https://github.com/jai0212/koach-sample-project

https://github.com/jai0212/koach-sample-project

Last synced: about 4 hours ago
JSON representation

• Host: GitHub
• URL: https://github.com/jai0212/koach-sample-project
• Owner: Jai0212
• Created: 2024-09-16T19:15:11.000Z (2 months ago)
• Default Branch: main
• Last Pushed: 2024-09-17T00:49:08.000Z (2 months ago)
• Last Synced: 2024-09-18T01:54:50.260Z (2 months ago)
• Language: Python
• Size: 2 MB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Koach Sample Project

A CNN using the **MNIST dataset** to identify the number from its image with an accuracy of **99.18%**, created in Python using TensorFlow.

## Model Architecture

* Convolutional Layer

* Pooling Layer

* Batch Normalization

* Dropout Layer

* Flattening Layer

* Dense Layer

### Adam Optimizer:

* Learning Rate: 0.001

* Loss: sparse_categorical_crossentropy

## Packages

* TensorFlow

* Keras

* SciKit Learn

* Numpy

* Matplotlib

* Pickle

## Files:

* **main.py** - model created and saved

* **predict.py** - used to predict any image, just put any image and put its title in line 9 of the code as image_path (ensure the number is of a lighter shade and the background is darker)

* **plot.py** - used to plot the training and validation loss graph

* **mnist_cnn_model.h5** - the CNN model

* **model_history.pkl** - stores model training history for plotting graph

* **test.png** - sample testing file

##





### Batch Normalization

This layer is used to normalize the input ensuring stability and preventing overfitting. It also increases the training speed. 

### Dropout Layers

This layer randomly drops out some neurons to prevent overfitting and helps in generalization.

### Use of these Layers

I have used these layers in this project to improve the model as it prevents overfitting and helps in generalization thereby improving accuracy. Since it is a simple project with a simple dataset it's not super useful. It was only slightly beneficial. For more complex models, these layers will be of great use.