1

I have performed a PCA analysis over my original dataset and from the compressed dataset transformed by the PCA I have also selected the number of PC I want to keep. Now I am struggling with the identification of the original features that are important in the reduced dataset and getting error

IndexError: list index out of range; the code is as below;

# Importing the libraries
import numpy as np
import matplotlib.pyplot as plt
from sklearn import datasets, linear_model
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import KFold
from sklearn.model_selection import GridSearchCV
import pandas as pd

dt = pd.read_excel('GENES.xlsx')


X = dt.iloc[:, 0:14808].values
y = dt.iloc[:, 14807:14808].values

# Splitting the dataset into the Training set and Test set
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)
X_train, X_val, y_train, y_val = train_test_split(X_train, y_train, test_size=0.2, random_state=0)

#Feature Scaling
from sklearn.preprocessing import StandardScaler
sc_X = StandardScaler()
X_train = sc_X.fit_transform(X_train)
X_test = sc_X.transform(X_test)
X_val = sc_X.transform(X_val)

# Applying Selecting Features
from sklearn.decomposition import PCA
pca=PCA(n_components=13)
X_train = pca.fit_transform(X_train)
X_val = pca.transform(X_val)
X_test = pca.transform(X_test)
explained_variance = pca.explained_variance_ratio_

# number of components
n_pcs= pca.components_.shape[0]

# get the index of the most important feature on EACH component
# LIST COMPREHENSION HERE
most_important = [np.abs(pca.components_[i]).argmax() for i in range(n_pcs)]

initial_feature_names = [ ' UBE2Q2P2 ' , ' SSX9 ' , ' CXORF67 ' , ' EFCAB8 ' , ' SDR16C6P ' , 
                         ' EFCAB12 ' , ' A1BG ' , ' A1CF ' , ' RBFOX1 ' , ' GGACT ' , ' A2ML1 ' , ' A2M ' , 
                         ' A4GALT ' , ' A4GNT ' , ' AAAS ' , ' AACSP1 ' , ' AACS ' , ' AADACL2 ' , ' AADACL3 ' , ' AADACL4 ' , 
                         ' AADAC ' , ' AAGAB ' , ' AAK1 ' , ' AAMP ' , ' AANAT ' , ' AARS2 ' , ' AARSD1 ' , ' AARS ' , ' AASDHPPT ' ,
                         ' AASDH ' , ' AASS ' , ' AATF ' , ' AATK ' , ' ABAT ' , ' ABCA10 ' , ' ABCA11P ' , ' ABCA12 ' , ' ABCA13 ' ,
                         ' ABCA17P ' , ' ABCA1 ' , ' ABCA2 ' , ' ABCA3 ' , ' ABCA4 ' , ' ABCA5 ' , ' ABCA6 ' , ' ABCA7 ' , 
                         ' ABCA8 ' , ' ABCA9 ' , ' ABCB10 ' , ' ABCB11 ' , ' ABCB1 ' , ' ABCB4 ' , ' ABCB6 ' , ' ABCB7 ' , 
                         ' ABCB8 ' , ' ABCB9 ' , ' ABCC11 ' , ' ABCC12 ' , ' ABCC13 ' , ' ABCC1 ' , ' ABCC2 ' , ' ABCC3 ' , ' ABCC4 ' , ' ABCC5 ' , ' ABCC6P1 ' , ' ABCC6 ' , ' ABCC8 ' , ' ABCC9 ' , ' ABCD1 ' , ' ABCD2 ' , ' ABCD3 ' , ' ABCD4 ' , ' ABCE1 ' , ' ABCF1 ' , ' ABCF2 ' , ' ABCG1 ' , ' ABCG2 ' , ' ABCG4 ' , ' ABCG5 ' , ' ABCG8 ' , ' ABHD10 ' , ' ABHD11 ' , ' ABHD12B ' , ' ABHD12 ' , ' ABHD13 ' , ' ABHD14A ' , ' ABHD14B ' , ' ABHD15 ' , ' ABHD1 ' , ' ABHD2 ' , ' ABHD3 ' , ' ABHD4 ' , ' ABHD5 ' , ' ABHD6 ' , ' ABHD8 ' , ' ABI1 ' , ' ABI2 ' , ' ABI3BP ' , ' ABI3 ' , ' ABL1 ' , ' ABL2 ' , ' ABLIM1 ' , ' ABLIM2 ' , ' ABLIM3 ' , ' ABO ' , ' AOC1 ' , ' ABRA ' , ' ABR ' , ' ABT1 ' , ' ABTB1 ' , ' ABTB2 ' , ' ACAA1 ' , ' ACAA2 ' , ' ACACA ' , ' ACACB ' , ' ACAD10 ' , ' ACAD11 ' , ' ACAD8 ' , ' ACAD9 ' , ' ACADL ' , ' ACADM ' , ' ACADSB ' , ' ACADS ' , ' ACADVL ' , ' ACAN ' , ' ACAP1 ' , ' ACAP2 ' , ' ACAP3 ' , ' ACAT1 ' , ' ACAT2 ' , ' ACBD3 ' , ' ACBD4 ' , ' ACBD5 ' , ' ACBD6 ' , ' ACBD7 ' , ' ASIC2 ' , ' ASIC1 ' , ' ASIC3 ' , ' ASIC4 ' , ' ASIC5 ' , ' ACCSL ' , ' ACCS ' , ' ACD ' , ' ACE2 ' , ' ACER1 ' , ' ACER2 ' , ' ACER3 ' , ' ACE ' , ' ACHE ' , ' ACIN1 ' , ' ACLY ' , ' ACMSD ' , ' ACO1 ' , ' ACO2 ' , ' ACOT11 ' , ' ACOT12 ' , ' ACOT13 ' , ' ACOT1 ' , ' ACOT2 ' , ' ACOT4 ' , ' ACOT6 ' , ' ACOT7 ' , ' ACOT8 ' , ' ACOT9 ' , ' ACOX1 ' , ' ACOX2 ' , ' ACOX3 ' , ' ACOXL ' , ' ACP1 ' , ' ACP2 ' , ' ACP5 ' , ' ACP6 ' , ' PXYLP1 ' , ' ACPP ' , ' ACPT ' , ' ACRBP ' , ' ACRC ' , ' ACRV1 ' , ' ACR ' , ' ACSBG1 ' , ' ACSBG2 ' , ' ACSF2 ' , ' ACSF3 ' , ' ACSL1 ' , ' ACSL3 ' , ' ACSL4 ' , ' ACSL5 ' , ' ACSL6 ' , ' ACSM1 ' , ' ACSM2A ' , ' ACSM2B ' , ' ACSM3 ' , ' ACSM4 ' , ' ACSM5 ' , ' ACSS1 ' , ' ACSS2 ' , ' ACSS3 ' , ' ACTA1 ' , ' ACTA2 ' , ' ACTBL2 ' , ' ACTB ' , ' ACTC1 ' , ' ACTG1 ' , ' ACTG2 ' , ' ACTL6A ' ]

