ExaScience / smurff

Bayesian Factorization with Side Information in C++ with Python wrapper
MIT License
70 stars 14 forks source link

IPython notebook crash #150

Closed tvandera closed 4 months ago

tvandera commented 5 months ago

The https://github.com/ExaScience/smurff/blob/master/docs/notebooks/different_noise_models.ipynb crashes in this test:

RMSE = 0.91
INFO:root:PythonSession {
  Data: {
    Type: ScarceMatrixData [with NAs]
    Component-wise mean: 6.35508
    Component-wise variance: 1.89449
    Noise: Probit Noise with threshold 6
    Size: 47424 [15073 x 346] (0.91%)
      Warning: 77 empty rows
      Warning: 1 empty cols
  }
  Model: {
    Num-latents: 32
    Dimensions: 15073,346
  }
  Priors: {
    0: NormalPrior
    1: NormalPrior
  }
  Result: {
    Test data: 11856 [15073 x 346] (0.23%)
    Binary classification threshold: 6.00
      54.92% positives in test data
  }
  Config: {
      Iterations: 100 burnin + 100 samples
      Save model: never
  }
}
tvandera commented 4 months ago

Seems to be the next one, with Probit + Macau. Goes out of memory in init phase

tvandera commented 4 months ago

Fixed in dca46080412a3af496d4b2caeee558b7b0abb3d2, merged in #151