coci 0.2.0

Collective Observation on Causal Inference

COCI

Collective Observation on Causal Inferences

Coci makes it easy to observe the changes in predictions from machine learning models based on the alterations of feature values.

Why Coci?

Machine learning has always been understood as a black box algorithm, which makes the decision makers hesitant to trust the predictions from this approach.

Shap and Lime has unveiled a lot of mysteries around the effects of the presence of each feature on outcomes. However, these methods cannot show the change in outcomes when features are tweaked.

Coci takes it a step further, and reveals the effects on outcomes when changing feature values.

Installation

pip install coci==0.1.7

Summary Plot

Sample code

import coci

explainer = coci.TreeExplainer(model)

explainer.sensitivity(X_test, 
                    feature_names=feature_names,
                    split_num=2,
                    sample_size=300)

explainer.summary_plot(max_display=10)

Reading the summary plot

Trend Plot

Sample code

import coci 

explainer = coci.TreeExplainer(model)

explainer.sensitivity(X_test, 
                    feature_names=feature_names,
                    split_num=2,
                    sample_size=300)

explainer.trend_plot(feature_name=['要介護認定等基準時間（食事）'])

## or show by index
explainer.trend_plot(feature_index=[1276])

## or show the top ranked features
explainer.trend_plot(max_display=10)

Reading the trend plot