Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

BOSSVS not working with a single feature #146

Open
roman-4erkasov opened this issue Feb 24, 2023 · 1 comment
Open

BOSSVS not working with a single feature #146

roman-4erkasov opened this issue Feb 24, 2023 · 1 comment

Comments

@roman-4erkasov
Copy link

roman-4erkasov commented Feb 24, 2023

Description

Class pyts.classification.BOSSVS doesn't accept timeseries of one feature.
Advices from the error message doesn't help, but leads to another error.

Steps/Code to Reproduce

I tried all the three possible versions to use timeseries of one feature:

  1. The first version is just 1D-array:
import numpy as np
from pyts.classification import BOSSVS
x_train = np.random.uniform(low=0.0, high=10.0, size=(300,))
y_train = np.random.randint(low=0, high=2, size=(300,))
est = BOSSVS().fit(x_train, y_train)

This example gives:

ValueError: Expected 2D array, got 1D array instead:
array=[1.84916215 7.16606073 5.69089018 ... ].
Reshape your data either using array.reshape(-1, 1) if your data has a single feature or array.reshape(1, -1) if it contains a single sample.
  1. After that I tried the following code
import numpy as np
from pyts.classification import BOSSVS
x_train = np.random.uniform(low=0.0, high=10.0, size=(300,))
y_train = np.random.randint(low=0, high=2, size=(300,))
est = BOSSVS().fit(x_train.reshape(-1, 1), y_train)

It gives the following error:
ValueError: If 'window_size' is an integer, it must be greater than or equal to 1 and lower than or equal to n_timestamps if 'drop_sum=False'.

  1. Finally I tried the following code:
import numpy as np
from pyts.classification import BOSSVS
x_train = np.random.uniform(low=0.0, high=10.0, size=(300,))
y_train = np.random.randint(low=0, high=2, size=(300,))
est = BOSSVS().fit(x_train.reshape(1, -1), y_train)

It gives the following error:
ValueError: Found input variables with inconsistent numbers of samples: [1, 300]

Versions

NumPy 1.23.4
SciPy 1.9.3
Scikit-Learn 1.2.0
Numba 0.56.4
Pyts 0.12.0

Thank You!

@johannfaouzi
Copy link
Owner

Sorry for the delayed response.

In order to train a classification algorithm, one needs several samples. In our case, a sample is a time series. This set of training samples is often called the training set.

The expected format for the training set is similar to the one used in scikit-learn (if you are familiar with it):

  • X_train is a 2D-array with shape (n_samples, n_timestamps): the first dimension corresponds to the samples (time series), while the second dimension corresponds to the time.
  • y_train is a 1D-array with shape (n_samples,): it contains the label associated with each sample (time series).

The format is identical for the test set.

Let's load a toy dataset to illustrate this:

>>> from pyts.datasets import load_gunpoint
>>> X_train, X_test, y_train, y_test = load_gunpoint(return_X_y=True)
>>> X_train.shape
(50, 150)  # there are 50 time series, each with 150 values.
>>> y_train.shape
(50,)  # there are 50 labels because there are 50 time series in the training set.
>>> y_train
array([2, 2, 1, 1, 2, ...])  # there are 2 labels (denoted as the integers 1 and 2).
>>> X_test.shape
(150, 150)  # there are 150 time series, each with 150 values.
>>> y_test.shape
(150,)  # there are 150 labels because there are 150 time series in the test set.

Now, one can perform classification using BOSSVS on this dataset:

>>> from pyts.classification import BOSSVS
>>> clf = BOSSVS()
>>> clf.fit(X_train, y_train)
BOSSVS()
>>> clf.score(X_test, y_test)
0.82  # accuracy score of 0.82 on the test set

Back to your example, I don't understand your data. It seems that you have 300 time series, but each time series has a single value. You cannot use BOSSVS with such data. You cannot do any time series analysis if the time series have a single value. It probably does not make sense to consider this kind of data as time series.

Hope this helps you a bit and I would be happy to give you more info if needed, but I'm not sure to understand your data.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants