Provisional Marks (subject to Faculty approval) are now available through SMS.
There will be pre-Exam Consultation on Tuesday 6 November from 12 to 2pm in K17-201B.
Marks for assignment 3 are now available on give.
Some late submissions are still being tested.
If you feel your mark is incorrect, please first note the following;
- If mark is greater than 7, the automarking scripts ran without error and the mark is unlikely to be changed. If less than 7, there was an error in your script and marks were manually assigned.
- A large amount of submissions did not complete 500 episodes in the specified 5min timeout period. Consequently, we extended the mark timeout to 7 minutes, however some were still over this. Suspicion that your script may have timed out is not a valid reason for a remark.
- Syntax errors in the submitted file are not a valid reason for a remark - it is your responsibility to test your submission prior to submitting it. Part marks are still assigned in these cases based on the report.
- Scripts failing due to importing external libraries such as pandas is not a valid reason for a remark.
If you still feel your mark is incorrect, follow the same steps outlined in the previous notice regarding A2. That is:
1) Clear explanation of why you suspect an error in the assigned mark
2) Clear explanation of what must be altered/changed for submission to be fairly assessed.
3) Screenshot of everything working locally at least up to episode 500.
Email the above to alex . long @unsw.edu.au with the subject: comp9444 <group number> prior to the 31st of Oct for consideration.
ALL PREVIOUS REQUESTS MUST BE RESENT ACCORDING TO THE BELOW TEMPLATE, AND MUST BE RECEIVED PRIOR TO WEDNESDAY 24th OCTOBER FOR CONSIDERATION. IF YOU DO NOT WISH TO REQUEST A REMARK YOU DO NOT NEED TO TAKE ANY ACTION.
Due to the higher than expected volume of remark requests, special consideration will be given for this assignment and some submissions will be manually reviewed. If you wish to appeal your A2 mark, please carry out the below steps and send an email to alex.long@unsw.edu.au in the format specified below. Requests that are not in this format will not be considered. If you have already emailed regarding a remark, you must do so again following the below instructions.
Note that if your mark is above 6, your model was successfully loaded and automatically graded and will not be reviewed/updated.
If you feel your mark is incorrect you should, prior to requesting a review:
1. Download and extract a fresh version of the src.zip to a new directory, then copy ONLY your submitted files to this directory, replacing implementation.py. Confirm that the model submitted runs locally with an unaltered runner.py with the eval flag, in python 3.6/3.7 and Tensorflow 1.9. Include a screenshot of this in your email.
2. Confirm you are not importing libraries outside of the stdlib (i.e. nltk)
3. Confirm your implementation is not loading external assets that are not present in the submission.
Only after following the above steps, send an email in the following format.
EMAIL TEMPLATE:
Subject: COMP9444: <Your group number here>
Body:
In every case, attach the screenshot of the final model accuracy on the eval set after step 1 above. If this fails, attach a screenshot of final model accuracy of your model running in your original development environment.
- If the model in step 1 runs fine, make the email body text "No modifications necessary".
- If you're model does not run after following step 1, include specific, step-by-step instructions on the modifications that must be made to allow the model to be evaluated.
- If you did not submit checkpoint files on time, due to excessive size, or naming restrictions, attach these to your email with unaltered timestamps.
Click on Assignments under the Course Work menu to the left.
Click on Assignments under the Course Work menu to the left.
Clarification has been added for the my_perceptron function, please see the FAQ
here
.
The first lecture for this course will be at 6pm on Monday 23 July, in Central Lecture Block 7.
The old COMP9444 Web site from 2017 can be found here .