https://github.com/mlflow/mlflow
- UI (tracking and comparing)
- Run ML projects with different parameters
- Serving
- reproduce with conda config file
- weak store (store in local file, slow to load)
- ugly code
- do not support save artifacts to user server
- model serving is weak
https://github.com/polyaxon/polyaxon
- use Docker
- UI
- Notebook plugins
- Tensorboard
- Hyperparameters tuning
- coupled to k8s
- hard to deploy and debug
- built for large data version control
- data saved in cloud or local or HDFS
- support reproduce experiments with user histroy
- not easy to use when user want to record the experiments
- beta