Note
Go to the end to download the full example code
Feature Drift#
This notebooks provides an overview for using and understanding feature drift check.
Structure:
What is a feature drift?#
Drift is simply a change in the distribution of data over time, and it is also one of the top reasons why machine learning model’s performance degrades over time.
Feature drift is a data drift that occurs in a single feature in the dataset.
For more information on drift, please visit our Drift Guide.
How Deepchecks Detects Feature Drift#
This check detects feature drift by using univariate measures on each feature column separately. Another possible method for drift detection is by a domain classifier which is used in the Multivariate Drift check.
Generate data & model#
Let’s generate a mock dataset of 2 categorical and 2 numerical features
import numpy as np
import pandas as pd
np.random.seed(42)
train_data = np.concatenate([np.random.randn(1000,2), np.random.choice(a=['apple', 'orange', 'banana'], p=[0.5, 0.3, 0.2], size=(1000, 2))], axis=1)
test_data = np.concatenate([np.random.randn(1000,2), np.random.choice(a=['apple', 'orange', 'banana'], p=[0.5, 0.3, 0.2], size=(1000, 2))], axis=1)
df_train = pd.DataFrame(train_data, columns=['numeric_without_drift', 'numeric_with_drift', 'categorical_without_drift', 'categorical_with_drift'])
df_test = pd.DataFrame(test_data, columns=df_train.columns)
df_train = df_train.astype({'numeric_without_drift': 'float', 'numeric_with_drift': 'float'})
df_test = df_test.astype({'numeric_without_drift': 'float', 'numeric_with_drift': 'float'})
Insert drift into test:#
Now, we insert a synthetic drift into 2 columns in the dataset
df_test['numeric_with_drift'] = df_test['numeric_with_drift'].astype('float') + abs(np.random.randn(1000)) + np.arange(0, 1, 0.001) * 4
df_test['categorical_with_drift'] = np.random.choice(a=['apple', 'orange', 'banana', 'lemon'], p=[0.5, 0.25, 0.15, 0.1], size=(1000, 1))
Training a model#
Now, we are building a dummy model (the label is just a random numerical column). We preprocess our synthetic dataset so categorical features are being encoded with an OrdinalEncoder
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import OrdinalEncoder
from sklearn.tree import DecisionTreeClassifier
from deepchecks.tabular import Dataset
model = Pipeline([
('handle_cat', ColumnTransformer(
transformers=[
('num', 'passthrough',
['numeric_with_drift', 'numeric_without_drift']),
('cat',
Pipeline([
('encode', OrdinalEncoder(handle_unknown='use_encoded_value', unknown_value=-1)),
]),
['categorical_with_drift', 'categorical_without_drift'])
]
)),
('model', DecisionTreeClassifier(random_state=0, max_depth=2))]
)
label = np.random.randint(0, 2, size=(df_train.shape[0],))
cat_features = ['categorical_without_drift', 'categorical_with_drift']
df_train['target'] = label
train_dataset = Dataset(df_train, label='target', cat_features=cat_features)
model.fit(train_dataset.data[train_dataset.features], label)
label = np.random.randint(0, 2, size=(df_test.shape[0],))
df_test['target'] = label
test_dataset = Dataset(df_test, label='target', cat_features=cat_features)
Run the check#
Let’s run deepchecks’ feature drift check and see the results
from deepchecks.tabular.checks import FeatureDrift
check = FeatureDrift()
result = check.run(train_dataset=train_dataset, test_dataset=test_dataset, model=model)
result.show()
Observe the check’s output#
As we see from the results, the check detects and returns the drift score per feature. As we expect, the features that were manually manipulated to contain a strong drift in them were detected.
In addition to the graphs, each check returns a value that can be controlled in order to define expectations on that value (for example, to define that the drift score for every feature must be below 0.05).
Let’s see the result value for our check
result.value
{'numeric_without_drift': {'Drift score': 0.05499999999999999, 'Method': 'Kolmogorov-Smirnov', 'Importance': 0.6911764705882353}, 'numeric_with_drift': {'Drift score': 0.6990000000000001, 'Method': 'Kolmogorov-Smirnov', 'Importance': 0.3088235294117647}, 'categorical_without_drift': {'Drift score': 0.005136700975462043, 'Method': "Cramer's V", 'Importance': 0.0}, 'categorical_with_drift': {'Drift score': 0.22862322289807285, 'Method': "Cramer's V", 'Importance': 0.0}}
Define a condition#
As we can see, we get the drift score for each feature in the dataset, along with the feature importance in respect to the model.
Now, we define a condition that enforce each feature’s drift score must be below 0.1. A condition is deepchecks’ way to enforce that results are OK, and we don’t have a problem in our data or model!
check_cond = check.add_condition_drift_score_less_than(max_allowed_categorical_score=0.2,
max_allowed_numeric_score=0.1)
result = check_cond.run(train_dataset=train_dataset, test_dataset=test_dataset)
result.show(show_additional_outputs=False)
As we see, our condition successfully detects and filters the problematic features that contains a drift!
Get an aggregated value#
Using the reduce_output
function we can combine the drift values per feature and get a collective score
that reflects the effect of the drift on the model, taking into account all the features.
In scenarios where labels are unavailable (either temporarily of permanently)
this value can be a good indicator of possible deterioration in the model’s performance.
We can define the type of aggregation we want to use via the aggregation_method parameter. The possible values are:
l3_weighted: Default. L3 norm over the ‘per-feature scores’ vector weighted by the feature importance, specifically,
sum(FI * PER_FEATURE_SCORES^3)^(1/3). This method takes into account the feature importance yet puts more weight on
the per-feature scores. This method is recommended for most cases.
l5_weighted: Similar to ‘l3_weighted’, but with L5 norm. Puts even more emphasis on the per-feature scores and
specifically on the largest per-feature scores returning a score closer to the maximum among the per-feature scores.
weighted: Weighted mean of per-feature scores based on feature importance.
max: Maximum of all the per-feature scores.
none: No averaging. Return a dict with a per-feature score for each feature.
check = FeatureDrift(aggregation_method='weighted')
result = check.run(train_dataset=train_dataset, test_dataset=test_dataset, model=model)
result.reduce_output()
{'Weighted Drift Score': 0.2538823529411765}
Total running time of the script: (0 minutes 0.918 seconds)