Project Repository Structure

This repository is structured to streamline data processing, modeling, and UI integration. Below is an overview of each folder and its purpose.

---

🚀 Main Application

(This section is currently empty. Fill in with relevant details about the main application.)

---

Data

(This section is currently empty. Fill in with relevant details about the data processing and structure.)

---

Data Processing

(This section is currently empty. Fill in with relevant details about data processing scripts and workflows.)

---

Documentation and Demos

(This section is currently empty. Fill in with relevant details about documentation, videos, and walkthroughs.)

---

Model

Dataset Structure

• Source (src): [batch_size, 22, 66]
  • Includes 6 match parameters and 60 player-specific parameters per player.
• Target (tgt): [batch_size, 22, 66]
  • Comprises 15 target parameters expanded to 66 dimensions for alignment.

---

Model Architecture

The model is based on a Transformer architecture, optimized for sequential data. Key components include:

1. Multi-Head Attention

   • Captures relationships between players and contextual match features using multiple attention heads.

2. Encoder

   • Processes the input sequence (match and player-level features) using stacked attention and feedforward layers.

3. Decoder

   • Generates player predictions by applying attention mechanisms to both the target and encoder outputs.

4. Loss Function

   • Uses a weighted Mean Absolute Error (MAE) loss:
     loss = weighted_mae_loss(prediction, target, lambda_val=10)
     • Weights are derived from Dream 11 fantasy points for each performance parameter.
     • Higher weight parameters (e.g., wickets, boundaries) are prioritized for accurate predictions.

---

Training Workflow

1. Data Loading and Batching

• Match data is structured into src (input) and tgt (output) tensors.
• A DataLoader is employed to process batches of size 32.
• Tensors are normalized for efficient training.

2. Training Loop

• Gradually increases the sequence length of the target (tgt) during training.
• Computes the loss at each step using the weighted MAE loss function.
• Updates model parameters using the Adam optimizer.

3. Saving the Model

• Saves the trained model weights to a .pth file.

---

Testing Workflow

1. Data Loading and Batching

• Processes match and player data into appropriate tensor structures.

2. Normalization

• Applies the same normalization procedure as in training.

3. Prediction

• Outputs are unnormalized to restore the original parameter scales.

---

Performance Metrics

1. Mean Absolute Error (MAE):
   • Computes the MAE between predicted and actual total team points.
2. Fantasy Points Evaluation:
   • Calculates fantasy points based on 14 predicted performance parameters.
   • Sorts players by predicted fantasy points and computes the top-11 total points.

---

---

Model Artifacts

(This section is currently empty. Fill in with relevant details about pre-trained models and artifacts.)

---

Out of Sample Data

(This section is currently empty. Fill in with relevant details about evaluation data and integration workflows.)

---

Rest

(This section is currently empty. Fill in with details about any miscellaneous requirements.)

---

UI

BASE_DIR: /src/UI/PRODUCT_UI

Key Components:

1. model.pt: ML Model file.
2. api/ml_model.py: ML model integrated with the Web UI.
3. api/agent.py: LLM Generative AI integrated feature.
4. api/data: Stores data for player-to-identifier mapping and player stats.

---

Deployed Site:

The Product UI is deployed and accessible at:
http://172.16.4.2:8080

The Product UI can be tested and utilized at the above URL.

---

Start the Application Locally

Frontend (cd frontend)

1. Install dependencies:

   pnpm i
   # or
   npm i --force

2. Start the development server:

   pnpm run dev
   # or
   npm run dev
   

Backend (cd backend)

1. Create and activate a virtual environment:

   python -m venv venv
   venv\Scripts\activate

2. Install dependencies:

   pip install -r requirements.txt

3. Apply database migrations:

   python manage.py makemigrations
   python manage.py migrate

4. Create a superuser:

   python manage.py createsuperuser

5. Enter the following details:
   1. Username: admin
   2. Email: admin@gmail.com
   3. Password: 1234
6. Update player data:

   python manage.py update_players.py

7. Update additional data:

   python manage.py update_data.py