Fake/real news classification [tensorflow (keras)]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 classification of news being fake or real, based on their text.
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
[21]:
%pip install giskard --upgrade
Import librariesยถ
[ ]:
import os
import string
from pathlib import Path
from urllib.request import urlretrieve
import numpy as np
import pandas as pd
from keras.layers import Dense, Embedding, LSTM
from keras.models import Sequential
from keras.optimizers import Adam
from keras.preprocessing.text import Tokenizer
from keras.utils import pad_sequences
from nltk.corpus import stopwords
from sklearn.metrics import accuracy_score
from sklearn.model_selection import train_test_split
from typing import Tuple, Callable
from giskard import Dataset, Model, scan, testing, GiskardClient, Suite
Notebook-level settingsยถ
[2]:
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '1'
Define constantsยถ
[3]:
# Constants.
MAX_TOKENS = 20000
MAX_SEQUENCE_LENGTH = 100
N_ROWS = 1000
STOPWORDS = stopwords.words('english')
TEXT_COLUMN_NAME = "text"
TARGET_COLUMN_NAME = "isFake"
RANDOM_SEED = 0
# Paths.
DATA_URL = "ftp://sys.giskard.ai/pub/unit_test_resources/fake_real_news_dataset/{}"
DATA_PATH = Path.home() / ".giskard" / "fake_real_news_dataset"
Dataset preparationยถ
Load and preprocess dataยถ
[ ]:
def fetch_from_ftp(url: str, file: Path) -> None:
"""Helper to fetch data from the FTP server."""
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 fetch_dataset() -> None:
"""Gradually fetch all necessary files from the FTP server."""
files_to_fetch = ("Fake.csv", "True.csv", "glove_100d.txt")
for file_name in files_to_fetch:
fetch_from_ftp(DATA_URL.format(file_name), DATA_PATH / file_name)
def load_data(**kwargs) -> pd.DataFrame:
"""Load data."""
real_df = pd.read_csv(DATA_PATH / "True.csv", **kwargs)
fake_df = pd.read_csv(DATA_PATH / "Fake.csv", **kwargs)
# Create target column.
real_df[TARGET_COLUMN_NAME] = 0
fake_df[TARGET_COLUMN_NAME] = 1
# Combine dfs.
full_df = pd.concat([real_df, fake_df])
full_df.drop(columns=["subject", "date"], inplace=True)
return full_df
fetch_dataset()
news_df = load_data(nrows=N_ROWS)
Train-test splitยถ
[5]:
X_train, X_test, Y_train, Y_test = train_test_split(news_df[["title", TEXT_COLUMN_NAME]], news_df[TARGET_COLUMN_NAME],
random_state=RANDOM_SEED)
Wrap dataset with Giskardยถ
To prepare for the vulnerability scan, make sure to wrap your dataset using Giskardโs Dataset class. More details here.
[6]:
raw_data = pd.concat([X_test, Y_test], axis=1)
giskard_dataset = Dataset(
df=raw_data,
# A pandas.DataFrame that contains the raw data (before all the pre-processing steps) and the actual ground truth variable (target).
target=TARGET_COLUMN_NAME, # Ground truth variable.
name="fake_and_real_news" # Optional.
)
Model buildingยถ
Define preprocessing stepsยถ
[7]:
def prepare_text(df: pd.DataFrame) -> np.ndarray:
"""Perform text-data cleaning: punctuation and stop words removal."""
# Merge text data into single column.
df[TEXT_COLUMN_NAME] = df[TEXT_COLUMN_NAME] + " " + df.title
df.drop(columns=["title"], inplace=True)
# Remove punctuation.
df[TEXT_COLUMN_NAME] = df[TEXT_COLUMN_NAME].apply(
lambda text: text.translate(str.maketrans('', '', string.punctuation)))
# Remove stop words.
df[TEXT_COLUMN_NAME] = df[TEXT_COLUMN_NAME].apply(
lambda sentence: ' '.join([_word for _word in sentence.split() if _word.lower() not in STOPWORDS]))
return df[TEXT_COLUMN_NAME]
X_train_prepared = prepare_text(X_train)
X_test_prepared = prepare_text(X_test)
def init_tokenizer() -> Tuple[Callable, Tokenizer]:
"""Initialize tokenization function with the Tokenizer in it's outer-scope."""
