Which 3 items' accuracy is affected most by the horizontal flipping?

Comment on the inverted pixels evaluation. Why is the performance so low?

• Optimize the hyperparameters so that you achieve at least an 80% test accuracy. Keep the data split the same as above for fair comparison.
• Upload a screenshot showing the train fraction, your selected hyperparameter values, and the accuracies.
• Comment on the performance of your optimized model. Is the model underfitted and/or overfitted? Why do you think the model doesn't reach 100% accuracy?

Example for the screenshot:

Comment on the results above. Is the classifier good? Does it suffer from underfitting and/or overfitting? 

Take a look at the 2D visualization of the prediction. What seems to be a common pattern/problem?