python xgboost explainer

inspired by xgboostExplainer of R language. addtionally fixed a bug about lambda

usage:

...
# calculate logit-odds of each node of each tree
tree_lst = xgb_exp.model2table(bst)

...

leaf_lst = bst.predict(sample, pred_leaf=True)
# sum the logit-odds contribution of each feature
dist = xgb_exp.logit_contribution(tree_lst, leaf_lst[0])

# print result
sum_logit = 0.0
for k in dist:
    sum_logit += dist[k]
    fn = feature_map[int(k[1:])] if k != "intercept" else k
    print(fn + ":", dist[k])

sample result

intercept: -1.4477233142227064

last_evaluation: -0.7779663104556692

average_montly_hours: -0.4159895659998113

time_spend_company: -0.25305760508286557

satisfaction_level: -0.6634059629069854

number_project: 0.02692557805173946

Work_accident: -1.2246923984381386

salary: 0.17029401462985444

sales: -0.023232836934958548

promotion_last_5years: 0.009318957559541447

you can check demo notebook for a full example