Support of category (with lightgbm)

[PierreCe]

I have a lightgbm model trained with pandas dataframes some columns of it have the type "category" (and are underlying strings)
All of the steps for loading and compiling this model work well. But when I try to make a prediction, I always get an error:

• Either, I pass the dataframe, and I get the error: TypeError: Cannot initialize DMatrix from DataFrame
• Either, I pass the associated ndarray and I get the error: ValueError: Unrecognized NumPy type: object

The code is a copy paste of the quick start example, and the error is on the line :
dmat = treelite_runtime.DMatrix(X.values)

The data is load with :

X = pd.read_csv("french_motors_test.csv")
for column, typ in X.dtypes.iteritems():
    if not is_numeric_dtype(typ):
        X[column] = X[column].astype('category')

Data is coming from open data, so I could share the data and the model (but where?).

[hcho3]

Please convert the matrix to numeric type. Treelite does not support string category type.

[hcho3]

Closing in favor of dmlc/tl2cgen#10.