mlem.api.save: ML frameworks and Data formats support

[aguschin]

This is an epic to collect all ML frameworks / Data formats we would like to support, when you call mlem.api.save(mymodel, "model") or mlem.api.save(mydataset, "dataset").

ML frameworks:

• Python functions (could be any function, but is treated as a ML model by MLEM. Specific framework support below means MLEM uses built-in framework serialization methods when possible)
• Sklearn Models
• Pytorch (including Keras)
• Catboost
• XGBoost
• LightGBM
• Tensorflow 2.0 (including Keras)
• Onnx
• Sparse matrices in scipy Support scipy sparse matrices #540
• Sklearn Transformers Support saving sklearn transformers #514
• Sklearn's CountVectorizer
• Sklearn's IsolationForest
• Chainer
• MXNet
• Huggingface Transformers Support huggingface transformers models and datasets #265
• fast.ai
• Pytorch Lightning

Data formats:

• Simple Python types (str, int, float, dict, etc)
• Pandas
• Numpy
• Lightgbm
• XGBoost

Please feel free to post a comment if you need something we don't support yet :)

[madhur-tandon]

we support lightgbm as well, updated the description.

[Snehashish-s]

I am getting a "mlem.core.errors.HookNotFound: No suitable ModelHook for object of type "IsolationForest" " exception when saving Isolationforest model, any way it can be solved ?

[aguschin]

Hi @Snehashish-s, rn MLEM doesn't support this, but it's pretty trivial to add by adding OutlierMixin to this line https://github.com/iterative/mlem/blob/main/mlem/contrib/sklearn.py#L31. Would you like contribute this and add a test?

[aguschin]

@Snehashish-s, I just supported IsolationForest in https://github.com/iterative/mlem/releases/tag/0.4.14