MantisClassifier

class MantisClassifier(checkpoint='paris-noah/MantisV2', model_version='v2', fine_tuning_type='full', seq_len=512, num_epochs=100, batch_size=64, predict_batch_size=256, base_learning_rate=0.0002, learning_rate_adjusting=True, device='auto', ignore_deps=False)[source]

Time series classifier using the Mantis foundation model.

Mantis [1] is a family of lightweight, calibrated foundation models designed for time series classification. This adapter wraps Mantis’s MantisTrainer scikit-learn-like interface as an sktime BaseClassifier.

Two usage modes are supported:

Fine-tuning (default): the Mantis backbone and/or classification head are fine-tuned on the supplied training data.
Zero-shot (fine_tuning_type="head" with very few epochs): only the linear classification head is trained on top of frozen Mantis embeddings.

Parameters:

checkpointstr or None, default=”paris-noah/MantisV2”

Hugging Face checkpoint to load via network.from_pretrained. Supported checkpoints:

"paris-noah/MantisV2" (MantisV2 backbone — recommended)
"paris-noah/MantisPlus" (MantisV1+ backbone)
"paris-noah/Mantis-8M" (MantisV1 backbone, smallest)

Pass None to use a randomly initialized backbone (no pre-training).

model_version{“v2”, “v1”}, default=”v2”

Mantis architecture family.

"v2" — use with "paris-noah/MantisV2".
"v1" — use with "paris-noah/Mantis-8M" or "paris-noah/MantisPlus".

fine_tuning_type{“full”, “head”, “adapter_head”, “scratch”}, default=”full”

Which parameters to update during training:

"full" — fine-tune the entire network (best accuracy).
"head" — train only the classification head on frozen embeddings (fastest, suitable as a zero-shot baseline).
"adapter_head" — train a learnable adapter and the head.
"scratch" — train everything from random initialisation.

seq_lenint, default=512

Sequence length passed to Mantis (must be a multiple of 32). Input time series that differ in length are resized to this value via linear interpolation.

num_epochsint, default=100

Number of fine-tuning epochs.

batch_sizeint, default=64

Batch size used during training.

predict_batch_sizeint, default=256

Batch size used during inference.

base_learning_ratefloat, default=2e-4

Initial learning rate for the AdamW optimizer.

learning_rate_adjustingbool, default=True

Whether to use the cosine learning-rate schedule built into Mantis.

devicestr, default=”auto”

Torch device. "auto" resolves to "cuda" when a GPU is available, otherwise "cpu".

ignore_depsbool, default=False

Skip soft-dependency checks (useful for testing without mantis-tsfm installed).

Attributes:

is_fitted: Whether fit has been called.

References

[1]

Feofanov et al., “Mantis: Lightweight Calibrated Foundation Model for User-Friendly Time Series Classification”, 2025. https://arxiv.org/abs/2502.15637

Examples

>>> from sktime.classification.foundation_models.mantis import MantisClassifier
>>> from sktime.datasets import load_unit_test
>>> X_train, y_train = load_unit_test(split="train", return_type="numpy3d")
>>> X_test, y_test = load_unit_test(split="test", return_type="numpy3d")
>>> clf = MantisClassifier(num_epochs=5, batch_size=16)
>>> clf.fit(X_train, y_train)
MantisClassifier(...)
>>> y_pred = clf.predict(X_test)

Methods

`check_is_fitted`([method_name])	Check if the estimator has been fitted.
`clone`()	Obtain a clone of the object with same hyper-parameters and config.
`clone_tags`(estimator[, tag_names])	Clone tags from another object as dynamic override.
`create_test_instance`([parameter_set])	Construct an instance of the class, using first test parameter set.
`create_test_instances_and_names`([parameter_set])	Create list of all test instances and a list of names for them.
`fit`(X, y)	Fit time series classifier to training data.
`fit_predict`(X, y[, cv, change_state])	Fit and predict labels for sequences in X.
`fit_predict_proba`(X, y[, cv, change_state])	Fit and predict labels probabilities for sequences in X.
`get_class_tag`(tag_name[, tag_value_default])	Get class tag value from class, with tag level inheritance from parents.
`get_class_tags`()	Get class tags from class, with tag level inheritance from parent classes.
`get_config`()	Get config flags for self.
`get_fitted_params`([deep])	Get fitted parameters.
`get_param_defaults`()	Get object's parameter defaults.
`get_param_names`([sort])	Get object's parameter names.
`get_params`([deep])	Get a dict of parameters values for this object.
`get_tag`(tag_name[, tag_value_default, ...])	Get tag value from instance, with tag level inheritance and overrides.
`get_tags`()	Get tags from instance, with tag level inheritance and overrides.
`get_test_params`([parameter_set])	Return testing parameter settings for the estimator.
`is_composite`()	Check if the object is composed of other BaseObjects.
`load_from_path`(serial)	Load object from file location.
`load_from_serial`(serial)	Load object from serialized memory container.
`predict`(X)	Predicts labels for sequences in X.
`predict_proba`(X)	Predicts labels probabilities for sequences in X.
`reset`()	Reset the object to a clean post-init state.
`save`([path, serialization_format])	Save serialized self to bytes-like object or to (.zip) file.
`score`(X, y)	Scores predicted labels against ground truth labels on X.
`set_config`(**config_dict)	Set config flags to given values.
`set_params`(**params)	Set the parameters of this object.
`set_random_state`([random_state, deep, ...])	Set random_state pseudo-random seed parameters for self.
`set_tags`(**tag_dict)	Set instance level tag overrides to given values.