coreylynch
can't save and load model correctly
I tried to save trained FM model and load it by using pickle. I could save and load it , but loaded model predicted anomalous values like all zero. How do I save and load trained model correctly ?
Hello, using the highest protocol available (-1) fixed the issue for me:
import cPickle as pickle
#Save the trained model to file
pickle.dump(model, open( "fittedmodel.pickle", "wb" ), -1) # protocol=-1 : using the highest protocol available
#Load the trained model from file
model= pickle.load(open( "fittedmodel.pickle", "rb" ))
print model.v
Hope this helps
same issue here. I can save it and load it, but the predict method returns all 0s.
I have the same issue.
joblib.dump(model, 'fm.pkl')
model = joblib.load('fm.pkl')
I ended up using fastFM and ultimately, when we went to production, we used ALS on pySpark. Another option is perhaps trying lightfm. Good luck!
As far as I can remember, both fastFm and lightFM saved the model with no problem. Regarding to fastFM I believe in their site they mentioned that is an academic project and not suited for production, but that might have changed have a look to their documenation. Regarding to lightFM, I believe the people at lyst used it (for me it did not work well). Good luck!
@jrzaurin thx very much.I will try to use lightFM. But I really don't kown why predict all 0s. fm<pyfm.pylibfm.FM instance at 0x31a5ef0> ('fm', array([[ 0.04463324, 0.16691201, -0.026428 , -0.09678349, 0.00591884, 0.151462 , 0.03368704, -0.03023687, -0.08386418], [-0.05177313, 0.02535079, -0.2484704 , -0.06514199, -0.17754692, 0.02342858, 0.12393999, -0.06503343, -0.02662149], [-0.05387198, 0.04207405, -0.00952863, -0.07698761, -0.01837245, 0.01768089, 0.07370562, 0.06938626, -0.08155303], [ 0.03876805, 0.0106537 , -0.03145769, -0.06997129, 0.00529883, 0.09341078, 0.14598721, -0.15788771, 0.09522664], [-0.06024594, 0.04614449, -0.00039591, 0.0777348 , -0.04931128, 0.06085309, 0.03142342, -0.07113038, 0.1005296 ], [ 0.0211259 , -0.08731488, 0.09841571, -0.00702032, 0.1778722 , -0.00426561, 0.00464802, 0.10008996, 0.13210467], [ 0.11710963, -0.07851948, 0.05037158, -0.22096869, 0.0221839 , -0.22044528, 0.06681716, -0.04526808, -0.14862625], [ 0.09367802, 0.03140169, 0.02186942, 0.06581881, 0.04394821, 0.02565542, 0.02248762, 0.14576877, -0.1542256 ], [-0.14858283, -0.04868577, 0.0234426 , -0.00569702, 0.15142498, -0.13743753, -0.04227515, -0.00504726, 0.07348803], [-0.07020803, -0.03101657, -0.1198239 , 0.08044487, 0.01856105, -0.28774227, 0.02996937, -0.0562063 , -0.06951357]])) model: ('model', array([[ 0.04463324, 0.16691201, -0.026428 , -0.09678349, 0.00591884, 0.151462 , 0.03368704, -0.03023687, -0.08386418], [-0.05177313, 0.02535079, -0.2484704 , -0.06514199, -0.17754692, 0.02342858, 0.12393999, -0.06503343, -0.02662149], [-0.05387198, 0.04207405, -0.00952863, -0.07698761, -0.01837245, 0.01768089, 0.07370562, 0.06938626, -0.08155303], [ 0.03876805, 0.0106537 , -0.03145769, -0.06997129, 0.00529883, 0.09341078, 0.14598721, -0.15788771, 0.09522664], [-0.06024594, 0.04614449, -0.00039591, 0.0777348 , -0.04931128, 0.06085309, 0.03142342, -0.07113038, 0.1005296 ], [ 0.0211259 , -0.08731488, 0.09841571, -0.00702032, 0.1778722 , -0.00426561, 0.00464802, 0.10008996, 0.13210467], [ 0.11710963, -0.07851948, 0.05037158, -0.22096869, 0.0221839 , -0.22044528, 0.06681716, -0.04526808, -0.14862625], [ 0.09367802, 0.03140169, 0.02186942, 0.06581881, 0.04394821, 0.02565542, 0.02248762, 0.14576877, -0.1542256 ], [-0.14858283, -0.04868577, 0.0234426 , -0.00569702, 0.15142498, -0.13743753, -0.04227515, -0.00504726, 0.07348803], [-0.07020803, -0.03101657, -0.1198239 , 0.08044487, 0.01856105, -0.28774227, 0.02996937, -0.0562063 , -0.06951357]])) model<pyfm.pylibfm.FM instance at 0x31aac68> predict:[ 5.],model:[ 0.]
