pca: A Python Package for Principal Component Analysis

2.10.2 · active · verified Thu Apr 16

pca is a Python package for Principal Component Analysis (PCA), providing extended functionality beyond basic scikit-learn implementations. It leverages sklearn's core for compatibility while offering features like SparsePCA and TruncatedSVD, comprehensive analysis, and advanced plotting capabilities such as biplots, explained variance plots, outlier detection, and feature importance extraction. The current version is 2.10.2, and it is actively maintained.

Common errors

```
ImportError: cannot import name 'pca' from 'pca'
```
cause You are trying to import 'pca' (lowercase) as a class directly from the package, but Python's import system might resolve 'pca' as the package itself rather than its main class/function if not structured carefully, or you might be trying to import 'PCA' (uppercase) which is not the main class of this library.

fix The correct import for the main PCA class in this library is `from pca import pca` (lowercase `pca` for the class name) or ensure your environment doesn't have conflicting modules named 'pca'.
```
AttributeError: 'PCA' object has no attribute 'biplot'
```
cause You are likely using scikit-learn's `sklearn.decomposition.PCA` class, which does not have a `biplot` method. The `biplot` method is a feature provided by the `erdogant/pca` library.

fix Ensure you are importing the `pca` class from the `pca` library: `from pca import pca`. Then, create an instance of this class to access its extended methods like `biplot`.
```
ValueError: not enough features to compute (n_components) components
```
cause This error occurs when the `n_components` parameter specified for PCA is greater than or equal to the number of features (columns) in your input data. PCA cannot create more components than existing features.

fix Reduce the value of `n_components` to be less than the number of features in your dataset. For `sklearn.decomposition.PCA` specifically, it must be strictly less than `min(n_samples, n_features)` when using solvers like 'arpack'.

Warnings

gotcha PCA is highly sensitive to the scale of input data. Features with larger value ranges or variances can disproportionately influence the principal components if data is not scaled (e.g., using `sklearn.preprocessing.StandardScaler`) prior to PCA.
Fix: Always preprocess your data using a scaling method like `StandardScaler` from scikit-learn before applying PCA, especially if features have different units or ranges.
gotcha When using the plotting features of `pca`, specifically with `matplotlib` versions 3.10 and above, you might encounter warnings related to `get_cmap`.
Fix: Ensure you are using the latest `pca` library version which often includes fixes for newer dependency versions. If warnings persist, refer to the library's GitHub issues for potential workarounds or updates, or consider temporarily downgrading `matplotlib` if critical for production.
gotcha There's a separate `PCA` class within scikit-learn (`sklearn.decomposition.PCA`). Attempting to use methods unique to the `erdogant/pca` package (like `.plot()` or `.biplot()`) on a `sklearn.decomposition.PCA` object will result in an `AttributeError`.
Fix: Ensure you are importing and instantiating the correct PCA object: `from pca import pca` for this package, or `from sklearn.decomposition import PCA` for scikit-learn's native implementation.

Install

pip install pca Install stable version

Imports

pca
```
from pca import pca
```
PCA
wrong:
```
from pca import PCA
```
correct:
```
from sklearn.decomposition import PCA
```
The `pca` library provides its own `pca` class (lowercase 'p'). The uppercase 'PCA' is typically imported from scikit-learn's `sklearn.decomposition` module. Confusing these can lead to missing methods.

Quickstart

Initializes the `pca` model, fits it to sample data, and transforms the data into principal components. It also shows how to access the explained variance and the transformed data. Plotting functionality is also available.

import numpy as np
from pca import pca
import pandas as pd

# Sample data
X = pd.DataFrame(np.random.rand(100, 10), columns=[f'feature_{i}' for i in range(10)])

# Initialize PCA model with 3 components
model = pca(n_components=3)

# Fit and transform the data
out = model.fit_transform(X)

print("Explained variance ratio:", model.results['explained_var'])
print("Principal Components (transformed data shape):", out['PC'].shape)

# To display the biplot (requires matplotlib to be installed)
# import matplotlib.pyplot as plt
# model.plot()
# plt.show()

view raw JSON →