# 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): most_important_names[14805] for i in range(n_pcs)}

# build the dataframe
df = pd.DataFrame(dic.items())
fuwiak
  • 1,373
  • 8
  • 13
  • 26
DataP
  • 31
  • 2

2 Answers2

1

PCA does not throw away the unimportant features.

In other words, features that you get from PCA are not the original ones. Mathematically PCA is orthogonal linear transformation that transforms the data to a new coordinate system such that the greatest variance by some scalar projection of the data comes to lie on the first coordinate (called the first principal component), the second greatest variance on the second coordinate, and so on...

Noah Weber
  • 5,669
  • 1
  • 12
  • 26
0

It's been three years, hopefully you you've found solution since then. This is for those who are having the same difficulty, as i was also too few days ago.

When running PCA using sklearn in python, in pca.components_, you will find the correlation between principal component found and the original variable. See the code below.

    import pandas as pd
    import numpy as np
    from sklearn.preprocessing import StandardScaler
    from sklearn.decomposition import PCA

# Here you read and initialize your dataset, it was a csv.
df = pd.read_csv('./mystery.csv')

# We scale our dataset to prepare it for pca
scaler = StandardScaler()
scaler.fit(np.asarray(df))
df_scaled = scaler.transform(df)

# Your scaled dataset as dataframe if interested to see how it looks
df=pd.DataFrame(df_scaled, columns=['V1','V2','V3'])

#Now we apply our PCA to have two principal components (it will takes the two first one, leave PCA() without argument if you want to see all of them
pca = PCA(n_components=2)
df_projected = pca.fit_transform(df_scaled)
pd.DataFrame(df_projected, columns=["F"+str(i+1) for i in range(2)])2    # our dataset with Principal Components as variables (columns)

#And here finally, you could see the correlation, which is the weight of any of your original variables(columns/caracteristic) on any principal components.
corr = pd.DataFrame(pca.components_, index=["F"+str(i+1) for i in range(2)], columns=["V"+str(j+1) for j in range(3)])
corr
Saraï SADJO LEGBANE
  • 1