dlcopilot 0.1.0

An assistant for training and analyzing deep learning models in PyTorch.

✅ README.md

DLCopilot 🚀

DLCopilot is your assistant for training and analyzing deep learning models in PyTorch. It provides a simple and extensible interface for visualizing model structure, input-output dimensions, training progress, and performance metrics — making your development process faster and more intuitive.

---

📦 Features

• 🔍 Inspect model layer-wise shapes and input-output dimensions
• 📊 Visualize class distribution in datasets
• 📈 Track and plot training and validation loss
• 🧪 Easy integration with your PyTorch training loop
• ✅ Lightweight and modular design

---

🛠️ Installation

Clone the repository and install the dependencies:

git clone https://github.com/KIREN2612/dlcopilot.git
cd dlcopilot
pip install -r requirements.txt

You can also install as a package (for development):

pip install -e .

---

🧪 Example Usage

from dlcopilot import DLCopilot
from your_model import MyModel  # Replace with your model
from torch.utils.data import DataLoader

# Initialize model, optimizer, loss, data loaders
model = MyModel()
optimizer = torch.optim.Adam(model.parameters())
criterion = torch.nn.CrossEntropyLoss()
train_loader = DataLoader(...)
val_loader = DataLoader(...)

# Create DLCopilot instance
copilot = DLCopilot(model, optimizer, criterion, train_loader, val_loader)

# Inspect architecture
copilot.inspect_input_output_shapes()
copilot.inspect_layer_shapes()

# Analyze data
copilot.class_distribution()

# Train and visualize
copilot.train_and_analyze(epochs=5)

See docs/usage.md for full examples.

---

🧪 Testing

Run tests with:

pytest tests/

Make sure to set the PYTHONPATH to the root of the project if needed:

export PYTHONPATH=$(pwd)  # On Linux/macOS
$env:PYTHONPATH = (Get-Location)  # On Windows PowerShell

---

📄 Documentation

• Usage Guide
• API Reference (Coming Soon)

---

🧑‍💻 Author

Kiren 📧 [e-mail] 🔗 github.com/KIREN2612

---

📜 License

This project is licensed under the terms of the MIT License. See LICENSE for more details.