tokenizer = Tokenizer(num_words=MAX_TOKENS)
tokenizer.fit_on_texts(X_train_prepared)
def tokenization_closure(df: pd.DataFrame) -> pd.DataFrame:
tokenized = tokenizer.texts_to_sequences(df)
return pad_sequences(tokenized, maxlen=MAX_SEQUENCE_LENGTH)
return tokenization_closure, tokenizer
tokenize, text_tokenizer = init_tokenizer()
X_train_tokens = tokenize(X_train_prepared)
X_test_tokens = tokenize(X_test_prepared)
Create embeddings matrixยถ
[8]:
def parse_line(word: str, *arr: list) -> Tuple[str, np.ndarray]:
"""Parse line from the file with embeddings.
The first value of the line is the word and the rest values are related glove embedding: (<word>, 0.66, 0.23, ...)."""
return word, np.asarray(arr, dtype='float32')
def init_embeddings_matrix(embeddings_dict: dict) -> np.ndarray:
"""Init a matrix, where each row is an embedding vector."""
num_embeddings = min(MAX_TOKENS, len(text_tokenizer.word_index))
stacked_embeddings = np.stack(list(embeddings_dict.values()))
embeddings_mean, embeddings_std, embeddings_dimension = stacked_embeddings.mean(), stacked_embeddings.std(), \
stacked_embeddings.shape[1]
embeddings_matrix = np.random.normal(embeddings_mean, embeddings_std, (num_embeddings, embeddings_dimension))
return embeddings_matrix
def get_embeddings_matrix() -> np.ndarray:
"""Create matrix, where each row is an embedding of a specific word."""
# Load glove embeddings.
embeddings_dict = dict(parse_line(*line.rstrip().rsplit(' ')) for line in open(DATA_PATH / "glove_100d.txt"))
# Create embeddings matrix with glove word vectors.
embeddings_matrix = init_embeddings_matrix(embeddings_dict)
for word, idx in text_tokenizer.word_index.items():
if idx >= MAX_TOKENS:
continue
embedding_vector = embeddings_dict.get(word, None)
if embedding_vector is not None:
embeddings_matrix[idx] = embedding_vector
return embeddings_matrix
embed_matrix = get_embeddings_matrix()
Build estimatorยถ
[ ]:
def init_model() -> Sequential:
"""Initialize new TF model."""
# Define model container.
model = Sequential()
# Non-trainable embedding layer.
model.add(Embedding(MAX_TOKENS, output_dim=100, weights=[embed_matrix], input_length=MAX_SEQUENCE_LENGTH,
trainable=False))
# LSTM stage.
model.add(LSTM(units=32, return_sequences=True, recurrent_dropout=0.25, dropout=0.25))
model.add(LSTM(units=16, recurrent_dropout=0.1, dropout=0.1))
# Dense stage.
model.add(Dense(units=16, activation='relu'))
model.add(Dense(units=1, activation='sigmoid'))
# Build model.
model.compile(optimizer=Adam(learning_rate=0.01), loss='binary_crossentropy', metrics=['accuracy'])
return model
# Fit model.
n_epochs = 5
batch_size = 256
classifier = init_model()
_ = classifier.fit(X_train_tokens, Y_train, batch_size=batch_size, validation_data=(X_test_tokens, Y_test),
epochs=n_epochs)
train_metric = classifier.evaluate(X_train_tokens, Y_train, verbose=0)[1]
test_metric = classifier.evaluate(X_test_tokens, Y_test, verbose=0)[1]
print(f"Train accuracy: {train_metric: .4f}")
print(f"Test accuracy: {test_metric: .4f}")
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:
"""Define a prediction function for giskard.Model."""
tokens = tokenize(prepare_text(df))
return classifier.predict(tokens, verbose=0)
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="fake_real_news_classification", # Optional.
feature_names=["title", TEXT_COLUMN_NAME], # Default: all columns of your dataset.
classification_labels=[0, 1], # Their order MUST be identical to the prediction_function's output order.
