pyts

0.13.0 · active · verified Thu Apr 16

pyts is a Python package dedicated to time series classification. It provides preprocessing and utility tools, along with implementations of various time series classification algorithms. The library maintains an active development status, with major versions often introducing new Python support and algorithms, currently at version 0.13.0.

Common errors

```
ModuleNotFoundError: No module named 'pyts'
```
cause The `pyts` package has not been installed in the current Python environment.

fix Run `pip install pyts` to install the library.
```
ImportError: cannot import name 'BagOfWords' from 'pyts.bag_of_words'
```
cause Attempting to import `BagOfWords` expecting its pre-v0.11.0 behavior, or from an incompatible version context.

fix For the functionality of `BagOfWords` before v0.11.0, import `WordExtractor` instead: `from pyts.bag_of_words import WordExtractor`. If you intend to use the newer `BagOfWords` functionality, ensure your `pyts` version is 0.11.0 or newer.
```
AttributeError: type object 'MyCustomEstimator' has no attribute '_estimator_type'
```
cause This can occur when custom `pyts`-compatible estimators or pipelines rely on `scikit-learn`'s internal mixin classes which `pyts` replaced with its own in version 0.12.0. The `_estimator_type` attribute might be expected by `scikit-learn` utilities interacting with `pyts` components, but the internal implementation has changed.

fix If this error occurs after updating `pyts` to v0.12.0 or newer, verify that any custom classes or integrations are compatible with `pyts`'s new internal mixin structure. It may require updating the custom estimator's inheritance or attributes to align with how `pyts` now defines estimator types internally, rather than directly relying on `sklearn`'s private APIs. Ensure that `pyts` estimators are used as intended within `scikit-learn` pipelines, which are generally supported.

Warnings

breaking Python 3.7 support was dropped in v0.13.0. Python 3.6 support was dropped in v0.12.0, and Python 3.5 in v0.11.0. Ensure your Python environment meets the current requirements.
Fix: Upgrade your Python interpreter to 3.8 or newer for pyts v0.11.0+, 3.9 or newer for pyts v0.12.0+, and 3.10 or newer for pyts v0.13.0+.
breaking The behavior of the `BagOfWords` algorithm was reworked in v0.11.0. If you relied on the old functionality, it has been moved to a new class, `WordExtractor`.
Fix: For the old `BagOfWords` functionality, use `from pyts.bag_of_words import WordExtractor` instead. If you want the new `BagOfWords` behavior, update your code to reflect its new parameters or expected output.
breaking `pyts` replaced `scikit-learn` mixin classes with its own internal mixin classes in v0.12.0. Custom estimators inheriting from deprecated `sklearn` mixins might encounter issues.
Fix: Review custom `pyts`-compatible estimators that directly inherited from `scikit-learn` mixins. Update them to use `pyts`'s own mixin classes or ensure compatibility with `pyts`'s internal mixin implementations.
gotcha Minimal versions of core dependencies (NumPy, SciPy, Scikit-Learn, Joblib, Numba) have been updated significantly in recent releases. Using an older `pyts` with newer dependencies, or a newer `pyts` with older dependencies, can lead to runtime errors or unexpected behavior.
Fix: Always install `pyts` in a clean virtual environment and allow `pip` to resolve the latest compatible versions of its dependencies. Check the official `pyts` documentation for the exact dependency requirements for your `pyts` version. `pip install pyts` should handle this for the latest version.

Install

pip install pyts Install stable version

Imports

TimeSeriesForest

from pyts.classification import TimeSeriesForest

BagOfWords
wrong:
```
from pyts.bag_of_words import BagOfWords # for pre-v0.11.0 behavior
```
correct:
```
from pyts.bag_of_words import BagOfWords
```
The BagOfWords algorithm was reworked in v0.11.0; its former version is now available as WordExtractor.
WordExtractor
```
from pyts.bag_of_words import WordExtractor
```
Introduced in v0.11.0, containing the functionality of the former BagOfWords.
load_gunpoint
```
from pyts.datasets import load_gunpoint
```

StandardScaler

from pyts.preprocessing import StandardScaler

BOSS
```
from pyts.transformation import BOSS
```

Quickstart

This quickstart loads the 'GunPoint' dataset, initializes a `TimeSeriesForest` classifier, trains it on the training data, and evaluates its accuracy on the test set. This demonstrates a common workflow for time series classification using `pyts`.

import numpy as np
from pyts.datasets import load_gunpoint
from pyts.classification import TimeSeriesForest

# Load the GunPoint dataset
X_train, X_test, y_train, y_test = load_gunpoint(return_X_y=True)

# Initialize and train the Time Series Forest classifier
clf = TimeSeriesForest(random_state=43)
clf.fit(X_train, y_train)

# Evaluate the classifier
accuracy = clf.score(X_test, y_test)
print(f"Accuracy of TimeSeriesForest: {accuracy:.4f}")

view raw JSON →