Overview
Concrete tool for querying KServe prediction and explanation endpoints and rendering interactive Alibi explainability visualizations provided by the KServe sample code.
Description
This module provides a collection of helper functions designed for use in Jupyter notebooks that demonstrate Alibi-based model explanations. It includes functions for: sending REST requests to KServe predict and explain endpoints (predict(), explain()), feature mapping (getFeatures()), and rendering interactive visualizations using Plotly bar charts (show_bar(), show_feature_coverage()) and IPython display utilities (show_anchors(), show_examples(), show_prediction(), show_row()).
Usage
Import these helper functions in a Jupyter notebook when running Alibi explanation demos against a deployed KServe model to make predictions, request explanations, and visualize anchor-based explanation results.
Code Reference
Source Location
Signature
def getFeatures(X, cmap):
...
def predict(X, name, ds, svc_hostname, cluster_ip):
...
def explain(X, name, svc_hostname, cluster_ip):
...
def show_bar(X, labels, title):
...
def show_feature_coverage(exp):
...
def show_anchors(names):
...
def show_examples(exp, fidx, ds, covered=True):
...
def show_prediction(prediction):
...
def show_row(X, ds):
...
Import
from alibi_helper import predict, explain, show_bar, show_anchors, show_examples
I/O Contract
Inputs
predict()
| Name |
Type |
Required |
Description
|
| X |
list |
Yes |
Input instances for prediction
|
| name |
str |
Yes |
Name of the KServe model
|
| ds |
object |
Yes |
Dataset object with target_names attribute
|
| svc_hostname |
str |
Yes |
Service hostname for the Host header
|
| cluster_ip |
str |
Yes |
Cluster IP address for the HTTP request
|
explain()
| Name |
Type |
Required |
Description
|
| X |
list |
Yes |
Input instances to explain
|
| name |
str |
Yes |
Name of the KServe model
|
| svc_hostname |
str |
Yes |
Service hostname for the Host header
|
| cluster_ip |
str |
Yes |
Cluster IP address for the HTTP request
|
show_bar()
| Name |
Type |
Required |
Description
|
| X |
list |
Yes |
Bar chart x-axis values (e.g., precision scores)
|
| labels |
list |
Yes |
Bar chart y-axis labels (e.g., feature names)
|
| title |
str |
Yes |
Title for the bar chart
|
show_examples()
| Name |
Type |
Required |
Description
|
| exp |
dict |
Yes |
Explanation result containing anchor and raw example data
|
| fidx |
int |
Yes |
Feature index for which to display examples
|
| ds |
object |
Yes |
Dataset object, optionally with feature_names attribute
|
| covered |
bool |
No |
If True, shows covered examples; if False, shows uncovered examples (default True)
|
Outputs
predict()
| Name |
Type |
Description
|
| prediction |
str or list |
The target name for the predicted class, or empty list on failure
|
explain()
| Name |
Type |
Description
|
| explanation |
dict or list |
JSON explanation response from the KServe explain endpoint, or empty list on failure
|
show_examples()
| Name |
Type |
Description
|
| examples |
pd.DataFrame |
DataFrame of covered or uncovered example rows
|
Usage Examples
Basic Usage
from alibi_helper import predict, explain, show_anchors, show_prediction
# Make a prediction
svc_hostname = "my-model.default.example.com"
cluster_ip = "10.0.0.1"
X = [[6.8, 2.8, 4.8, 1.4]]
prediction = predict(X, "iris-model", iris_dataset, svc_hostname, cluster_ip)
show_prediction(prediction)
# Get and display explanation
explanation = explain(X, "iris-model", svc_hostname, cluster_ip)
show_anchors(explanation["anchor"])
Related Pages
