Reduce Dimensions — reduceDimensions

Compress selected numeric features into principal components for exploration.

Use cases

• Inspect a lower-dimensional representation of correlated numeric features.
• Retain IDs and targets while replacing explicitly selected feature columns.

What it does

Fits PCA on the selected columns, removes those original columns and appends components named <prefix>_1, <prefix>_2, and so on. Other columns and row order are retained. If columns is absent or empty, every numeric column is selected, including numeric IDs or targets.

An integer component count is capped to the smaller of the number of selected columns and rows. A fraction between zero and one requests an explained-variance fraction. The node emits only an out dataframe and refits PCA on each input; it does not expose a reusable fitted PCA model.

Configuration

Generated Python code

Download the customer sample CSV and save it as customers.csv beside the script. These are complete exports from File Input → this node → File Output; additional preparation steps are stated below.

The flow standardizes tenure, monthly_charges and support_calls with Scale Features, then fits two PCA components with seed: 42. It writes reduce-dimensions.csv with customer_id, churn, pc_1 and pc_2. The ID and target stay unchanged.

The node configuration is:

{
  "method": "pca",
  "columns": ["tenure", "monthly_charges", "support_calls"],
  "n_components": 2,
  "prefix": "pc",
  "seed": 42
}

Both flow engines run this ML step through pandas/scikit-learn. The Polars script includes the actual conversion into pandas and back into Polars.

pandas

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

df_customers = pd.read_csv('customers.csv')
df_customers = df_customers.copy()
df_customers[['tenure', 'monthly_charges', 'support_calls']] = StandardScaler().fit_transform(df_customers[['tenure', 'monthly_charges', 'support_calls']])
_cols = ['tenure', 'monthly_charges', 'support_calls']
_pca = PCA(n_components=min(2, len(_cols), len(df_customers)), random_state=42)
_comp = _pca.fit_transform(df_customers[_cols])
_names = [f'pc_{i + 1}' for i in range(_comp.shape[1])]
df_customers = pd.concat([df_customers.drop(columns=_cols), pd.DataFrame(_comp, columns=_names, index=df_customers.index)], axis=1)
df_customers.to_csv('reduce-dimensions.csv', index=False)

Polars

import polars as pl
import pandas as pd
from sklearn.decomposition import PCA
from sklearn.preprocessing import StandardScaler

df_customers = pl.read_csv('customers.csv')
_eager_1 = df_customers.to_pandas()
_eager_2 = _eager_1.copy()
_eager_2[['tenure', 'monthly_charges', 'support_calls']] = StandardScaler().fit_transform(_eager_2[['tenure', 'monthly_charges', 'support_calls']])
df_1 = pl.from_pandas(_eager_2)
_eager_3 = df_1.to_pandas()
_cols = ['tenure', 'monthly_charges', 'support_calls']
_pca = PCA(n_components=min(2, len(_cols), len(_eager_3)), random_state=42)
_comp = _pca.fit_transform(_eager_3[_cols])
_names = [f'pc_{i + 1}' for i in range(_comp.shape[1])]
_eager_4 = pd.concat([_eager_3.drop(columns=_cols), pd.DataFrame(_comp, columns=_names, index=_eager_3.index)], axis=1)
df_2 = pl.from_pandas(_eager_4)
df_2.write_csv('reduce-dimensions.csv')

Tips & common mistakes

• Choose feature columns explicitly to avoid replacing an ID or target.
• PCA centers values but does not scale different units automatically; this exploratory example scales first.
• 1 means one component; 1.0 is not a valid variance fraction for this node.
• Scaling and PCA here refit on the incoming frame. Do not fit them independently on held-out data; retain fitted transformations in a pipeline for evaluation or prediction.
• Integer counts are capped; fractions are passed through to scikit-learn.

See also

• Scale Features · Select Features
• Train/Test Split
• PCA in scikit-learn