evfro / polara Goto Github PK

data_model = ItemColdStartData(
training_data,
*training_data.columns, # userid, itemid
item_features=content_feature_df,
seed=seed)

print(data_model)

HERE IM GETTING ERROR: NotImplementedError: Data has duplicate values

My dataframe has multiple entries for a user. cant drop them. any help here

Referring this blogs code https://www.eigentheories.com/blog/lightfm-vs-hybridsvd/
The class ItemColdStartData has attribute item_features.
which is removed in the current version of the package.

Is their another way to add item features in the data model or the code behaves the same without it.

I just found that there no package on Conda and Pip repository. It would be great to put it into both registries :)

I am currently working with an implicit feedback dataset and trying to evaluate a model based on the precision@K measure. I am having difficulties understanding the implemented precision calculation in a case when the list of recommendations consists only of top K recommendations.

@evfro Can you please clarify if the calculated precision is a precision@K or a regular precision measure?

Hi,
While running data_model.prepare(), There is error on following line.
features_melted = self.item_features.agg(lambda x: [f for l in x for f in l], axis=1)

Error is TypeError: 'int' object is not iterable

However I tried converting every element in the item_tag dataframe in str then into np array, But doesn't seem to work.

I tried to run https://github.com/evfro/polara/blob/master/examples/Comparing%20LightFM%20with%20HybridSVD.ipynb, had
to change the variable catch=None to catch=(RuntimeWarning,). It doesn't run if you don't make this change.

Hi! I clone a git repository with polara and run setup.py in appropriate folder. But then trying to import it got a ModuleNotFoundError: No module named 'polara.preprocessing'.
Is there some hints to use it or it's a bug?

Hi @evfro , I love this great Polara framework!

However, when I run ScaledSVD notebook (https://github.com/evfro/recsys19_hybridsvd/blob/master/ScaledSVD.ipynb), I got this issue:

---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
<ipython-input-1-3b2c68df4112> in <module>
      8                                          set_config)
      9 
---> 10 from scaledsvd import ScaledSVD, ScaledSVDItemColdStart
     11 from data_preprocessing import (get_movielens_data,
     12                                 get_bookcrossing_data,

~/bukalapak/recsys19_hybridsvd/scaledsvd.py in <module>
      3 from scipy.sparse.linalg import norm as spnorm
      4 from polara import SVDModel
----> 5 from polara.recommender.coldstart.models import SVDModelItemColdStart
      6 
      7 

~/miniconda3/envs/erwin-svd/lib/python3.6/site-packages/polara-0.6.5.dev0-py3.6.egg/polara/recommender/coldstart/models.py in <module>
      5 from polara import SVDModel
      6 from polara.recommender.models import RecommenderModel, ScaledMatrixMixin
----> 7 from polara.recommender.hybrid.models import LCEModel, HybridSVD
      8 from polara.recommender.external.lightfm.lightfmwrapper import LightFMWrapper
      9 from polara.lib.similarity import stack_features

ModuleNotFoundError: No module named 'polara.recommender.hybrid'

How to solve the issue? Thanks!

Hi @evfro ,

I want to get recommendation based on user's preference, not user_id. For example, there are 10 items in training data, and (new) user's preference is [0,1,0,0,0,0,1,1,0,0]. Theoretically, we just multiply the vector and item matrix (p . V . Vt). How can I achieve that?

Not able to install using python 3.7.7 venv with following requirement versions.
jupyter>=1.0.0
numba>=0.21.0
numpy>=1.10.1
matplotlib>=1.4.3
pandas>=0.17.1
requests>=2.7.0
scipy>=0.16.0

According to this code, LightFM features are normalized by default

I have faced with the issue that normalization decrease model precision with the same epochs counts. The same problem already reported in LightFM repository:
lyst/lightfm#413

As I understood, there no need to normalize one-hot encoded features (e.g., category).

Hi @evfro ,

In this blog post https://www.eigentheories.com/blog/lightfm-vs-hybridsvd/, it is mentioned that SVD from polara has deterministic output but each run in my pipeline is giving me different outputs. I am using hybrid svd and I am being careful with every seed and random_state instantiation but still the issue is persisting. How can I achieve the same output for different runs?

The code is not compatible with Python 3. The usage of "Print" function is obsolete.

ModuleNotFoundError Traceback (most recent call last)
Input In [50], in
----> 1 from polara.recommender.coldstart.models import LightFMItemColdStart

ModuleNotFoundError: No module named 'polara'

Even after doing pip install polara facing the issue.