All that is left to do for the final project is to take the ideas you laid out in HW4 and implement them. The assignment will be graded hollistically, but an approximate rubric is given below.
Again, the goal of the project is to say something interesting about computational representations of language–i.e. how do implementation choices affect the types of linguistic phenomena our models can or cannot capture? We understand that many experiments won’t produce the expected results, and that those of you with more ambitious projects may not complete all the necessary steps needed to draw the sweeping conclusion you hope to. Prioritize completing part of the project well over completing many disparate experiments sloppily.
You should use your HW4 as a starting point, and just refine/extend the intro sections if needed, populate the results section, and add some discussion interpretting your findings. I will leave comments in your HW4s before 12/16, you should take them into account when writing your final report.
If you get stuck and are not able to produce the results you thought you could produce, add a section on “Technical Challenges” or similar, and use it to communicate to us what you tried, why it did or didn’t work, where you got stuck, etc. If its clear you put in the appropriate level and quality work, you won’t be punished for not producing the desired results.
Again, the assignment will be graded hollistically. A rough scale to help you set your priorities is as follows.
On the last point, I know this is vague. If you are stuck on something, please reach out so we can help you debug in a principled way that will allow you to give a “compelling discussion” of your debugging strategy. The type fo thing that would not constitute “compelling” is: “the model wouldn’t train so I just tried 5 different dropouts and it still didn’t train so I turned it in.”
Submit your writeup as a single pdf. Run cs2952d_handin final to submit every file within the current directory.