Skip to content

TemporAI

Data Tutorial 04: Data splitting

Data Tutorial 04: Data splitting¶

This tutorial shows how a Dataset can be split or KFold-ed.

1. Splitting data¶

[ ]:

# Get a dataset.

from tempor import plugin_loader

sine_dataset = plugin_loader.get("prediction.one_off.sine", plugin_type="datasource").load()

sine_dataset

OneOffPredictionDataset(
    time_series=TimeSeriesSamples([100, *, 5]),
    static=StaticSamples([100, 4]),
    predictive=OneOffPredictionTaskData(targets=StaticSamples([100, 1]))
)

The method train_test_split (same API as sklearn.model_selection.train_test_split) can be used to split the dataset into train and test sets.

[ ]:

sine_dataset_train, sine_dataset_test = sine_dataset.train_test_split(test_size=0.2)

print("Training set:")
print(sine_dataset_train)
print("Test set:")
print(sine_dataset_test)

Training set:
OneOffPredictionDataset(
    time_series=TimeSeriesSamples([80, *, 5]),
    static=StaticSamples([80, 4]),
    predictive=OneOffPredictionTaskData(targets=StaticSamples([80, 1]))
)
Test set:
OneOffPredictionDataset(
    time_series=TimeSeriesSamples([20, *, 5]),
    static=StaticSamples([20, 4]),
    predictive=OneOffPredictionTaskData(targets=StaticSamples([20, 1]))
)

2. Using `KFold` and other `sklearn` splitter classes¶

Any of the sklearn splitter classes can be used with Dataset split method.

The following example illustrates this with the `KFold splitter <https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.KFold.html#sklearn.model_selection.KFold>`__:

[ ]:

from sklearn.model_selection import KFold

kfold = KFold(n_splits=5, random_state=123, shuffle=True)

splits = list(sine_dataset.split(splitter=kfold))

print(len(splits))

[ ]:

train_set_0, test_set_0 = splits[0]

print("0th split training set:")
print(sine_dataset_train)
print("0th split test set:")
print(sine_dataset_test)

0th split training set:
OneOffPredictionDataset(
    time_series=TimeSeriesSamples([80, *, 5]),
    static=StaticSamples([80, 4]),
    predictive=OneOffPredictionTaskData(targets=StaticSamples([80, 1]))
)
0th split test set:
OneOffPredictionDataset(
    time_series=TimeSeriesSamples([20, *, 5]),
    static=StaticSamples([20, 4]),
    predictive=OneOffPredictionTaskData(targets=StaticSamples([20, 1]))
)