neuralnet_visualization

Data visualization tool for neural networks

Tech Stack

• D3.js

Installation

1. Clone the repo and cd into the directory.

   git clone https://github.com/vanillabase1lb/neuralnet_visualization.git
   cd neuralnet_visualization/

2. Install required packages using npm install.

3. Run ./test_data_generator/ann_data_generator.py to get sample data (nodes.json and weights.json) or generate own data.

4. Place nodes.json and weights.json in ./public/data/.

5. Start webpack dev server using npm run dev.

Custom ANN visualization

The application only reads data from nodes.json and weights.json. Any such ANN data structured in the below mentioned "standard" may be accepted. The ./test_data_generator/ directory also contains a reference implementation using custom callbacks in Keras to get nodes and weights data after every epoch.

nodes.json format:

[ --> only 1 should be present at the beginning and ending of JSON
[ --> 1 for every epoch
[ --> 1 for every layer of the ANN
[ --> 1 for every batch
   N comma separated values where N = no. of nodes in the layer
]
]
]
]

weights.json format:

[ --> only 1 should be present at the beginning and ending of JSON
[ --> 1 for every epoch
[ --> 1 for every 2 adjacent layers of the ANN (should be equal to no. of layers - 1)
[ --> should be only 2. First one is weight matrix(n x m), second is bias array(m). Second one is currenly ignored, but should be present nevertheless
[ --> N number of arrays where N = no. of nodes in the first out of the 2 adjacent layers
    M comma separated values where M = no. of nodes in the second out of the 2 adjacent layers
]
]
]
]
]