# classification_threshold=0.5 # Default: 0.5
)
# Validate wrapped model.
Y_test_pred_wrapper = giskard_model.predict(giskard_dataset).prediction
wrapped_test_metric = accuracy_score(Y_test, Y_test_pred_wrapper)
print(f"Wrapped test accuracy: {wrapped_test_metric: .4f}")
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)
[12]:
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.
[13]:
test_suite = results.generate_test_suite("My first test suite")
test_suite.run()
Executed 'Recall on data slice โ`text` contains "october"โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.slice.QueryBasedSliceFunction object at 0x1426c2770>, 'threshold': 0.9306910569105691}:
Test failed
Metric: 0.87
Executed 'Accuracy on data slice โ`avg_whitespace(title)` >= 0.147 AND `avg_whitespace(title)` < 0.152โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.text_slicer.MetadataSliceFunction object at 0x13e0318a0>, 'threshold': 0.9386}:
Test failed
Metric: 0.91
Executed 'Recall on data slice โ`text_length(title)` >= 92.500 AND `text_length(title)` < 97.500โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.text_slicer.MetadataSliceFunction object at 0x13dfb8d90>, 'threshold': 0.9306910569105691}:
Test failed
Metric: 0.9
Executed 'Recall on data slice โ`text` contains "decision"โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.slice.QueryBasedSliceFunction object at 0x142206ef0>, 'threshold': 0.9306910569105691}:
Test failed
Metric: 0.91
Executed 'Recall on data slice โ`text` contains "texas"โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.slice.QueryBasedSliceFunction object at 0x141eadff0>, 'threshold': 0.9306910569105691}:
Test failed
Metric: 0.91
Executed 'Recall on data slice โ`text` contains "life"โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.slice.QueryBasedSliceFunction object at 0x14245ee90>, 'threshold': 0.9306910569105691}:
Test failed
Metric: 0.92
Executed 'Accuracy on data slice โ`text` contains "guilty"โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.slice.QueryBasedSliceFunction object at 0x1426cab90>, 'threshold': 0.9386}:
Test failed
Metric: 0.93
Executed 'Recall on data slice โ`text` contains "september"โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.slice.QueryBasedSliceFunction object at 0x1426553f0>, 'threshold': 0.9306910569105691}:
Test failed
Metric: 0.92
Executed 'Accuracy on data slice โ`avg_word_length(text)` < 4.884 AND `avg_word_length(text)` >= 4.835โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.text_slicer.MetadataSliceFunction object at 0x13e1099f0>, 'threshold': 0.9386}:
Test failed
Metric: 0.93
Executed 'Accuracy on data slice โ`avg_digits(text)` >= 0.007 AND `avg_digits(text)` < 0.008โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.text_slicer.MetadataSliceFunction object at 0x13e1082b0>, 'threshold': 0.9386}:
Test failed
Metric: 0.93
Executed 'Accuracy on data slice โ`text_length(text)` >= 2446.500 AND `text_length(text)` < 2572.500โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.text_slicer.MetadataSliceFunction object at 0x13e108a90>, 'threshold': 0.9386}:
Test failed
Metric: 0.93
Executed 'Accuracy on data slice โ`title` contains "video"โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.slice.QueryBasedSliceFunction object at 0x141fbeb90>, 'threshold': 0.9386}:
Test failed
Metric: 0.94
Executed 'Recall on data slice โ`avg_digits(text)` >= 0.003 AND `avg_digits(text)` < 0.004โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.text_slicer.MetadataSliceFunction object at 0x13e10ad10>, 'threshold': 0.9306910569105691}:
Test failed
Metric: 0.93
Executed 'Accuracy on data slice โ`text` contains "elections"โ' with arguments {'model': <giskard.models.function.PredictionFunctionModel object at 0x13a691b40>, 'dataset': <giskard.datasets.base.Dataset object at 0x10ca8e170>, 'slicing_function': <giskard.slicing.slice.QueryBasedSliceFunction object at 0x1421a2ad0>, 'threshold': 0.9386}:
Test failed
Metric: 0.94
[13]:
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()