Better answers in Linear Regression exercise

[jorgeorpinel]

This notebook is linked from https://developers.google.com/machine-learning/crash-course/linear-regression/programming-exercise

Suggested improvements:

1. Better code to get the most common payment type
2. Improved explanation around how fast the first experiment converges:
   • Epoch no. 5 in the plot is actually the 6th epoch (since they're 0-indexed)
   • Quantified how well the model fits the data, putting RMSE in context
3. Expands explanation around a larger batch size (2nd set of experiments): The reason 20 epochs are not enough for the training to converge is that the model "learns" less times in each epoch.

Note: A helper notebook used for this PR can be found and run at https://colab.research.google.com/drive/16xShaKthamjBGkSWM2gqIo3pB1xvYwLA .

[jorgeorpinel]

Here, the answer could use this code:

training_corr['FARE'].abs().nlargest(2).index[1]

[jorgeorpinel]

And here:

training_corr['FARE'].abs().nsmallest(2).index[0]

[jorgeorpinel]

Just noticed another potential improvement ☝🏼

Left comments instead of updating the PR code since the repo seems to be inactive (no response at all here)...