Multiverso unable to learn on the Criteo Dataset

[jcarreira]

I can't seem to train a model using Multiverso on the Criteo dataset (L2-regularized Logistic regression).

I can train a model successfully on the same data using Vowpal Wabbit and some simple Python code. I have tried multiple configurations of Multiverso but without much luck.

My config file is:

input_size=14
output_size=1
objective_type=logistic
regular_type=L2
updater_type=sgd
train_epoch=20
sparse=false
use_ps=false
minibatch_size=20
#train_file=/mnt/efs/criteo_derivatives/day_1.csv_parsed
test_file=/mnt/efs/test_file
learning_rate_coef=0.0001
regular_coef=0.0007