This example is going to use an employee attrition dataset to help us train a model to determine the likelihood of an employee leaving a company. Beyond this, we will generate an Explanations Dashboard which helps indicate the local and global feature importance with respect to the all-up prediction.

We're training a sklearn regression model here.

The explainer is created in the script.

You can deploy the model and explainer from the CLI or the SDK. There are examples for both in this folder.

import requests
import json

headers = {'Content-Type':'application/json'}

# send request to service
resp = requests.post(service.scoring_uri, sample, headers=headers)
print("prediction:", resp.text)
result = json.loads(resp.text)

This will return you a JSON file which contains the prediction (attrition / no attrition) and the weighted feature importance score.

If you want to reproduce this chart yourself, you can use this convenient script:

result = json.loads(resp.text)
labels = x_test[:1].columns.to_list()
objects = labels
y_pos = np.arange(len(objects))
performance = result["local_importance_values"][0][0]

plt.bar(y_pos, performance, align='center', alpha=0.5)
plt.xticks(y_pos, objects)
locs, labels = plt.xticks()
plt.setp(labels, rotation=90)
plt.ylabel('Feature impact - leaving vs not leaving')
plt.title('Local feature importance for prediction')

plt.show()