brain2vec_PCA.py

#!/usr/bin/env python3

"""
pca_autoencoder.py

Adjustments requested:
  1. Only fit on scans with a 'train' label in the inputs.csv 'split' column.
  2. An option to either run incremental PCA or standard PCA.

Example usage:
    python pca_autoencoder.py \
        --inputs_csv /path/to/inputs.csv \
        --output_dir ./pca_outputs \
        --pca_type standard \
        --n_components 100
"""

import os
import argparse
import numpy as np
import pandas as pd

import torch
from torch.utils.data import DataLoader

from monai import transforms
from monai.data import Dataset, PersistentDataset

# We'll import both PCA classes, and decide which to use based on CLI arg.
from sklearn.decomposition import PCA, IncrementalPCA


###################################################################
# Constants for your typical config
###################################################################
RESOLUTION = 2
INPUT_SHAPE_AE = (80, 96, 80)
DEFAULT_N_COMPONENTS = 1200


###################################################################
# Helper: get_dataset_from_pd (same as in brain2vec_linearAE.py)
###################################################################
def get_dataset_from_pd(df: pd.DataFrame, transforms_fn, cache_dir: str):
    """
    Returns a monai.data.Dataset or monai.data.PersistentDataset
    if `cache_dir` is defined, to speed up loading.
    """
    if cache_dir and cache_dir.strip():
        os.makedirs(cache_dir, exist_ok=True)
        dataset = PersistentDataset(data=df.to_dict(orient='records'),
                                    transform=transforms_fn,
                                    cache_dir=cache_dir)
    else:
        dataset = Dataset(data=df.to_dict(orient='records'),
                          transform=transforms_fn)
    return dataset


###################################################################
# PCAAutoencoder
###################################################################
class PCAAutoencoder:
    """
    A PCA 'autoencoder' that can use either standard PCA or IncrementalPCA:
      - fit(X): trains the model
      - transform(X): get embeddings
      - inverse_transform(Z): reconstruct data from embeddings
      - forward(X): returns (X_recon, Z)

    If using standard PCA, we do a single call to .fit(X).
    If using incremental PCA, we do .partial_fit on data in batches.
    """
    def __init__(self, n_components=DEFAULT_N_COMPONENTS, batch_size=128, pca_type='incremental'):
        """
        Args:
            n_components (int): number of principal components to keep
            batch_size (int): chunk size for either partial_fit or chunked .transform
            pca_type (str): 'incremental' or 'standard'
        """
        self.n_components = n_components
        self.batch_size = batch_size
        self.pca_type = pca_type.lower()

        if self.pca_type == 'standard':
            self.ipca = PCA(n_components=self.n_components, svd_solver='randomized')
        else:
            # default to incremental
            self.ipca = IncrementalPCA(n_components=self.n_components)

    def fit(self, X: np.ndarray):
        """
        Fit the PCA model. If incremental, calls partial_fit in batches.
        If standard, calls .fit once on the entire data matrix.
        X: shape (n_samples, n_features)
        """
        if self.pca_type == 'standard':
            # Potentially large memory usage, so be sure your system can handle it.
            self.ipca.fit(X)
        else:
            # IncrementalPCA
            n_samples = X.shape[0]
            for start_idx in range(0, n_samples, self.batch_size):
                end_idx = min(start_idx + self.batch_size, n_samples)
                self.ipca.partial_fit(X[start_idx:end_idx])

    def transform(self, X: np.ndarray) -> np.ndarray:
        """
        Project data into the PCA latent space in batches for memory efficiency.
        Returns Z with shape (n_samples, n_components)
        """
        results = []
        n_samples = X.shape[0]
        for start_idx in range(0, n_samples, self.batch_size):
            end_idx = min(start_idx + self.batch_size, n_samples)
            Z_chunk = self.ipca.transform(X[start_idx:end_idx])
            results.append(Z_chunk)
        return np.vstack(results)

    def inverse_transform(self, Z: np.ndarray) -> np.ndarray:
        """
        Reconstruct data from PCA latent space in batches.
        Returns X_recon with shape (n_samples, n_features).
        """
        results = []
        n_samples = Z.shape[0]
        for start_idx in range(0, n_samples, self.batch_size):
            end_idx = min(start_idx + self.batch_size, n_samples)
            X_chunk = self.ipca.inverse_transform(Z[start_idx:end_idx])
            results.append(X_chunk)
        return np.vstack(results)

    def forward(self, X: np.ndarray) -> tuple[np.ndarray, np.ndarray]:
        """
        Mimics a linear AE's forward() returning (X_recon, Z).
        """
        Z = self.transform(X)
        X_recon = self.inverse_transform(Z)
        return X_recon, Z


