Bug Description
When using the principal method with lapack SVD instead of randomized, the loadings matrix returned by FactorAnalyzer is always given in full, it has shape n_cols x n_cols, instead of selecting only loadings for the n_factors desired. When using the randomized SVD, there is no issue.
Reproducible Code
import pandas as pd
import numpy as np
num_rows = 1000
num_cols = 6
df = pd.DataFrame(
np.random.standard_normal(size=(num_rows, num_cols)),
columns=[f'col{i+1}' for i in range(num_cols)])
# shape is correct with randomized
efa = FactorAnalyzer(n_factors=2, rotation='promax', method='principal', svd_method='randomized')
efa.fit(df)
print(efa.loadings_.shape)
# shape is incorrect with lapack
efa = FactorAnalyzer(n_factors=2, rotation='promax', method='principal', svd_method='lapack')
efa.fit(df)
print(efa.loadings_.shape)
Expected behavior
The shape of the .loadings_ attribute should be n_cols x n_factors.
Versions (please complete the following information):
Bug Description When using the principal method with lapack SVD instead of randomized, the loadings matrix returned by FactorAnalyzer is always given in full, it has shape n_cols x n_cols, instead of selecting only loadings for the n_factors desired. When using the randomized SVD, there is no issue.
Reproducible Code
Expected behavior The shape of the .loadings_ attribute should be n_cols x n_factors.
Versions (please complete the following information):
factor_analyzer
: 0.5.1 /numpy
: 1.26.1 /scipy
: 1.11.3 /pandas
: 2.1.1