This repository contains an implementation of an anomaly detection framework for multivariate time series data using PyTorch. The framework integrates contrastive learning and a Generative Adversarial Network (GAN) to address overfitting and learn robust representations.

• Data preprocessing with MinMaxScaler from scikit-learn
• Data augmentation using geometric distribution masks
• Transformer-based Autoencoder for time series reconstruction
• Contrastive loss for learning robust representations
• GAN architecture with a generator and discriminator
• Adversarial loss for training the GAN
• Anomaly detection based on reconstruction error

The code is organized into several components:

1. Data Preprocessing: The preprocess_data function normalizes the input data using MinMaxScaler.

2. Data Augmentation: The geometric_mask function applies a geometric distribution mask to augment the input time series data.

3. Transformer-based Autoencoder: The TransformerAutoencoder class defines the Autoencoder model, which uses PyTorch's nn.TransformerEncoder and nn.TransformerDecoder modules.

4. Contrastive Loss: The contrastive_loss function computes the contrastive loss between the input and reconstructed time series.

5. GAN Models: The Generator and Discriminator classes define the generator and discriminator models for the GAN architecture.

6. Adversarial Loss: The adversarial_loss function computes the adversarial loss for training the GAN.

7. Training: The train function orchestrates the training process, iterating over the data in batches. It performs data augmentation, trains the Autoencoder with the contrastive loss, trains the GAN with the adversarial loss, and updates the discriminator with samples from the Autoencoder's latent space.

8. Anomaly Detection: The anomaly_detection function takes the trained Autoencoder and the input data, reconstructs the data using the Autoencoder, computes the reconstruction error, and identifies anomalies based on a threshold.

1. Prepare your multivariate time series dataset.
2. Import the necessary libraries and functions from the provided code.
3. Load your dataset and preprocess it using preprocess_data.
4. Set the desired hyperparameters (e.g., batch size, latent size, number of layers, and number of heads).
5. Call the train function with your data and hyperparameters to train the Autoencoder, Generator, and Discriminator models.
6. Use the anomaly_detection function with the trained Autoencoder and input data to identify anomalies based on the reconstruction error.

Note: You may need to adjust the model architectures and hyperparameters based on your specific dataset and requirements.

• PyTorch
• scikit-learn

Make sure to install the required dependencies before running the code.