def test_trainer(self):
# generate random data
num_entries = 50
num_features = 10
# seed random number generator for consistency
np.random.seed(1238927)
x = np.random.rand(num_entries, num_features)
y = (30 * x[:, 0]) - (10 * x[:, 2]) + np.random.rand(1, num_entries)
# run forward
t = Trainer(x, y)
# run forward selection
t.run_forward_selection([1], None)
# initial column should still be in there when training
self.assertIn(1, t.column_indices)
# should have at least one more column than initial data
self.assertGreater(len(t.column_indices), 1)
# run forward selection
t.run_forward_selection()
self.assertIn(1, t.column_indices)
# run backward selection
t.run_backward_selection(range(0, num_features), None)
# column 0 should still be in there with 99.9% likelihood
self.assertIn(0, t.column_indices)
# should have at least one more column than initial data
self.assertLess(len(t.column_indices), num_features)
import numpy as np
from lintrain import Trainer
# must be placed outside the package and reference updated to use the parallel trainer
# due to limitation related to relative references
# from parallel.trainer import Trainer
if "__main__" == __name__:
# generate random data
num_entries = 50
num_features = 10
x = np.random.rand(num_entries, num_features)
y = (30 * x[:, 0]) - (10 * x[:, 2]) + np.random.rand(1, num_entries)
# create trainer
t = Trainer(x, y)
t.debug = 2
# run
#t.run_forward_selection()
#t.run_backward_selection()
t.run_bidirectional_selection([1, 3])
# print output
print "COLUMN COEFFICIENTS"
print t.fit
print ""
print "COLUMNS USED"
print t.column_indices
print ""
from lintrain import Trainer
from lintrain.solvers import LogisticRegression
# use parallel trainer instead
# from lintrain import ParallelTrainer as Trainer
if "__main__" == __name__:
# generate random data
num_entries = 500
num_features = 10
x = np.random.rand(num_entries, num_features)
y = (5 * x[:, 0]) + (2 * x[:, 2]) + np.random.rand(1, num_entries)
y = 1 * (y > 3.5)
# create trainer
t = Trainer(x, y, solver=LogisticRegression)
t.debug = 2
# run
t.run_forward_selection()
#t.run_backward_selection()
#t.run_bidirectional_selection([1, 3])
# print output
print "COLUMN COEFFICIENTS"
print t.fit
print ""
print "COLUMNS USED"
print t.column_indices
print ""
import numpy as np
from lintrain import Trainer
from lintrain.solvers import RidgeRegression
# use parallel trainer instead
# from lintrain import ParallelTrainer as Trainer
if "__main__" == __name__:
# generate random data
num_entries = 50
num_features = 10
x = np.random.rand(num_entries, num_features)
y = (30 * x[:, 0]) - (10 * x[:, 2]) + np.random.rand(1, num_entries)
# create trainer
t = Trainer(x, y, solver=RidgeRegression)
t.debug = 2
# run
#t.run_forward_selection()
#t.run_backward_selection()
t.run_bidirectional_selection([1, 3])
# print output
print "COLUMN COEFFICIENTS"
print t.fit
print ""
print "COLUMNS USED"
print t.column_indices
print ""
