metadata

title: MnistStudio
emoji: 🐨
colorFrom: red
colorTo: indigo
sdk: docker
app_port: 8000
pinned: false
license: mit
short_description: Train and perform inference on MNIST dataset

MNIST Application

Overview

This is a simple application that can be used to train a convolutional neural network model to classify images of handwritten digits. The same application can also be used to perform inference of the digits drawn by the user.

Application Description

The landing page consists of two buttons, one for training the model and one for performing inference.
On clicking the inference button, a new page is loaded where the user can draw a digit on the canvasand select the model to perform inference.
The inference results are displayed on the same page.
On clicking the training button, a new page is loaded where two buttons are displayed, one for training single model and another for comparing multiple models.
On clicking the train single model button, a new page is loaded where the user can select following options:
- Number of kernels of three blocks of the network
- Optimizer [Admam, SGD]
- Batch Size [32, 64, 128]
- Number of Epochs [1, 2, 3]
Once these parameters are selected, the user can click on the train button to start the training. Training and validation loss, accuracy are displayed on the same page.
On clicking the train and compare models button, a new page is loaded where the user can select following options for both the models:
- Number of kernels of three blocks of the network for each model
- Optimizer [Admam, SGD] for each model
- Batch Size [32, 64, 128] for each model
- Number of Epochs [1, 2, 3] for each model
Once these parameters are selected, the user can click on the train button to start the training. Training and validation loss, accuracy are displayed on the same page.