Phi-Alpha PCA
Project description
Alpha-PCA
Alpha-PCA is more robust to outliers than standard PCA.
Standard PCA is a special case of alpha PCA (when alpha=1).
Usage
The model is inherited from a sklearn module and works the same way as the standard PCA.
It also supports PyTorch tensors (on cpu and GPU).
from alpha_pca import AlphaPCA
import torch
X = torch.randn(16, 10) # also works with numpy
pca = AlphaPCA(n_components=5, alpha=0.7, random_state=123) # alpha=1 -> standard PCA
pca.fit(X)
# to project X in the latent space
X_transformed = pca.transform(X) # (16, 10) -> (16, 5)
# fit inverse
X_ = pca.inverse_transform(X_transformed) # (16, 5) -> (16, 10)
# directly approximate X_ == inverse_transform(transform(X))
X_ = pca.approximate(X) # (16, 10) -> (16, 10)
# Find the optimal alpha via a reconstruction loss
best_alpha = pca.compute_optimal_alpha(X, n_components=5)
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
alpha-pca-1.0.2.tar.gz
(6.1 kB
view hashes)
Built Distribution
Close
Hashes for alpha_pca-1.0.2-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | d735d6718d5f943b4596716c41fc3e65ab1efd79b191d5ffc5825fd572c5ee29 |
|
MD5 | 6708b12fe29ffed460ada8a222eef81b |
|
BLAKE2b-256 | 75720773a1130049fafa8e14af6c6259351df88a63c8a0ab9870c7b0e0dcaaf9 |