https://github.com/debasish-dutta/handwritten-digit-classifier

A web-app based on a handwritten digit recognizer TF model and its neural network created from scratch and is deployed using a web-app with tensorflow.js
https://github.com/debasish-dutta/handwritten-digit-classifier

deep-neural-networks tensorboard-mnist-digit tensorflow2

Last synced: 22 days ago
JSON representation

A web-app based on a handwritten digit recognizer TF model and its neural network created from scratch and is deployed using a web-app with tensorflow.js

• Host: GitHub
• URL: https://github.com/debasish-dutta/handwritten-digit-classifier
• Owner: debasish-dutta
• Created: 2020-07-15T15:04:45.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2022-09-19T18:51:14.000Z (over 2 years ago)
• Last Synced: 2024-11-18T05:35:06.486Z (3 months ago)
• Topics: deep-neural-networks, tensorboard-mnist-digit, tensorflow2
• Language: Jupyter Notebook
• Homepage: https://debasish-dutta.github.io/math-garden-website/
• Size: 8.42 MB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Handwritten-Digit-Classifier

## Problem

In this I tend to create a neural network model to classify handwritten digits.

![nn]

## Data

The data is taken from MINST dataset and then is divided into training, validation and test datasets of about 50k, 10k and 10k subsecuently.

The data are greyscale images of 28 by 28 px. These are then flattened into about 784px.

## Model Creatiion

The model is created withh 3 laters - the input layer with 2 hidden layer and the output layer with a learning rate of le-3. The first hidden layer consists of 512 nodes and the 2nd hidden layer with 64 nodes. The model is then trained in batched of 1000 and total of 50 epochs.

![graph]

## Evaluation

The model is evaluated on 2 metrices - accuracy and loss. It uses the Adam optimizer.

Accuracy Graph

![graph_acc]

Loss Graph

![graph_cost]

---

The histogram chart of the model on different layers are

layer1_biases

![layer1_biases]

layer1_weights_train

![layer1_weights_train]

out_biases

![out_biases]

out_weights

![out_weights]

## Saving

The model is saved using the tf.v1 Saved Model method.

## Accuracy on test set is 97.92%.

[nn]:tensorboard_mnist_digit_logs/graphs_models/nn.jpg

[graph]:tensorboard_mnist_digit_logs/graphs_models/graph.png

[graph_acc]:tensorboard_mnist_digit_logs/graphs_models/graph_acc.png

[graph_cost]:tensorboard_mnist_digit_logs/graphs_models/graph_cost.png

[layer1_biases]:tensorboard_mnist_digit_logs/graphs_models/layer1_biases.png

[layer1_weights_train]:tensorboard_mnist_digit_logs/graphs_models/layer1_weights_train.png

[out_biases]:tensorboard_mnist_digit_logs/graphs_models/out_biases.png

[out_weights]:tensorboard_mnist_digit_logs/graphs_models/out_weights.png