Binder badge Colab badge

Date Train Validation Leakage Overlap

[1]:
from deepchecks.checks.methodology import DateTrainTestLeakageOverlap
from deepchecks.base import Dataset, Suite
from datetime import datetime
import pandas as pd

def dataset_from_dict(d: dict, datetime_name: str = None) -> Dataset:
    dataframe = pd.DataFrame(data=d)
    return Dataset(dataframe, datetime_name=datetime_name)

Synthetic example dates before last training

[2]:
train_ds = dataset_from_dict({'col1': [
        datetime(2021, 10, 1, 0, 0),
        datetime(2021, 10, 1, 0, 0),
        datetime(2021, 10, 1, 0, 0),
        datetime(2021, 10, 2, 0, 0),
        datetime(2021, 10, 2, 0, 0),
        datetime(2021, 10, 2, 0, 0),
        datetime(2021, 10, 3, 0, 0),
        datetime(2021, 10, 3, 0, 0),
        datetime(2021, 10, 3, 0, 0),
        datetime(2021, 10, 4, 0, 0),
        datetime(2021, 10, 4, 0, 0),
        datetime(2021, 10, 4, 0, 0),
        datetime(2021, 10, 5, 0, 0),
        datetime(2021, 10, 5, 0, 0)
    ]}, 'col1')
test_ds = dataset_from_dict({'col1': [
        datetime(2021, 9, 4, 0, 0),
        datetime(2021, 10, 6, 0, 0),
        datetime(2021, 10, 6, 0, 0),
        datetime(2021, 10, 7, 0, 0),
        datetime(2021, 10, 7, 0, 0),
        datetime(2021, 10, 8, 0, 0),
        datetime(2021, 10, 8, 0, 0),
        datetime(2021, 10, 9, 0, 0),
        datetime(2021, 10, 9, 0, 0)
    ]}, 'col1')

DateTrainTestLeakageOverlap().run(train_dataset=train_ds, test_dataset=test_ds)

Date Train-Test Leakage (overlap)

Check test data that is dated earlier than latest date in train. Read More...

Additional Outputs
11.11% of test data dates before last training data date (2021/10/05 00:00:00.000000 )

Synthetic example no date leakage

[3]:
train_ds = dataset_from_dict({'col1': [
        datetime(2021, 10, 3, 0, 0),
        datetime(2021, 10, 3, 0, 0),
        datetime(2021, 10, 4, 0, 0),
        datetime(2021, 10, 4, 0, 0),
        datetime(2021, 10, 4, 0, 0),
        datetime(2021, 10, 5, 0, 0),
        datetime(2021, 10, 5, 0, 0)
    ]}, 'col1')
test_ds = dataset_from_dict({'col1': [
        datetime(2021, 11, 4, 0, 0),
        datetime(2021, 11, 4, 0, 0),
        datetime(2021, 11, 5, 0, 0),
        datetime(2021, 11, 6, 0, 0),

    ]}, 'col1')

DateTrainTestLeakageOverlap().run(train_dataset=train_ds, test_dataset=test_ds)

Date Train-Test Leakage (overlap)

Check test data that is dated earlier than latest date in train. Read More...

Additional Outputs

Nothing found