Competition : https://www.kaggle.com/c/digit-recognizer
Detailed Explanation : MNIST.ipynb (CNN keras, with 0.992% accuracy on kaggle)
- 1. Data preparation
- 1.1 Load train data
- 1.2 Check for null and missing values
- 1.3 Normalization
- 1.4 Reshape
- 1.5 Label encoding
- 1.6 Split training and valdiation set
- 2. CNN
- 2.1 Define the model
- 2.2 Set the optimizer and annealer
- 3. Evaluate the model
- 3.1 Training and validation curves
- 3.2 Confusion matrix
- 4. Prediction and submition
- 4.1 Load test data
- 4.1 Predict and Submit results