Movie Review Sentiment Classification with DISTILL-BERT [scikit-learn + torch preprocessing]ยถ
Giskard is an open-source framework for testing all ML models, from LLMs to tabular models. Donโt hesitate to give the project a star on GitHub โญ๏ธ if you find it useful!
In this notebook, youโll learn how to create comprehensive test suites for your model in a few lines of code, thanks to Giskardโs open-source Python library.
Use-case:
Binary sentiment classification of moviesโ reviews.
Outline:
Detect vulnerabilities automatically with Giskardโs scan
Automatically generate & curate a comprehensive test suite to test your model beyond accuracy-related metrics
Upload your model to the Giskard Hub to:
Debug failing tests & diagnose issues
Compare models & decide which one to promote
Share your results & collect feedback from non-technical team members
Install dependenciesยถ
Make sure to install the giskard
[ ]:
%pip install giskard --upgrade
Import librariesยถ
[1]:
from pathlib import Path
from urllib.request import urlretrieve
import numpy as np
import pandas as pd
import torch
import transformers as ppb
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score
from sklearn.model_selection import train_test_split
from giskard import Model, Dataset, scan, testing, Suite
from giskard.client.giskard_client import GiskardClient
Define constantsยถ
[2]:
# Constants.
TARGET_COLUMN = "label"
TEXT_COLUMN = "text"
PRETRAINED_WEIGHTS_NAME = "distilbert-base-uncased"
RANDOM_STATE = 0
# Paths.
DATA_URL = "ftp://sys.giskard.ai/pub/unit_test_resources/movie_review_sentiment_classification_dataset/train.jsonl"
DATA_PATH = Path.home() / ".giskard" / "movie_review_sentiment_classification_dataset" / "train.jsonl"
Dataset preparationยถ
Load dataยถ
[ ]:
def fetch_from_ftp(url: str, file: Path) -> None:
if not file.parent.exists():
file.parent.mkdir(parents=True, exist_ok=True)
if not file.exists():
print(f"Downloading data from {url}")
urlretrieve(url, file)
print(f"Data was loaded!")
def load_data(**kwargs) -> pd.DataFrame:
"""Load data."""
fetch_from_ftp(DATA_URL, DATA_PATH)
df = pd.read_json(DATA_PATH, lines=True, **kwargs)
df = df.drop(columns="label_text")
return df
reviews_df = load_data(nrows=2000)
Train-Test splitยถ
[4]:
train_df, test_df = train_test_split(reviews_df, random_state=RANDOM_STATE)
Wrap dataset with Giskardยถ
To prepare for the vulnerability scan, make sure to wrap your dataset using Giskardโs Dataset class. More details here.
[5]:
giskard_dataset = Dataset(
df=test_df,
# A pandas.DataFrame that contains the raw data (before all the pre-processing steps) and the actual ground truth variable (target).
target=TARGET_COLUMN, # Ground truth variable.
name="Movie reviews dataset" # Optional.
)
Model buildingยถ
Define preprocessing stepsยถ
[ ]:
embedder = ppb.DistilBertModel.from_pretrained(PRETRAINED_WEIGHTS_NAME)
tokenizer = ppb.DistilBertTokenizer.from_pretrained(PRETRAINED_WEIGHTS_NAME)
def get_max_sequence_length(corpus: pd.Series) -> int:
"""Define a length of the longest tokenized document."""
max_length = max(len(tokenizer.encode(document, add_special_tokens=True)) for document in corpus)
return max_length
max_sequence_length = get_max_sequence_length(reviews_df[TEXT_COLUMN])
def tokenize_documents(corpus: pd.Series) -> torch.Tensor:
"""Tokenization step."""
tokens_matrix = corpus.apply(lambda document: tokenizer.encode(document, add_special_tokens=True)).values
tokens_matrix = torch.tensor(
[tokens_row + [0] * (max_sequence_length - len(tokens_row)) for tokens_row in tokens_matrix])
return tokens_matrix
def get_documents_embeddings(tokens_matrix: torch.Tensor) -> np.ndarray:
"""Calculate sentence embeddings using distill-BERT model."""
attention_mask = torch.where(tokens_matrix != 0, 1, 0)
embedder.eval()
with torch.no_grad():
tokens_representations = embedder(tokens_matrix, attention_mask=attention_mask)
# Take just 'cls token' embeddings, which represent whole sentence embedding.
documents_embeddings = tokens_representations[0][:, 0, :].numpy()
return documents_embeddings
def preprocess_text(df: pd.DataFrame) -> np.ndarray:
"""Preprocessing function to be also used in 'giskard.Model'."""
return get_documents_embeddings(tokenize_documents(df[TEXT_COLUMN]))
X_train, Y_train = preprocess_text(train_df), train_df.label
X_test, Y_test = preprocess_text(test_df), test_df.label
Build estimatorยถ
[ ]:
classifier = LogisticRegression()
classifier.fit(X_train, Y_train)
# Validate model.
train_score = classifier.score(X_train, Y_train)
print(f"Train accuracy: {train_score: .2f}")
test_score = classifier.score(X_test, Y_test)
print(f"Test accuracy: {test_score: .2f}")
Wrap model with Giskardยถ
To prepare for the vulnerability scan, make sure to wrap your model using Giskardโs Model class. You can choose to either wrap the prediction function (preferred option) or the model object. More details here.
