ldssa / batch2-blu12

batch2-BLU12-instructors

What is in this BLU?

In this final BLU in the Recommender Systems (RS) specialization we learn taxonomy and strategies to extend the algorithms and techniques learned so far. We also learn about evaluation metrics and Surprise, for fast prototyping. Finally, we zoom in on building RS based on implicit feedback and unary data. Good luck!

How to use this repo

1. Install all the needed dependencies, specified in the requirements.txt file

• Install via pip

pip install -r requirements.txt

2. Go through the Learning Notebooks (they are in the Learning Notebooks/ folder)
3. Do the Exercise notebook, and submit it on the portal as usual

"I need help understanding something."

You can and should ask for help, be it about Learning Notebooks, Exercises, or anything else. Please check out the How to Ask for Help, and remember not to share code when asking for help about the exercises!

"I think I've found a bug."

This repo is entirely open source and is continuously improving over time. When you spot a mistake, check the issues. If it hasn't, please open an issue, explaining in details where it is (e.g., in what notebook, and on what line), and how to reproduce the error. If it is an easy fix, feel free to make a pull request.

batch2-blu12's Issues

Exercise 3: learning_rate parameter

I think there is a bug in the exercise description.
* With a learning rate of 0.0005 should be * With a learning rate of 0.00005 that is equal to 5e-05 as in the assert

LN3 1.2 Duplicate Entries

After converting the matrix from COO to CSR it is said:
The reduction from 500 to 460 stored element in the matrix is due to the consolidation.
We can confirm this by calling .max() on it.

Did not get the point of using max, in particular since calling max() on both COO and CSR the output is the same, what am I missing?

Exercise 3 baseline_cross_validate - assert problems

There are two problems in the asserts:

• when converting dict to list you cannot predict the order, so one should use sets; otherwise these asserts can only pass by chance:

assert(list(baseline.bsl_options.values()) == ['sgd', 5e-05, 0.05])
assert(list(baseline_results.keys()) == ['test_rmse', 'test_mae', 'fit_time', 'test_time'])

• it is asked a learning rate of 0.0005 but the assert contains 5e-05, instead of 5e-04 which is the equivalent.

(not sure how the grader handles this since I did not submit yet)

PS-Propose to change those 2 asserts to:

assert(set(baseline.bsl_options.values()) == {'sgd', 5e-04, 0.05})
assert(set(baseline_results.keys()) == {'test_rmse', 'test_mae', 'fit_time', 'test_time'})

Exercise Notebook 5

I am getting this error on the assert, i didn't see the reason.

Exercise 4 k-NN (with Baseline)

As in issue #2, this exercise has the same problems in the asserts:

• uses list instead of set
• parameter learning rate of 0.0005 is compared with 5.-e5 instead of 5.-e4

PS-I propose to replace those asserts with:

assert(set(baseline.bsl_options.values()) == {'sgd', 5e-04, 0.05})
assert(set(baseline_results.keys()) == {'test_rmse', 'test_mae', 'fit_time', 'test_time'})