# Track your machine learning experiments
###### tags: `track`, `machine learning`, `trackor`, `mlflow`, `sacred`

Tools that you can use to track your machine learning experiments. I've split them into 3 styles in terms of their usages.

## Usage styles
1. 'logger' style
- Pro: Easy to use, exactly like using logging;
- Con: You need to decide what to track yourself
- [weights&biases](https://github.com/wandb/client)
- [losswise](https://github.com/Losswise/losswise-python)
- [commet](https://github.com/comet-ml/comet-examples)
- [modelchimp](https://github.com/ModelChimp/modelchimp)
- [randopt](https://github.com/seba-1511/randopt)
- [bnb](https://github.com/elanmart/bnb)

2. 'Sacred' style
- Pro: Clean, Flexible
- Con: sacred.automain is the trackor, which means that if you have your own pipeline, you need to re-structure them using the 'sacred' manner.
- [sacred](https://github.com/IDSIA/sacred)

3. Framework style
- Pro: Not much extra efforts since most info you want to track has been implemented in the framework
- Con: As for what wasn't implemented, you need to figure out how to track yourself
- Import Python Module
- [mlflow](https://github.com/mlflow/mlflow)
- [lore](https://github.com/instacart/lore)
- [pachyderm](https://github.com/pachyderm/pachyderm)
- 'Virtual Environment' style; cli interface
- [datmo](https://github.com/datmo/datmo)
- [studio](https://github.com/studioml/studio)
- [polyaxon](https://github.com/polyaxon/polyaxon)

## Reference
- [reddit](https://www.reddit.com/r/MachineLearning/comments/bx0apm/d_how_do_you_manage_your_machine_learning/)
- [reddit2](https://www.reddit.com/r/MachineLearning/comments/8zlj0w/d_what_tools_are_used_in_practice_to_schedule/)
- [blog](https://elanmart.github.io/2018-02-02-staying-sane/)