Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Gurneet1928/TryMLEasy/llms.txt

Use this file to discover all available pages before exploring further.

After the optional preprocessing step and before the ML model itself, TryMLEasy can apply a dimensionality reduction technique to your feature set. This step sits as the second stage of the scikit-learn Pipeline and is configured on the Feature Selection page via the “Select Decomposition Step” dropdown. Reducing the number of dimensions can improve model generalisation, reduce training time, and help remove noise or highly correlated features.
Decomposition is applied only for Traditional ML models. It is not part of the neural network workflow — the neural network pages receive the raw train/test splits directly from the Feature Selection page.

None

Selecting None skips decomposition entirely. All selected feature columns are passed as-is to the model (or to the scaler first, if one was chosen).
  • scikit-learn class: (not applied)
  • When to use: When your feature count is already small, when interpretability of original features is important, or when you want to assess model performance on the uncompressed feature space first.

PCA

Principal Component Analysis (PCA) is a linear dimensionality reduction technique. It projects the data onto a lower-dimensional space by finding the directions (principal components) that maximise explained variance.
  • scikit-learn class: sklearn.decomposition.PCA
  • Linear: Yes
  • When to use: When your dataset has many correlated numeric features and the underlying structure is approximately linear. PCA is most effective after a zero-mean scaling step such as Standard Scaler, because it is sensitive to the scale of features.

Kernel PCA

Kernel PCA extends standard PCA by applying the kernel trick — it implicitly maps the data to a higher-dimensional feature space via a kernel function before performing PCA. This allows it to capture non-linear relationships that standard PCA would miss.
  • scikit-learn class: sklearn.decomposition.KernelPCA
  • Linear: No
  • When to use: When your data has non-linear structure that PCA cannot unfold. A natural upgrade to try when PCA does not improve — or actively hurts — model performance.
The dropdown label for this option in the TryMLEasy UI reads “Kernal PCA” (a known typo in the source). The correct spelling is “Kernel PCA”, as used throughout this documentation.
TryMLEasy exposes Kernel PCA with its default kernel (rbf). Advanced kernel selection is not available in the current version.

FastICA

FastICA (Fast Independent Component Analysis) decomposes a multivariate signal into statistically independent additive subcomponents. Unlike PCA, which maximises variance, ICA maximises statistical independence between components.
  • scikit-learn class: sklearn.decomposition.FastICA
  • Linear: Yes (linear mixing assumed)
  • When to use: When you suspect the observed features are mixtures of independent source signals — for example in sensor data, physiological measurements, or audio feature sets. Less common for general tabular ML, but useful for signal separation tasks.

Comparison Table

Methodscikit-learn ClassLinearBest For
None(not applied)N/ASmall feature sets; interpretability-first workflows
PCAPCAYesCorrelated numeric features; paired with Standard Scaler
Kernel PCAKernelPCANoNon-linear feature relationships
FastICAFastICAYesIndependent signal separation; sensor / physiological data
PCA after Standard Scaler is a classic and effective combination. First, Standard Scaler ensures all features contribute equally (zero mean, unit variance); then PCA extracts the directions of greatest variance. Try this pairing as your default when working with more than ~10 correlated features before exploring Kernel PCA or FastICA.

Build docs developers (and LLMs) love