src.train_autoencoder

train_autoencoder.py

Loads preprocessed sensor data from a .hdf5 file, applies normalization, trains a Transformer autoencoder with configurable parameters, and saves models and training logs.

Usage:

python train_transformer.py –input data.hdf5 –epochs 100 –batch_size 32 –head_size 128 –num_heads 4 –ff_dim 256 –dropout 0.2 –num_blocks 3 –output models/

Functions

build_transformer_autoencoder(timesteps, ...)	Builds a Transformer-based autoencoder for time series data.
load_and_normalize_from_hdf5(hdf5_path[, ...])	Load and normalize 3D sensor data from an HDF5 file.
main()	Entry point for training a Transformer autoencoder on time series data using K-Fold cross-validation.
set_seeds([seed])	Set seeds for reproducibility across NumPy, TensorFlow, and Python.
train_model_kfold(X[, output_dir, epochs, ...])	Train a Transformer autoencoder using K-Fold cross-validation.
transformer_encoder(inputs[, head_size, ...])	Creates a single Transformer encoder block with multi-head attention and feed-forward layers.