Course talk:CPSC532:StaRAI:2017:Bahare
- [View source↑]
- [History↑]
Contents
Thread title | Replies | Last modified |
---|---|---|
Feedback | 0 | 20:18, 26 April 2017 |
Feedback | 2 | 23:58, 19 April 2017 |
Feedback | 1 | 22:46, 19 April 2017 |
Overall good.
I specifically agreed with the future work section, I do believe regular neural net should have a reasonable performance in our method since a CNN-style network did.
Since we test PCA with Neural Net, it's kind of regretful we don't test a method that is similar to it but use PCA with logistic regression, which could be seen as another future work because this is essentially a simplification of PCA with Neural Net. Such method is also comparable to MF with logistic regression because they can be seen using different dimension reduce tools.
Hi Bahare,
I like the structure of your page and I think that briefly explaining what each of the methods does is helpful too. One thing that caught my attention, is that our neural nets performance is very different, we must've used different parameters... but that happens so I think it's not something to worry about.
I agree with Moumita on that including some code snippets would be helpful.
In the results table: "MF and LR (k = 5) / rating", does "rating" mean "rating >= 4"?
Thanks, Alex
Hi Alex,
Thanks for your feedback. For the neural net I first reduced the dimension of the data-set (suing PCA) and then tried neural nets. That might be one reason why our results are different. My system crashed using all the data-set for building a neural net. Did you use all the data-set to build a neural net? Which library did you use for implementing the neural net?
I added some links to my page to referring to the codes.
"rating" means the actual ratings. I did not include these results in the paper.
Cheers,
Bahare
That indeed could be one reason... Also, I only used training dataset to build the network. As for the library, I used scikit's MLP classifier.
Alright, now I understand.
Thanks! :)
Alex
Hi Bahare,
I think you nicely covered all the information needed. Just want to add may be you could post your codes for each of the methods you tried in the same place where you described them.
Good Luck Moumita
Thanks Moumita!
I have difficulties posting my code in git-hub. But I will do that.