Convert Xgboost model to PMML 4.3 version

README.md

Overview

We are developing a model which needs to be integrated into an external system where it only supports PMML 4.3

Challenge

We were using the latest xgboost model and the latest sklearn2pmml and it could only produce PMML in 4.4 version.

These are some errors that we were getting

SEVERE: Failed to parse PKL
RuntimeError: The JPMML-SkLearn conversion application has failed. The Java executable should have printed more information about the failure into its standard output and/or standard error streams

SEVERE: Failed to parse learner
java.io.IOException: Expected 27-element array of zeroes, got [1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]

Downgrades

There are components that need to be downgraded.

1. xgboost

This post suggests that xgboost version needs to be in 1.0.0

2. sklearn2pmml

This post mentions 4.4 was introduced sometime in May 2020. I just looked through the releases and found a jar close to that date.

Dockerfile

FROM python

RUN apt-get update
RUN apt-get install -y openjdk-11-jdk
RUN python -m pip install --trusted-host pypi.org --trusted-host pypi.python.org --trusted-host files.pythonhosted.org sklearn xgboost==1.0.0 sklearn2pmml ipython
RUN wget https://github.com/jpmml/jpmml-xgboost/releases/download/1.4.0/jpmml-xgboost-executable-1.4.0.jar
RUN python train.py
RUN java -jar jpmml-xgboost-executable-1.3.16.jar --model-input iris.model --fmap-input Audit.fmap --target-name Adjusted --pmml-output XGBoostAudit.pmml

train.py

from sklearn import datasets
from sklearn import metrics
import pandas as pd
from sklearn import svm
from xgboost import XGBClassifier

iris = datasets.load_iris() #dataset loading

clf = XGBClassifier()

df = pd.DataFrame(data= np.c_[iris['data'], iris['target']],
                     columns=iris['feature_names'] + ['target'])

from sklearn.model_selection  import train_test_split
X_train, X_test, y_train, y_test = train_test_split(df[iris["feature_names"]], df['target'], test_size=0.2, random_state=42)

clf.fit(X_train, y_train)

clf.save_model("iris.model")

from sklearn2pmml.xgboost import make_feature_map

audit_fmap = make_feature_map(df, enable_categorical = False)
audit_fmap.save("Audit.fmap")