
- Installation or getting started
- Pricing & features: https://www.wandb.com/pricing
- Documentations
- https://docs.wandb.com/docs/started.html
- Sweeps: https://docs.wandb.com/sweeps
- Framework docs: https://docs.wandb.com/library/frameworks
- Group: https://docs.wandb.com/library/advanced/grouping
- Resuming tracking: https://docs.wandb.com/library/advanced/resuming
- Examples
- Code & concepts
- Videos
- GitHub
- Organisation: https://github.com/wandb
- W&B Client: https://github.com/wandb/client
- W&B User issues: https://github.com/wandb/user-issues
- W&B Examples: https://github.com/wandb/examples
- W&B Artifacts examples: https://github.com/wandb/artifacts-examples
- W&B CodeSearchNet: https://github.com/wandb/codesearchnet (an F/OSS benchmarking project in collaboration with GitHub)
- W&B Training: https://github.com/wandb/wandb-training
- W&B Tutorial: https://github.com/wandb/tutorial
- W&B GitBook: https://github.com/wandb/gitbook
- W&B Docs: https://github.com/wandb/docs
- Additional resources
- Error caused by missing input_shape in your first layer
- Bloomberg summary colab notebook
- Implementation of W&B for model param tuning
- https://talktotransformer.com/ - Adam Daniel King's implementation of GPT-2 on the back of the PyTorch version
- https://app.wandb.ai/sayakpaul/jigsaw-toxic/reports/Kaggle-Starter-Kernel---Jigsaw-Multilingual-Toxic-Comment-Classification--Vmlldzo3NjE1MQ
- https://www.wandb.com/kaggle | https://www.youtube.com/watch?v=MFJnkgUTMrM&
- https://www.wandb.com/articles/model-explorations-and-hyperparameter-search-with-w-b-and-kubernetes (Robert Porsch)
- https://www.kaggle.com/robertehshi/basics-of-nfl-expost (Robert Lutz)
- https://app.wandb.ai/cayush/pytorchlightning/reports/Use-Pytorch-Lightning-with-Weights-%26-Biases--Vmlldzo2NjQ1Mw
- Automate Kaggle model training with Skorch and W&B:
- W&B Sweep: NeRF-–-Representing-Scenes-as-Neural-Radiance-Fields-for-View-Synthesis | Notebook 1 | Notebook 2
- Track model performance
- Hyper parameter sweeps
- GPU metrics
- Visualise model prediction
- Visualise Scikit Models
- Save and restore models
- Keras and W&B
- Pytorch and W&B
- Tracking experiments
- Implementing W&B sweeps: Sweeps | Notebook
- ...for more see this
- [AI/ML/DL Library / Package / Framework: applicable]
- [Inexpensive crowd-sourced infrastructure sharing: applicable]
- [Data cleaning: manual / no tools available]
- [Data querying: manual / tools available]
- [Data analytics: manual / tools available]
- [Data visualisation: manual / tools available]
- [Data validation: manual / no tools available]
- [Feature extraction: manual / no tools available]
- [Model creation: available]
- [Execute experiments: available]
- [Track experiments: available]
- [Hyper parameter tuning: available]
- [Model saving: available]
- [Visualisations: available]
Back to Programs and Tools.
Back to Data page.