Kaggle MNIST experiences

Working with MNIST data set
The next Kaggle competition which I attempted was working with MNIST data set which is also known as Digit recogniser. I used CNN(Convolutional Neural Networks) for predictions. 
Let us dig deeper step by step with the dataset. Kaggle has already spilt the datasets into train and test.
Visualizing the data
This is the image which we obtain// Couldnt paste an image over here, will be uploaded separately

Specific Counts for each and every digit
1    4684
7    4401
3    4351
9    4188
2    4177
6    4137
0    4132
4    4072
8    4063
5    3795
So it is evident that we have almost similar amount of digits.
We have to check if any column is required or not and if there is something which can be avoided, we can remove the specific column.
I don’t see any necessity to remove any column, so we will keep the data untouched.
[The code can be seen in the corresponding python file].
We perform a reshaping operation and there is a specific reasoning behind it, the KERAS convolution layers expect a tenor in the following shape:
[batch_size, image_width, image_height, channels]
In which image width and height are self-explanatory. Batch size points out to the size of the subset of examples when performing gradient descent. Channels refer to number of colour channels.
The next step which I performed is one hot encoding i.e. using to_categorical ten classes are created,  one for each digit.
Step-2 
Once we are done with data pre-processing, it is now the turn to create the Neural Network Model. I have followed the one provided in TensorFlow tutorial.
Convolution->ReLu->MaxPooling->Convolution->ReLu->MaxPool->Flatten ,in between drop out details are provided as well. Dropouts are used to overcome the problem of overfitting i.e. a percentage of what is learnt by the model is erased. 
Once the model is created, next step is to perform fit and predict. Initially I have used 5 epochs and no data augmentation is not done yet. I will be updating the details further in coming days.
The accuracy is now 99.02 and it will improve gradually as I will be introducing data augmentation and running with higher epochs to achieve better accuracy.
The accuracy was 92% initially and now it is 99.02%.  
Look into my_MNIST.py file for the code and further details.