nyoka support for xgboost>=1.0.0

[Robert-Christensen-visa]

I have been able to successfully export xgboost models using nyoka when I was using xgboost<1.0.0. When I upgrade to xgboost>=1.0.0, exporting the xgboost model fails.

The following is from xgboost example 1 I get an error when doing the export.

import pandas as pd
from sklearn import datasets
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from xgboost import XGBRegressor,XGBClassifier

iris = datasets.load_iris()
irisd = pd.DataFrame(iris.data,columns=iris.feature_names)
irisd['Species'] = iris.target

features = irisd.columns.drop('Species')
target = 'Species'

pipeline_obj = Pipeline([
    ('Xgbc',XGBClassifier())
])

pipeline_obj.fit(irisd[features],irisd[target])

from nyoka import xgboost_to_pmml

xgboost_to_pmml(pipeline_obj,features,target,"xgbc_pmml.pmml")

The stack trace thrown from running to the last line:

---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
<ipython-input-2-5a7704b3a862> in <module>
      1 from nyoka import xgboost_to_pmml
      2 
----> 3 xgboost_to_pmml(pipeline_obj,features,target,"xgbc_pmml.pmml")

~/condapv/envs/new_python/lib/python3.6/site-packages/nyoka/xgboost/xgboost_to_pmml.py in xgboost_to_pmml(pipeline, col_names, target_name, pmml_f_name, model_name, description)
     61                                       mining_imp_val,
     62                                       categoric_values,
---> 63                                       model_name)
     64         pmml = pml.PMML(
     65             version=PMML_SCHEMA.VERSION.value,

~/condapv/envs/new_python/lib/python3.6/site-packages/nyoka/xgboost/xgboost_to_pmml.py in get_PMML_kwargs(model, derived_col_names, col_names, target_name, mining_imp_val, categoric_values, model_name)
    103                                                       mining_imp_val,
    104                                                       categoric_values,
--> 105                                                       model_name)}
    106     return algo_kwargs
    107 

~/condapv/envs/new_python/lib/python3.6/site-packages/nyoka/xgboost/xgboost_to_pmml.py in get_ensemble_models(model, derived_col_names, col_names, target_name, mining_imp_val, categoric_values, model_name)
    138     mining_models.append(pml.MiningModel(
    139         modelName=model_name if model_name else "XGBoostModel",
--> 140         Segmentation=get_outer_segmentation(model, derived_col_names, col_names, target_name, mining_imp_val,categoric_values,model_name),
    141         **model_kwargs
    142     ))

~/condapv/envs/new_python/lib/python3.6/site-packages/nyoka/xgboost/xgboost_to_pmml.py in get_outer_segmentation(model, derived_col_names, col_names, target_name, mining_imp_val, categoric_values, model_name)
    178         segmentation = pml.Segmentation(
    179             multipleModelMethod=get_multiple_model_method(model),
--> 180             Segment=get_segments(model, derived_col_names, col_names, target_name, mining_imp_val,categoric_values,model_name)
    181         )
    182     return segmentation

~/condapv/envs/new_python/lib/python3.6/site-packages/nyoka/xgboost/xgboost_to_pmml.py in get_segments(model, derived_col_names, col_names, target_name, mining_imp_val, categoric_values, model_name)
    211     segments = None
    212     if 'XGBClassifier'  in str(model.__class__):
--> 213         segments=get_segments_for_xgbc(model, derived_col_names, col_names, target_name, mining_imp_val,categoric_values,model_name)
    214     elif 'XGBRegressor' in str(model.__class__):
    215         segments=get_segments_for_xgbr(model, derived_col_names, col_names, target_name, mining_imp_val,categoric_values)

~/condapv/envs/new_python/lib/python3.6/site-packages/nyoka/xgboost/xgboost_to_pmml.py in get_segments_for_xgbc(model, derived_col_names, feature_names, target_name, mining_imp_val, categoric_values, model_name)
    479                                                        mining_schema_for_1st_segment, out, index)
    480             segments.append(segments_equal_to_class)
--> 481         reg_model=sklToPmml.get_regrs_models(model,oField,oField,target_name,mining_imp_val,categoric_values,model_name)[0]
    482         reg_model.normalizationMethod=REGRESSION_NORMALIZATION_METHOD.SOFTMAX.value
    483         last_segment = pml.Segment(True_=pml.True_(), id=model.n_classes_ + 1,

~/condapv/envs/new_python/lib/python3.6/site-packages/nyoka/skl/skl_to_pmml.py in get_regrs_models(model, derived_col_names, col_names, target_name, mining_imp_val, categoric_values, model_name)
   1685     regrs_models.append(pml.RegressionModel(
   1686         modelName=model_name if model_name else model.__class__.__name__,
-> 1687         RegressionTable=get_regrs_tabl(model, derived_col_names, target_name, categoric_values),
   1688         **model_kwargs
   1689     ))

~/condapv/envs/new_python/lib/python3.6/site-packages/nyoka/skl/skl_to_pmml.py in get_regrs_tabl(model, feature_names, target_name, categoric_values)
   1714     """
   1715     merge = list()
-> 1716     if hasattr(model, 'intercept_'):
   1717         import numpy as np
   1718         func_name = get_mining_func(model)

~/condapv/envs/new_python/lib/python3.6/site-packages/xgboost/sklearn.py in intercept_(self)
    743                                  .format(self.booster))
    744         b = self.get_booster()
--> 745         return np.array(json.loads(b.get_dump(dump_format='json')[0])['bias'])
    746 
    747 

KeyError: 'bias'

I am using nyoka 4.2.0 and xgboost 1.0.0.

[Nirmal-Neel]

Hi @Robert-Christensen-visa, Nyoka supports xgboost <= 0.90. We will be adding support for latest release very soon. To know about the supported libraries along with version, please refer to Libraries Supported by Nyoka

[victornoel]

Hi @Nirmal-Neel, any news about supporting recent xgboost and sklearn? Cheers

[Nirmal-Neel]

Hi @victornoel, we are still not there yet. But that is not so far.

[victornoel]

Very nice, thank you for the feedback!

[Nirmal-Neel]

Hi @victornoel, nyoka release 5.2.0 adds support for xgboost 1.x.x.

[victornoel]

@Nirmal-Neel really cool, thx for the heads-up!

While I have you here, any plan for sklearn 1.x? With both xgboost 1.x and sklearn 1.x supported, we could have a suite of stabilized tools!