###################################################################
# Load and Flatten Data
###################################################################
def load_and_flatten_dataset(csv_path: str, cache_dir: str, transforms_fn) -> np.ndarray:
    """
    1) Reads CSV.
    2) Filters rows if 'split' in columns => only keep 'split' == 'train'.
    3) Applies transforms to each image, flattening them into a 1D vector (614,400).
    4) Returns a NumPy array X: shape (n_samples, 614400).
    """
    df = pd.read_csv(csv_path)

    # Filter only 'train' if the split column exists
    if 'split' in df.columns:
        df = df[df['split'] == 'train']
    # If there is no 'split' column, we assume the entire CSV is for training.

    dataset = get_dataset_from_pd(df, transforms_fn, cache_dir)
    loader = DataLoader(dataset, batch_size=1, num_workers=0)

    # We'll store each flattened volume in a list, then stack
    X_list = []
    for batch in loader:
        # batch["image"] shape => (1, 1, 80, 96, 80)
        img = batch["image"].squeeze(0)  # => (1, 80, 96, 80)
        img_np = img.numpy()
        flattened = img_np.flatten()  # => (614400,)
        X_list.append(flattened)

    if len(X_list) == 0:
        raise ValueError("No training samples found (split='train'). Check your CSV or 'split' values.")

    X = np.vstack(X_list)
    return X


###################################################################
# Main
###################################################################
def main():
    parser = argparse.ArgumentParser(description="PCA Autoencoder with MONAI transforms and 'split' filtering.")
    parser.add_argument("--inputs_csv", type=str, required=True,
                        help="Path to CSV with at least 'image_path' column, optional 'split' column.")
    parser.add_argument("--cache_dir", type=str, default="",
                        help="Cache directory for MONAI PersistentDataset (optional).")
    parser.add_argument("--output_dir", type=str, default="./pca_outputs",
                        help="Where to save PCA model and embeddings.")
    parser.add_argument("--batch_size_ipca", type=int, default=128,
                        help="Batch size for partial_fit or chunked transform.")
    parser.add_argument("--n_components", type=int, default=1200,
                        help="Number of PCA components to keep.")
    parser.add_argument("--pca_type", type=str, default="incremental",
                        choices=["incremental", "standard"],
                        help="Which PCA algorithm to use: 'incremental' or 'standard'.")
    args = parser.parse_args()

    os.makedirs(args.output_dir, exist_ok=True)

    # define transforms as in brain2vec_linearAE.py
    transforms_fn = transforms.Compose([
        transforms.CopyItemsD(keys={'image_path'}, names=['image']),
        transforms.LoadImageD(image_only=True, keys=['image']),
        transforms.EnsureChannelFirstD(keys=['image']),
        transforms.SpacingD(pixdim=RESOLUTION, keys=['image']),
        transforms.ResizeWithPadOrCropD(spatial_size=INPUT_SHAPE_AE, mode='minimum', keys=['image']),
        transforms.ScaleIntensityD(minv=0, maxv=1, keys=['image']),
    ])

    print("Loading and flattening dataset from:", args.inputs_csv)
    X = load_and_flatten_dataset(args.inputs_csv, args.cache_dir, transforms_fn)
    print(f"Dataset shape after flattening: {X.shape}")

    # Build the PCAAutoencoder with chosen type
    model = PCAAutoencoder(
        n_components=args.n_components,
        batch_size=args.batch_size_ipca,
        pca_type=args.pca_type
    )

    # Fit the PCA model
    print(f"Fitting {args.pca_type.capitalize()}PCA in batches...")
    model.fit(X)
    print("Done fitting PCA. Transforming data to embeddings...")

    # Get embeddings & reconstruction
    X_recon, Z = model.forward(X)
    print("Embeddings shape:", Z.shape)         # (n_samples, n_components)
    print("Reconstruction shape:", X_recon.shape)  # (n_samples, 614400)

    # Save
    embeddings_path = os.path.join(args.output_dir, "pca_embeddings.npy")
    recons_path = os.path.join(args.output_dir, "pca_reconstructions.npy")
    np.save(embeddings_path, Z)
    np.save(recons_path, X_recon)
    print(f"Saved embeddings to {embeddings_path}")
    print(f"Saved reconstructions to {recons_path}")

    # Optionally save the actual PCA model with joblib
    # from joblib import dump
    # ipca_model_path = os.path.join(args.output_dir, "pca_model.joblib")
    # dump(model.ipca, ipca_model_path)
    # print(f"Saved PCA model to {ipca_model_path}")


if __name__ == "__main__":
    main()