I am trying to work out how to use PCA to determine the most important features. I think I have done that below.
I am wondering then, how do I pass the most important features, with their original column names (from a pandas dataframe) back into the new dataframe I am creating at the bottom - so I can use that as the new 'lightweight' dataset?
This way, if I set n_components to 10; I would have 10 feature columns (with names) being passed into the new dataframe.
Any ideas?
from sklearn.decomposition import PCA
# PCA (principal component analysis) aims to reduce the number of dimensions in the dataset, without losing those which are very relevant to the model
# it provides a score, you can drop those with poor scores.
X_pc = PCA(n_components=2).fit_transform(train_features)
pd.DataFrame({'PC1': X_pc[:, 0], 'PC2': X_pc[:, 1], 'Y': train_labels.ravel()}).sample(10)
PCA reduced the dimensions to 2 by linearly combining the initial features. After transformation, the output is a matrix with [samples, components] size and thus, it is not possible to create a dataframe since you cannot project back the names/features.
The important features are the ones that influence more the components and thus, have a large absolute value on the component.
If you change the code you can get the most important features on the PCs:
from sklearn.decomposition import PCA
import pandas as pd
import numpy as np
np.random.seed(0)
# 10 samples with 5 features
train_features = np.random.rand(10,5)
model = PCA(n_components=2).fit(train_features)
X_pc = model.transform(train_features)
# number of components
n_pcs= model.components_.shape[0]
# get the index of the most important feature on EACH component
# LIST COMPREHENSION HERE
most_important = [np.abs(model.components_[i]).argmax() for i in range(n_pcs)]
initial_feature_names = ['a','b','c','d','e']
# get the names
most_important_names = [initial_feature_names[most_important[i]] for i in range(n_pcs)]
# LIST COMPREHENSION HERE AGAIN
dic = {'PC{}'.format(i+1): most_important_names[i] for i in range(n_pcs)}
# build the dataframe
df = pd.DataFrame(sorted(dic.items()))
This prints:
0 1
0 PC1 e
1 PC2 d
So on the PC1 the feature named e is the most important and on PC2 the d.