Dimensionality Reduction

Endpoint: POST /dimensionality-reduction

Request Parameters

File Upload

• file (required): The file to be processed. Supported formats include CSV, Excel, etc.

Form Parameters

• columns (optional): A comma-separated list of columns to include in the dimensionality reduction. If not provided, all numeric columns are used.
• auto_detect_numeric (optional, default: True): Whether to auto-detect numeric columns if columns is not provided.
• algorithm (required): The dimensionality reduction algorithm to apply. Supported values:
  • pca: Principal Component Analysis (PCA).
  • tsne: t-Distributed Stochastic Neighbor Embedding (t-SNE).
  • umap: Uniform Manifold Approximation and Projection (UMAP).
  • autoencoder: Autoencoder-based dimensionality reduction.
• n_components (required): The number of components (dimensions) to reduce the data to.
• normalize (optional, default: True): Whether to normalize the data before applying dimensionality reduction.
• check_missing_values (optional, default: False): Whether to check for missing values in the dataset.
• check_numeric (optional, default: False): Whether to ensure all selected columns are numeric.
• random_seed (optional): Random seed for reproducibility.

Algorithm-Specific Parameters

• PCA:
  • pca_whiten (optional, default: False): Whether to whiten the data.
  • pca_svd_solver (optional, default: auto): The SVD solver to use. Supported values: auto, full, arpack, randomized.
• t-SNE:
  • tsne_perplexity (optional, default: 30.0): The perplexity parameter for t-SNE.
  • tsne_learning_rate (optional, default: 200.0): The learning rate for t-SNE.
• UMAP:
  • umap_neighbors (optional, default: 15): The number of neighbors for UMAP.
  • umap_min_dist (optional, default: 0.1): The minimum distance for UMAP.
• Autoencoder:
  • autoencoder_layers (optional): A comma-separated list of layer sizes for the autoencoder (e.g., 64,32).
  • autoencoder_epochs (optional, default: 50): The number of training epochs for the autoencoder.
  • autoencoder_learning_rate (optional, default: 0.001): The learning rate for the autoencoder.

Processing Logic

1. Read the uploaded file: The file is converted into a Pandas DataFrame.
2. Validate inputs:
   • Checks if the dataset is empty.
   • Validates the selected columns (if provided).
   • Auto-detects numeric columns if auto_detect_numeric is True.
   • Checks for missing values if check_missing_values is True.
   • Ensures all selected columns are numeric if check_numeric is True.
   • Validates the number of components (n_components).
3. Normalize data: If normalize is True, the data is standardized using StandardScaler.
4. Apply dimensionality reduction:
   • PCA: Applies Principal Component Analysis.
   • t-SNE: Applies t-Distributed Stochastic Neighbor Embedding.
   • UMAP: Applies Uniform Manifold Approximation and Projection.
   • Autoencoder: Trains an autoencoder and uses the bottleneck layer for dimensionality reduction.
5. Return the transformed data: The reduced dataset is provided for download with the filename appended with _dimensionality_reduction.

Example Request

curl -X POST "http://localhost:8000/dimensionality-reduction" \
-F "file=@data.csv" \
-F "algorithm=pca" \
-F "n_components=2" \
-F "pca_whiten=False" \
-F "pca_svd_solver=auto" \
-F "normalize=True"