Well, I mean, how many bugs have you implemented till now? I want to checkout.

[Heermosi]

No need for exactly the locations, I want numbers for each directory.

[brylie]

@Heermosi have you found a specific bug to report?

[Heermosi]

Oh, yeah, our team worked on this branch for weeks to make it a working demo. And we discovered a lot of bugs, especially this one from [evaluation_runner_test.py]:

        # figure out what to do with result ;)
        self.assertLess(result['mae'], decimal.Decimal(1.7))
        self.assertLess(result['pak'], decimal.Decimal(0.7))
        self.assertLess(result['rak'], decimal.Decimal(0.7))

I'm almost certain you manually injected such "bugs",.... It's alright, but did you have a counter on how many such bugs exist?

As you requested, I'll list part below:

1. templates/moviegeek/index.html & templates/moviegeek/detail.html

                           a.setAttribute('onclick', "add_impression({{user_id}}, 'more_details', " +
                                                     movie_id + ", '{{ session_id }}','{{ csrf_token }}')")

   This would collect incorrect movie id (stripped out of prev 0s), however would not be a bug if you turn column movie_id to type int, then it was yet another bug

2. moviegeeks/views.py

   
   def genre(request, genre_id):
   
   if genre_id:
       selected = Genre.objects.filter(name=genre_id)[0]
       movies = selected.movies.all().order_by('-year')
   else:
       movies = Movie.objects.all().order_by('-year')

This would cause duplicated items browsing specific genre when the data set have some null in year column. Yes some null in populated dataset.(and some 0 star explicit scoring ......????)

3. The populated data set have a lot of incorrect movie ids. 
My team member is fixing on this...
I'm sure this can only be manually injected.

4. moviegeeks/views.py line 119

context_dict = {
    'genres': genres,
    'movies': mov.values(),
    'api_key': api_key,
}

The search page cannot display the recommend items correctly because of lacking some data, (userid)

5. builder/matrix_factorization_calculator.py

def predict(self, user, item):

    pq = np.dot(self.item_factors[item], self.user_factors[user].T)
    b_ui = self.all_movies_mean + self.user_bias[user] + self.item_bias[item]
    prediction = b_ui + pq

    if prediction > 10:
        prediction = 10
    elif prediction < 1:
        prediction = 10
    return prediction

Hmmm, what kind of bug is it..... a loop in prediction???

This is only part of bugs detected till now. I don't want to make this post toooooo long.

[kimfalk]

Thank you for looking at the code. Im sure the code is not perfect, so it is great that you are using the code and helping to find bug. I would recommend making a branch of the repo and commit your fixes to the code. Which can then be merged with a PR.

Having your fixes in a branch would also make it easier to understand where you believe the bugs are. I'm not sure I understand all of your message above.

[Heermosi]

Well, if you really want this , I'll make a full list of known bugs fixed... I originally thought this is a challenge to green hands.

As agreed I'll put the rest of the list of bugs...

6. <builder/item_similarity_calculator.py> line 106 && <builder/lda_model_calculator.py> line 195

        cur.execute('truncate table similarity')

This would cause sequence number out of range after rounds of model building, use "restart identity"

7. <builder/lda_model_calculator.py> line 3

import tqdm where in line 94

for d in tqdm(data): In fact it was using tqdm.tqdm to perform actions.

8. <builder/lda_model_calculator.py> line 114-117

        index = similarities.MatrixSimilarity(corpus)

        self.save_lda_model(lda_model, corpus, dictionary, index)
        self.save_similarities(index, docs)

This means the lda model is not really in use.... I don't know it was a bug or not... calculate using the word vector is intended? Well, at least the coo_matrix conversion is not recommended because it would use up to 64 GB ram. You may always fetch a row at a time. And to the result generation: the result table is much too large, it could reach up to 130GB when used lda calculation... cut it or ... wait for 6-12 hours.

9. recs/fwls_recommender.py Lacks an auto model loading process by default, the model is not loaded and only triggered by set_save_path This is only called in evaluator To make it available in web pages(analytics) it should be loaded automatically

[kimfalk]

You do realize its code used for as demo code for a book right?

[undead747]

You do realize its code used for as demo code for a book right?

Sorry sir but I'm very worried, is it true that I need 130GB to save IDA results ?

[kimfalk]

Sorry sir but I'm very worried, is it true that I need 130GB to save IDA results? No, you should not.

Oh, yeah, our team worked on this branch for weeks to make it a working demo. And we discovered a lot of bugs, especially this one from [evaluation_runner_test.py]:

        # figure out what to do with result ;)
        self.assertLess(result['mae'], decimal.Decimal(1.7))
        self.assertLess(result['pak'], decimal.Decimal(0.7))
        self.assertLess(result['rak'], decimal.Decimal(0.7))

I'm almost certain you manually injected such "bugs",....

It's a test case to test everything is working. So I create a small dataset run it through the system and verify that the results look good.

[faisalnazik]

It is not unusual to have bugs. but of these types ... ?? btw thanx for the code.

[brylie]

I'm not sure this issue is productive. Should it be closed?