[ ]:
def prediction_function(df: pd.DataFrame) -> np.ndarray:
x = preprocess_text(df)
return classifier.predict_proba(x)
giskard_model = Model(
model=prediction_function,
# A prediction function that encapsulates all the data pre-processing steps and that could be executed with the dataset used by the scan.
model_type="classification", # Either regression, classification or text_generation.
name="Movie reviews sentiment classifier", # Optional.
classification_labels=classifier.classes_.tolist(),
# Their order MUST be identical to the prediction_function's output order.
feature_names=[TEXT_COLUMN], # Default: all columns of your dataset.
# classification_threshold=0.5 # Default: 0.5.
)
Y_test_pred_wrapped = giskard_model.predict(giskard_dataset).prediction
wrapped_test_score = accuracy_score(Y_test, Y_test_pred_wrapped)
print(f"Wrapped test accuracy: {wrapped_test_score: .2f}")
Detect vulnerabilities in your modelยถ
Scan your model for vulnerabilities with Giskardยถ
Giskardโs scan allows you to detect vulnerabilities in your model automatically. These include performance biases, unrobustness, data leakage, stochasticity, underconfidence, ethical issues, and more. For detailed information about the scan feature, please refer to our scan documentation.
[ ]:
results = scan(giskard_model, giskard_dataset)
[10]:
display(results)
Generate comprehensive test suites automatically for your modelยถ
Generate test suites from the scanยถ
The objects produced by the scan can be used as fixtures to generate a test suite that integrate all detected vulnerabilities. Test suites allow you to evaluate and validate your modelโs performance, ensuring that it behaves as expected on a set of predefined test cases, and to identify any regressions or issues that might arise during development or updates.
[11]:
test_suite = results.generate_test_suite("My first test suite")
test_suite.run()
Executed 'Overconfidence on data slice โ`avg_whitespace(text)` >= 0.172โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x14aaccee0>, 'dataset': <giskard.datasets.base.Dataset object at 0x14aaa4550>, 'slicing_function': <giskard.slicing.text_slicer.MetadataSliceFunction object at 0x14d3fd870>, 'threshold': 0.4033333333333333, 'p_threshold': 0.5}:
Test failed
Metric: 0.43
Executed 'Precision on data slice โ`text` contains "movie"โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x14aaccee0>, 'dataset': <giskard.datasets.base.Dataset object at 0x14aaa4550>, 'slicing_function': <giskard.slicing.slice.QueryBasedSliceFunction object at 0x150f8c1f0>, 'threshold': 0.808091286307054}:
Test failed
Metric: 0.64
[11]:
Customize your suite by loading objects from the Giskard catalogยถ
The Giskard open source catalog will enable to load:
Tests such as metamorphic, performance, prediction & data drift, statistical tests, etc
Slicing functions such as detectors of toxicity, hate, emotion, etc
Transformation functions such as generators of typos, paraphrase, style tune, etc
To create custom tests, refer to this page.
For demo purposes, we will load a simple unit test (test_f1) that checks if the test F1 score is above the given threshold. For more examples of tests and functions, refer to the Giskard catalog.
[ ]:
test_suite.add_test(testing.test_f1(model=giskard_model, dataset=giskard_dataset, threshold=0.7)).run()
Debug and interact with your tests in the Giskard Hubยถ
At this point, youโve created a test suite that is highly specific to your domain & use-case. Failing tests can be a pain to debug, which is why we encourage you to head over to the Giskard Hub.
Play around with a demo of the Giskard Hub on HuggingFace Spaces using this link.
More than just debugging tests, the Giskard Hub allows you to:
Compare models to decide which model to promote
Automatically create additional domain-specific tests through our automated model insights feature
Share your test results with team members and decision makers
The Giskard Hub can be deployed easily on HuggingFace Spaces.
Hereโs a sneak peek of automated model insights on a credit scoring classification model.
Upload your test suite to the Giskard Hubยถ
The entry point to the Giskard Hub is the upload of your test suite. Uploading the test suite will automatically save the model, dataset, tests, slicing & transformation functions to the Giskard Hub.
[ ]:
# Create a Giskard client after having install the Giskard server (see documentation)
api_key = "<Giskard API key>" #This can be found in the Settings tab of the Giskard hub
#hf_token = "<Your Giskard Space token>" #If the Giskard Hub is installed on HF Space, this can be found on the Settings tab of the Giskard Hub
client = GiskardClient(
url="http://localhost:19000", # Option 1: Use URL of your local Giskard instance.
# url="<URL of your Giskard hub Space>", # Option 2: Use URL of your remote HuggingFace space.
key=api_key,
# hf_token=hf_token # Use this token to access a private HF space.
)
project_key = "my_project"
my_project = client.create_project(project_key, "PROJECT_NAME", "DESCRIPTION")
# Upload to the project you just created
test_suite.upload(client, project_key)
Download a test suite from the Giskard Hubยถ
After curating your test suites with additional tests on the Giskard Hub, you can easily download them back into your environment. This allows you to:
Check for regressions after training a new model
Automate the test suite execution in a CI/CD pipeline
Compare several models during the prototyping phase
[ ]:
test_suite_downloaded = Suite.download(client, project_key, suite_id=...)
test_suite_downloaded.run()