Fixes #234, add converter for IsolationForest

[xadupre]

No description provided.

[lgtm-com]

This pull request introduces 2 alerts when merging 5799d65 into 2bd006e - view on LGTM.com

new alerts:

• 2 for Unused import

[m3at]

@xadupre thank you for working on this. I tried your branch, but I cannot run the created model with onnxruntime, am I missing something?

Here is a snippet adapted from your test:

import numpy as np
import onnxruntime as rt
from sklearn.ensemble import IsolationForest
from skl2onnx import to_onnx

data = np.random.randn(100, 4).astype(np.float32)
model = IsolationForest(n_estimators=2, random_state=0).fit(data)

# No issue to create the model
model_onnx = to_onnx(model, data)

with open("/tmp/model.onnx", "wb") as f:
    f.write(model_onnx.SerializeToString())

# NotImplemented error
sess = rt.InferenceSession("/tmp/model.onnx")

Which raise

NotImplemented: [ONNXRuntimeError] : 9 : NOT_IMPLEMENTED : Could not find an implementation for the node path_length0_LabelEncoder:LabelEncoder(2)

Which itself point to an issue coming from around here.

Any clue?

[xadupre]

Which version of onnxruntime are you using?

[m3at]

The latest version, 1.4 (installed with pip)

[xadupre]

onnxruntime does not implement parameter keys_floats of LabelEncoder. I used keys_int64s instead but it adds a cast to the graph. It should fix your issue.

[m3at]

It does indeed. Thank you for the quick fix 👍