How to get the 1.0 score in Digit Recognizer?

Hello! My name’s Makar Baderko, and I’m a 12 years old machine learning developer. Today we will try to get the top place in one of Kaggle's most popular computer vision competitions — Digit Recognition, with the famous MNIST dataset. It will be not very difficult, so let’s get started!

I recommend you to run this code in Kaggle Notebooks

At first, let’s import some needed libraries:

After that, let’s load our data and look at sample submission:

We would like to take a closer look at some of our pictures.

On this step, you need to download full MNIST dataset from this website:
https://www.kaggle.com/oddrationale/mnist-in-csv

Here, you will create an Image Data Generator, that will make the data look best for our model.

Here we will create a TensorFlow model:

You can find the meaning of all that params here:
https://www.tensorflow.org/js/guide/models_and_layers

Let’s fit our model, to the MNIST data.

I’ll make predictions on Kaggle’s test dataset

And save them to a CSV file

Thank you for reading, follow me on Medium! :)