https://github.com/yashas-naidu/conv_idd_64

https://github.com/yashas-naidu/conv_idd_64

Last synced: 2 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/yashas-naidu/conv_idd_64
• Owner: Yashas-naidu
• License: mit
• Created: 2025-03-08T10:20:56.000Z (3 months ago)
• Default Branch: main
• Last Pushed: 2025-03-28T08:41:47.000Z (3 months ago)
• Last Synced: 2025-04-12T04:16:58.719Z (2 months ago)
• Language: Python
• Size: 89.6 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Training and Prediction Guide

This document provides step-by-step instructions for setting up the environment, training the model, and generating predictions.

## Environment Setup

1. **Create a Conda Environment**

   ```sh

   conda create -p venv python=3.12.8 -y

   ```

   **Note:** Ensure that Python 3.12.8 was the version i had in my case,but u can replace with your version.

2. **Activate the Environment**

   ```sh

   conda activate 

   ```

3. **Install Required Packages**

   ```sh

   pip install -r requirements.txt

   ```

4. Download the pytorch package from [https://pytorch.org/get-started/locally/](https://pytorch.org/get-started/locally/)

   In my  case :

   ```sh

   pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

   ```

## Loading MovingMnist Data

To load the MovingMNIST dataset from the gzip file, use the following command:

   ```sh

   python data/mm.py

   ```

## Training the Model

To start training, run:

   ```sh

   python main.py

   ```

You can customize the number of epochs and batch size as per your requirements by modifying the respective parameters in `main.py`.

### Model Checkpoints

- The trained model is saved in the current directory in the following format:

  

  *checkpoint__.pth.tar*

  

- It is recommended to use a checkpoint with the lowest validation loss for better predictions.

## Generating Predictions

To generate predictions, load the desired checkpoint and run:

   ```sh

   python prediction.py

   ```

![](https://github.com/Yashas-naidu/ConvLSTM/blob/main/images/movingmnist.png)

## Viewing Logs in TensorBoard

Logs for **Epoch vs Loss** are stored inside the `runs` directory. To visualize them using TensorBoard, execute:

   ```sh

   tensorboard --logdir=./runs

   ```

This will launch TensorBoard, where you can analyze the training progress visually.

---

**Note:** Ensure that all dependencies are correctly installed before proceeding with training and prediction.