1. Dataset files for training and validating Shared Frame Data File: A consolidated binary stream of raw pixel data extracted from behavioral videos of multiple subjects. By using a fixed frame size and a predictable storage order, this format allows for efficient, random-access retrieval of any specific frame. Storage Order: Temporal (sequential) by extraction timestamp, concatenated across subjects. Specifications: Little-Endian, 224x224 pixels, Grayscale (1 Channel), 8-bit depth (uint8). Frame Size: 50,176 bytes (224 * 224 * 1).

   Shared Sequence Data Offset File:
   Shared Sequence Index Purpose: Acts as a lookup table to map sequence start points to their exact byte locations in the Shared Frame Data File, specifically accounting for the stride (the frame interval) between the start of one sequence and the next. Data Entry: Each entry represents the byte offset of the first frame in a sequence. Sequence Definition: A group of N sequential frames. Data Type: 64-bit Unsigned Integer (uint64) per offset.

   Shared sequence label data file: Shared Sequence Labels Purpose: Stores a label for every sequence indexed in the Shared Sequence Data Offset File. Data Mapping: 1:1 relationship—each entry corresponds to the sequence at the same index in the offset file. Entry Definition: Represents the label associated with the final frame of the sequence. Data Type: 32-bit Float (float32).

2. Training project (.\train) Overview datasets.py: Features memory-mapped loading for training and validation datasets, including logic to balance positive and negative samples. model.py: Contains the CNN-LSTM architecture implementation. train.py: Manages the main training loop, model optimization, and validation logic. main.py: The entry point for training, used to configure dataset paths, shapes, hyperparameters, schedulers, logging, and model saving.

3. Prediction Project (.\prediction) Overview dataset.py: Features memory-mapped loading for efficient handling of video datasets. model.py: Contains the CNN-LSTM architecture implementation. prediction.py: Manages the main prediction logic. main.py: The entry point for the project. It initializes the training settings required to load models and video datasets and configures the prediction parameters.

4. requirements.txt: Lists the Python dependencies required to run the training project.

5. Trained model (.\model) temporal_lick_classifier.pth: The saved weights for the trained model.