kserve/docs/samples/explanation/alibi/moviesentiment/movie_review_explanations.ipynb

Movie Review Explanations

We will use an SKLearn classifier built on movie sentiment data which predicts positive or negative sentiment for review text.

The Kfserving resource provides:

• A pretrained sklearn model stored on a Google bucket
• A Text Seldon Alibi Explainer. See the Alibi Docs for further details.

In [ ]:

!pygmentize moviesentiment.yaml

In [ ]:

!kubectl apply -f moviesentiment.yaml

In [ ]:

CLUSTER_IPS = !(kubectl -n istio-system get service istio-ingressgateway -o jsonpath='{.status.loadBalancer.ingress[0].ip}')
CLUSTER_IP = CLUSTER_IPS[0]
print(CLUSTER_IP)

In [ ]:

SERVICE_HOSTNAMES = !(kubectl get inferenceservice moviesentiment -o jsonpath='{.status.url}' | cut -d "/" -f 3)
SERVICE_HOSTNAME = SERVICE_HOSTNAMES[0]
print(SERVICE_HOSTNAME)

In [ ]:

import sys

sys.path.append("../")
from alibi_helper import *

In [ ]:

from alibi.datasets import fetch_movie_sentiment

movies = fetch_movie_sentiment()

In [ ]:

idxNeg = 37
idxPos = 5227
for idx in [idxNeg, idxPos]:
    print(movies.data[idx])
    show_prediction(
        predict(
            movies.data[idx : idx + 1],
            "moviesentiment",
            movies,
            SERVICE_HOSTNAME,
            CLUSTER_IP,
        )
    )

Get Explanation for Negative Prediction

In [ ]:

exp = explain(
    movies.data[idxNeg : idxNeg + 1], "moviesentiment", SERVICE_HOSTNAME, CLUSTER_IP
)

In [ ]:

show_anchors(exp["data"]["anchor"])

Show precision. How likely predictions using the Anchor features would produce the same result.

In [ ]:

show_bar([exp["data"]["precision"]], [""], "Precision")
show_bar([exp["data"]["coverage"]], [""], "Coverage")

In [ ]:

show_feature_coverage(exp["data"])

In [ ]:

show_examples(exp["data"], 0, movies)

In [ ]:

show_examples(exp["data"], 0, movies, False)

Get Explanation for Positive Example

In [ ]:

exp = explain(
    movies.data[idxPos : idxPos + 1], "moviesentiment", SERVICE_HOSTNAME, CLUSTER_IP
)

In [ ]:

show_anchors(exp["data"]["anchor"])

Show precision. How likely predictions using the Anchor features would produce the same result.

In [ ]:

show_bar([exp["data"]["precision"]], [""], "Precision")
show_bar([exp["data"]["coverage"]], [""], "Coverage")

In [ ]:

show_feature_coverage(exp["data"])

In [ ]:

show_examples(exp["data"], 0, movies)

In [ ]:

show_examples(exp["data"], 0, movies, False)

Teardown

In [ ]:

!kubectl delete -f moviesentiment.yaml

In [ ]: