Project 1 FAQ

This page was last updated: Friday, 03-Jul-2020 13:39:30 AEST by blair@cse.unsw.edu.au

    1. Have any of the files been updated ?

    Yes, hw1/src/part2.py was updated at 7pm on Friday 11 October and at 2pm on Tuesday 15 October (hw1.zip was also updated).

    Changes made at 7pm on Friday 11 October:

    114d113
    <         cost = 0
    120c119
    <             cost += model.loss(output, y)
    ---
    >             cost = model.loss(output, y)
    132c131
    <         print(f" Cost: {cost/len(inputs):.2f} Accuracy: {num_correct / len(inputs) * 100:.2f}%")
    ---
    >         print(f" Cost: {cost:8.6f} Accuracy: {num_correct / len(inputs) * 100}%")
    
    When the code is run, the new version correctly prints the average of the loss function at each epoch. You are encouraged to either download the new version, or to mirror the changes in your own code (although, it will not make any difference to the automarking).

    Changes made at 2pm on Tuesday 15 October:

    75c75
    <         dE/ds = z - t (which is passed in as diff)
    ---
    >         -dE/ds = t - z (which is passed in as diff)
    
    This is a change in the comments, which now correctly specifies the sign of the diff parameter.

    2. Was there a problem with the Figure showing the output for Part 2 ?

    Yes, somehow the pickle file got changed after the figure was generated. We have now replaced this with a newly generated figure. The accuracy reaches 92% or 94%, with one of the points lying very close to the boundary.

Back to Project 1 | Back to the main page