Matteo Interlandi

To reproduce: ``` import numpy as np from sklearn.ensemble import RandomForestClassifier from onnxconverter_common.data_types import FloatTensorType from onnxmltools.convert np.random.seed(0) X = np.random.rand(1, 1) X = np.array(X, dtype=np.float32) y = np.random.randint(1, size=1)...

I tried to run `LabelEncoder` and `OneHotEncoder` using TVM and it breaks. I think `OneHotEncoder` should work, but for `LabelEncoder` we need TVM VM.

Some of the test files are getting quite long (e.g., `sklearn_decision_tree`). We should refactor them in a way that makes them more easy to parse.

Sklearn models accept single inputs. If you pass a pandas Dataframe, for instance, it will create a single input. We instead consider Dataframes columns as different inputs, e.g., if we...

We should have at least one test for each of the available configurations. Now those tests are spread in different files.

Right now in the `convert` function we ask users to explicitly create a `extra_configs` dictionary. We can skip this process by using `**kwargs` instead. Similarly we expect users to pack...

The test `test_lgbm_classifier_random_forest` returns wrong results with the latest version of lgbm.