pyunit_pubdev_5493_glrm_seed.py

from __future__ import print_function
import sys

sys.path.insert(1, "../../../")
import h2o
from tests import pyunit_utils
from h2o.estimators.glrm import H2OGeneralizedLowRankEstimator

'''
PUBDEV-5493: GLRM return different results regardless of seed setting.
'''

def test_glrm_seeds():
  print("Importing iris_wheader.csv data...")
  irisH2O = h2o.upload_file(pyunit_utils.locate("smalldata/iris/iris_wheader.csv"))
  irisH2O.describe()
  initMethods = ["random", "svd", "plus_plus", "user"] # user mode without init values is equivalent to randomized
  seeds = [123456789, 987654321]


  for initM in initMethods:
    # first two models are trained with same seed and should be the same
    glrm_h2o_seed0 = setupTrainModel(initM, seeds[0])
    predict_seed0 = predGLRM(irisH2O, glrm_h2o_seed0)

    glrm_h2o_seed0Same = setupTrainModel(initM, seeds[0])
    predict_seed0Same = predGLRM(irisH2O, glrm_h2o_seed0Same)

    # trained with same seed, reconstructed datasets should be the same
    pyunit_utils.compare_frames_local(predict_seed0[0:4], predict_seed0Same[0:4],
                                      prob=1.0)  # compare and make sure reconstructed frames are the same

    # trained with different seed, reconstructed datasets should be different
    glrm_h2o_seed1 = setupTrainModel(initM, seeds[1])
    predict_seed1 = predGLRM(irisH2O, glrm_h2o_seed1)
    assert not (pyunit_utils.compare_frames_local(predict_seed0[0:4], predict_seed1[0:4], prob=1.0, returnResult=True)), \
      "GLRM return same results with different random seed."


def setupTrainModel(initM, seed):
  rank = 3
  gx = 0.25
  gy = 0.25
  trans = "STANDARDIZE"

  return H2OGeneralizedLowRankEstimator(k=rank, loss="Quadratic", gamma_x=gx, gamma_y=gy, transform=trans,
                                                init=initM, seed=seed)

def predGLRM(dataset, model):
  '''
  Simple method to train GLRM model and return prediction result.

  :param dataset: dataset to be scored and trained on
  :param model: glrm model to be trained
  :return: reconstructed dataframe.
  '''
  model.train(x=dataset.names, training_frame=dataset)
  return model.predict(dataset)

if __name__ == "__main__":
  pyunit_utils.standalone_test(test_glrm_seeds)
else:
  test_glrm_seeds()