CODEWITHSUNDEEP
pythonmachine-learningcollaborative-filteringrecommendation-engine
Olivier D.
Olivier D.

Reputation: 234

LightFM recommendation: Inconsistent error with interaction data

I have the following basic code with the LightFM recommendation module:

# Interactions
A=[0,1,2,3,4,4] # users
B=[0,0,1,2,2,3] # items
C=[1,1,1,1,1,1] # weights
matrix = sparse.coo_matrix((C,(A,B)),shape=(max(A)+1,max(B)+1))
# Create model
model = LightFM(loss='warp')
# Train model
model.fit(matrix, epochs=30)
# Predict
scores = model.predict(1, np.array([0,1,2,3]))
print(scores)

This returns the following error:

> C:\Program
> Files\Python\Python36\lib\site-packages\numpy\core\_methods.py:32:
> RuntimeWarning: invalid value encountered in reduce   return
> umr_sum(a, axis, dtype, out, keepdims) Traceback (most recent call
> last):   File "run.py", line 15, in <module>
>     model.fit(matrix, epochs=100)   File "C:\Program Files\Python\Python36\lib\site-packages\lightfm\lightfm.py", line 476,
> in fit
>     verbose=verbose)   File "C:\Program Files\Python\Python36\lib\site-packages\lightfm\lightfm.py", line 580,
> in fit_partial
>     self._check_finite()   File "C:\Program Files\Python\Python36\lib\site-packages\lightfm\lightfm.py", line 410,
> in _check_finite
>     raise ValueError("Not all estimated parameters are finite," ValueError: Not all estimated parameters are finite, your model may
> have diverged. Try decreasing the learning rate or normalising feature
> values and sample weights

Strangely enough, making some changes in the interaction data makes it work, as with:

# Interactions
A=[0,1,2,3,4,4]
B=[0,0,1,2,2,10] # notice the 10 here
C=[1,1,1,1,1,1]

Could anyone help me with that please?

Upvotes: 1

Views: 1214

Answers (1)

Alexandra Lorenzo
Alexandra Lorenzo

Reputation: 51

#Predict
scores = model.predict(1, np.array([0,1,2,3]))
print(scores)

[-0.17697991 -0.55117112 -0.37800685 -0.57664376]

It works fine for me, update the lightFM version?

Upvotes: 1

Related Questions