Scraply

A no-code, deep learning platform 🚀 -- The "Scratch" for Neural Networks

• Previously awarded Best Developer Tool (HackPrinceton'24)
• Watch our new demo video

1) Drag and drop neural network layers. View your model's PyTorch configuration

2) Set training parameters - updated live

After training, you can also download a python notebook. See the code for everything you just did on Scraply!

3) View outputs - includes ✨️special✨️ visualization with image datasets

What are PEEK Maps?:

The heat maps shown when initializing a CNN are generated from a method called PEEK¹. It was developed by Mackenzie Meni and the NETS Lab at the Florida Institute of Technology. PEEK visualizes neural network decision-making by computing entropy-based maps of convolutional layers. It is able to create heat maps by highlighting the most information-rich regions in the input. Check out the paper here!

Some cool PEEK maps from the CIFAR10 image dataset:

Running locally:

The scraply server isn't deployed yet, therefore you need to run your own backend! We are working on cost-effective and possible funding/sponsor options to allow users to train their Scraply models for free :)

1. visit scraply (server status shows offline)
2. clone github repo git clone https://github.com/the-AMA-team/scraply.git
3. go to the api directory cd scraply/dynamic-model-api/
4. download python packages pip install -r requirements.txt
5. run server uvicorn app:socket_app --host 0.0.0.0 --port 5000 --reload

Updates in Summer'25 Release:

• 👾 explainability features
• 👾 outputs tab with model insights
• 👾 live training graph
• 👾 stop/resume training

Coming Soon Someday:

• 👾 ability to run in browser (using tf.js)
• 👾 uploading custom datasets and data pre-processing
• 👾 encoder support for transformers

Developed by the-AMA-team

Alan 🧑‍🍳: Cloud Ops

Mehek 🤓: Backend/AI

Adi 🤩: Frontend/UI

Footnotes

1. M. Meni, T. Mahendrakar, O. D. Raney, R. T. White, M. L. Mayo, and K. R. Pilkiewicz (2024). Taking a PEEK into YOLOv5 for Satellite Component Recognition via Entropy-based Visual Explanations. AIAA SCITECH 2024 Forum. https://arc.aiaa.org/doi/abs/10.2514/6.2024-2766 ↩