search

aksnzhy / xlearn

High performance, easy-to-use, and scalable machine learning (ML) package, including linear model (LR), factorization machines (FM), and field-aware factorization machines (FFM) for Python and CLI interface.
https://xlearn-doc.readthedocs.io/en/latest/index.html
Apache License 2.0
3.09k stars 519 forks source link

log_loss is nan when using fm model #316

Open sjoke opened 4 years ago

sjoke commented 4 years ago

// train_x shape: (51960, 7193), train_y shape: (51960,) // test_x shape: (22269, 7193), test_y shape: (22269,) xdm_train = xl.DMatrix(train_x.toarray(), train_y)
xdm_test = xl.DMatrix(test_x.toarray(), test_y) fm_model = xl.create_fm()
fm_model.setTrain(xdm_train) fm_model.setValidate(xdm_test) xparams = { 'task': 'binary', 'metric': 'auc', 'lr': 0.02, 'lambda': 0.0002, 'epoch': 5,

'opt': 'FTRL',

'nthread': 1

} fm_model.fit(xparams, 'xmodel.out')

result: [------------] Epoch Train log_loss Test log_loss Test AUC Time cost (sec) [ 10% ] 1 nan nan 0.500000 11.07 [ 20% ] 2 nan nan 0.500000 11.18 [ 30% ] 3 nan nan 0.500000 11.50 [ 40% ] 4 nan nan 0.500000 10.99 [ 50% ] 5 nan nan 0.500000 10.72 [ 60% ] 6 nan nan 0.500000 11.90 [ 70% ] 7 nan nan 0.500000 11.87 [ 80% ] 8 nan nan 0.500000 11.12 [ 90% ] 9 nan nan 0.500000 11.02 [ 100% ] 10 nan nan 0.500000 11.16

when using linear model, the result is fine. so what's the problem? wish your help, :)

naihaishy commented 4 years ago

reduce lr, maybe learning rate is too high

Murray132 commented 4 years ago

Maybe your train data has a zero row.

kyle-pjk commented 3 years ago

Have you slove the problem? I had the same problem. Can you help me? thanks.