scikit-fda

0.10.1 verified Fri May 01 auth: no python

scikit-fda is a Python package for functional data analysis (FDA). Version 0.10.1 requires Python >=3.10. It provides tools for representation, preprocessing, and statistical analysis of functional data, following scikit-learn like API. Releases are irregular, roughly 1-2 per year.

pip install scikit-fda

Common errors

error ModuleNotFoundError: No module named 'skfda.representation' ↓

cause Incorrect import path; in older docs, some examples use deep nested paths that no longer exist.

fix

Use the new top-level imports: from skfda import FDataGrid instead of from skfda.representation.grid import FDataGrid.

error ValueError: Data must be 2D array with shape (n_samples, n_points) or (n_samples, n_points, n_dimensions) ↓

cause Data passed to FDataGrid has wrong shape (e.g., transposed).

fix

Reshape data to (n_samples, n_points). For multivariate functional data, shape should be (n_samples, n_points, n_dimensions).

error AttributeError: 'FData' object has no attribute 'regularize' ↓

cause The `regularize` method was removed in newer versions (>=0.9).

fix

Use least_squares_warping from skfda.preprocessing.registration instead.

Warnings

breaking In version 0.9, the top-level imports were reorganized. Importing from `skfda.representation.grid` or `skfda.representation.basis` directly is deprecated. Use `from skfda import FDataGrid` and `from skfda.representation.basis import FDataBasis`. ↓

fix Update imports to the new paths as shown in the docs.

deprecated The `regularize` function from `skfda.preprocessing.registration` was deprecated in 0.9. Use `least_squares_warping` instead. ↓

fix Replace `regularize` calls with `least_squares_warping` from the same module.

gotcha FDataGrid expects data_matrix of shape (n_samples, n_points) by default, not (n_points, n_samples). Common mistake: transposed data leads to weird errors. ↓

fix Ensure data_matrix has shape (n_samples, n_points) or specify `sample_points` argument accordingly.

gotcha When using basis expansion (e.g., `FDataBasis`), the coefficients array shape must match the basis. Dimensions mismatch leads to obscure NumPy errors. ↓

fix Check that the shape of coefficients is (n_samples, n_basis) for unidimensional basis.

Install

pip install scikit-fda[plot]

Imports

FDataGrid
wrong
```
from skfda.representation import FDataGrid
```
correct
```
from skfda import FDataGrid
```
In older versions, FDataGrid was under skfda.representation.grid, but since 0.9 it is directly importable from skfda.

FPCA

from skfda.preprocessing.dim_reduction import FPCA

No common wrong import; use full path.

regularize

wrong

from skfda.registration import regularize

correct

from skfda.preprocessing.registration import least_squares_warping

The registration module was reorganized; now use least_squares_warping directly.

Quickstart

Creates a simple FDataGrid object and performs functional principal component analysis (FPCA).

import numpy as np
from skfda import FDataGrid
from skfda.preprocessing.dim_reduction import FPCA

# Generate functional data: 10 curves, each with 50 points
t = np.linspace(0, 1, 50)
data_matrix = np.random.randn(10, 50)  # 10 samples, 50 time points
fd = FDataGrid(data_matrix, grid_points=t)

# Perform FPCA
fpca = FPCA(n_components=3)
fpca.fit(fd)
scores = fpca.transform(fd)
print(scores.shape)