from sklearn.preprocessing import StandardScaler

from sklearn.decomposition import PCA

​

scaler = StandardScaler()

std_df = scaler.fit_transform(df)

pca = PCA()

print(pca.fit_transform(std_df))

# See which principal components explains the variance more

print(pca.explained_variance_ratio_)

# See equation of principal components : PC 1 = 0.71 x Feature 1 + 0.71 x Feature 2

print(pca.components_)

​

### Alternative approach: Pipeline

from sklearn.pipeline import Pipeline

pipe = Pipeline([('scaler', StandardScaler()),

                ('reducer', PCA(n_components=3)), # n_components=0.9 for capturing 90% variance

                ('classifier', RandomForestClassifier())])

# principal components

pipe['reducer'].components_

# No of principal components

pipe['reducer'].n_components_

# See PCA explained variance

pipe['reducer'].explained_variance_ratio_

# Visualize elbow plot for PCA tuning

plt.plot(pipe['reducer'].explained_variance_ratio_)

# Visualize PCA plot

pc = pipe['reducer'].fit_transform(df)

df['PC 1'] = pc[:,0]

df['PC 2'] = pc[:,1]

sns.scatterplot(data=df, x='PC 1', y='PC 2', hue='cat_col', alpha=0.4)

# Rebuilding back to original Data

pc = pipe['reducer'].transform(X)

X_rebuilt = pipe['reducer'].inverse_transform(pc)