The notebook contains model to classify the MNIST dataset using Tensorflow.

The notebook is divided into five main parts:

• Data Pre-processing
• Building the Model
• Optimizing the Model
• Evaluating the Model
• Training the Model

Classify.MNIST.with.Tensorflow.Video.mp4

Dependencies required for this notebook:

import tensorflow as tf

from tensorflow.keras.layers import Dense, Flatten, Conv2D
from tensorflow.keras import Model
import numpy as np

Note: TensorFlow version: 2.8.0 is used.

Notebook file in detail:

1. Data Pre-processing:
   • Import MNIST dataset and split the data into train and test set and add a channels dimensions.
   • Data Augmentation is conducted by rotating the images by 180 degrees.
   • Data is shuffled and batch size of 32 defined.
2. Building the Model:
   • Model is built applying Conv2D to build simple Convolutional Neural Network layer with activation function.
   • Flatten and Dense layer is applied.
3. Optimizing the Model:
   • Adam Optimizer is applied.
   • Sparse Categorical cross entropy loss is used to computes the crossentropy loss between the labels and predictions.
4. Evaluating the Model:
   • Performance of the model is evaluated using loss and accuracy for train and test state.
5. Training the Model:
   • Model is trained using tf.GradientTape which is Tensorflow's API for automatic differentiation.
   • Model is trained on epoch 6 with test accuracy of 98.45.

The following results could be improved by using Dropout and conducting more comprehensive data augmentation processes.

This notebook file is referenced from Tensorflow tutorials: https://github.com/tensorflow/docs/tree/master/site/en/tutorials