How can i use another metric similarity in KNN about surprise HOT 12 CLOSED

Hi,

What have you tried and what does not work?

from surprise.

Error: Wrong sim name euclidean. Allowed values are msd, pearson_baseline, cosine, pearson.

I already add euclidean sim in algo.base.py

construction_func = {'cosine': sims.cosine,
'msd': sims.msd,
'pearson': sims.pearson,
'pearson_baseline': sims.pearson_baseline
'euclidean':sims.euclidean}

but still doesnt work

from surprise.

There's a missing comma in your dict but I suppose your code is correct, else you would have gotten an error there?

You seem to have done it correctly, so my guess would be that you are not using your custom version of surprise but rather the one that you have installed from PyPI? Maybe try a dirty print('BLABLA') in your code and see if it's printed?

Also, have you installed it locally in dev mode with pip install -e .? The -e is very important. In addition, as you have modified a Cython file, you will need to reinstall the package at each change.

from surprise.

Ok thanks, i have recorrect the coma. But it still doesnt work. I create euclidean metric is base in surprise lib. I only change the calculation
Cosine:
denum = np.sqrt(sqi[xi, xj] * sqj[xi, xj])
sim[xi, xj] = prods[xi, xj] / denum

sim[xj, xi] = sim[xi, xj]

To:
sim[xi, xj] = (proda[xi, xj] - prods[xi, xj])**2
sim[xi, xj] = np.sqrt(sim[xi, xj])

sim[xj, xi] = sim[xi, xj]

which proda[] already defined above

from surprise.

Are you sure you are running the code that you have modified?

from surprise.

i need to install cython to run them?
i run it again with python give me an error in " cdef "

from surprise.

im really sorry, i just unfamiliar with Linux

from surprise.

No problem.

Yes you will need Cython and a C compiler.

Cython will compile the .pyx files into C code, and the C compiler will compile C code into an external extension / module.

You may be interesting in checking the contribution guidelines (you can forget coding style guidelines if you're not planning on a pull request)

from surprise.

Now i got the error.
similarities.c:4:20: fatal error: Python.h: No such file or directory

from surprise.

Try Googling this, it is a known issue (also it is not related to Surprise).

from surprise.

i already compile it from pyx/pyc and c. for similarities and algo.base. But the error still popping around.
There is no other way to create other similarity metric for KNN?

from surprise.

Well another way of doing this would be to extent the k-NN class and to redefine your own train() method, where you would compute the similarities here.

This is dirty but it would work. Note though that you would note be able to use the classic workflow for defining the similarity options.

from surprise.