Any help for minirocket on UEA multivariate time series classification

[cc19860606]

hello， has any result report on minirocket on UEA multivariate time series classification archive？ @angus924
I use the minirocket_multivariate to handle PenDigits dataset in UEA multivariate，but there is NaN in X_training_transform.
And the result on UEA is poor compared to the result of minirocket_dv on UCRArchive_2018, can give me some suggestion?
Code:
parameters = fit(X_training,num_features = 10_000)
X_training_transform = transform(X_training, parameters)
print('X_training_transform:',X_training_transform)
print('type(X_training_transform):',type(X_training_transform))
print("X_training_transform.shape:", X_training_transform.shape)
print("np.isnan(X_training_transform).any():", np.isnan(X_training_transform).any())
classifier = RidgeClassifierCV(alphas = np.logspace(-3, 3, 10), normalize = True)
classifier.fit(X_training_transform, Y_training)
X_test_transform = transform(X_test, parameters)
predictions = classifier.predict(X_test_transform)
Report:
last_X_training.shape: (7494, 2, 8)
last_X_test.shape: (3498, 2, 8)
last_Y_training.shape: (7494,)
last_Y_test.shape: (3498,)
X_training_transform: [[0. 0. 0. ... 0.625 0.875 0.375]
[0. 0. 0. ... 0.625 1. 0.125]
[0. 0. 0. ... 0.375 0.625 0.25 ]
...
[0. 0. 0. ... 0.375 0.875 0.125]
[0. 0. 0. ... 0.25 1. 0.125]
[0. 0. 0. ... 0.5 0.875 0.125]]
type(X_training_transform): <class 'numpy.ndarray'>
X_training_transform.shape: (7494, 9996)
np.isnan(X_training_transform).any(): True
Traceback (most recent call last):
File "cc-test.py", line 68, in
classifier.fit(X_training_transform, Y_training)
File "/newhome/chenc/miniforge3/envs/AIcocahing/lib/python3.6/site-packages/sklearn/linear_model/_ridge.py", line 1943, in fit
multi_output=True, y_numeric=False)
File "/newhome/chenc/miniforge3/envs/AIcocahing/lib/python3.6/site-packages/sklearn/base.py", line 433, in _validate_data
X, y = check_X_y(X, y, **check_params)
File "/newhome/chenc/miniforge3/envs/AIcocahing/lib/python3.6/site-packages/sklearn/utils/validation.py", line 63, in inner_f
return f(*args, **kwargs)
File "/newhome/chenc/miniforge3/envs/AIcocahing/lib/python3.6/site-packages/sklearn/utils/validation.py", line 878, in check_X_y
estimator=estimator)
File "/newhome/chenc/miniforge3/envs/AIcocahing/lib/python3.6/site-packages/sklearn/utils/validation.py", line 63, in inner_f
return f(*args, **kwargs)
File "/newhome/chenc/miniforge3/envs/AIcocahing/lib/python3.6/site-packages/sklearn/utils/validation.py", line 721, in check_array
allow_nan=force_all_finite == 'allow-nan')
File "/newhome/chenc/miniforge3/envs/AIcocahing/lib/python3.6/site-packages/sklearn/utils/validation.py", line 106, in _assert_all_finite
msg_dtype if msg_dtype is not None else X.dtype)
ValueError: Input contains NaN, infinity or a value too large for dtype('float32').

[angus924]

Hi @cc19860606,

Thanks for your question, sorry for the slow reply.

has any result report on minirocket on UEA multivariate time series classification archive?

Good question, it doesn't look like it. I can't find any.

It looks like there are some results here, from this paper. (I haven't checked any of this, so I can't say if it's correct or not, but maybe it's useful.)

there is NaN in X_training_transform

I think what might be happening is this: the input time series are of length 8, but MiniRocket needs time series of at least length 9. So I think you will have to pad the time series by putting zeros at the start and/or end of the series. I think at the moment some kind of strange undefined behaviour is occurring.

Let me know if padding with zeros fixes the problem. If not, I can try to figure out what else might not be working.

Thanks very much.

Best,

Angus