def testMetricHistoriesFiles(self):
logdir = tf.test.get_temp_dir()
tf.gfile.MkDir(os.path.join(logdir, 'job1'))
tf.gfile.MkDir(os.path.join(logdir, 'job2'))
p = early_stop.MetricHistory.Params().Set(logdir=logdir)
mh1 = early_stop.MetricHistory(
p.Set(jobname='job1', metric='m1', local_filesystem=True))
mh2 = early_stop.MetricHistory(
p.Set(jobname='job2', metric='m2', local_filesystem=True))
early_stop.MetricHistory.ConditionalAppend('job1', 'm1', 1, 10.0)
early_stop.MetricHistory.ConditionalAppend('job1', 'm2', 1, 10.0)
early_stop.MetricHistory.ConditionalAppend('job2', 'm2', 1, 10.0)
early_stop.MetricHistory.ConditionalAppend('job1', 'm1', 2, 5.0)
self.assertTrue(tf.gfile.Exists(mh1.hist_file))
self.assertTrue(tf.gfile.Exists(mh2.hist_file))
with tf.gfile.GFile(mh1.hist_file) as f:
lines = f.readlines()
self.assertEqual(len(lines), 2)
self.assertEqual(lines[0].rstrip(), '1 10.000000')
self.assertEqual(lines[1].rstrip(), '2 5.000000')
with tf.gfile.GFile(mh2.hist_file) as f:
lines = f.readlines()
self.assertEqual(len(lines), 1)
self.assertEqual(lines[0].rstrip(), '1 10.000000')
def testMetricHistoriesMapUniqueness(self):
# pylint: disable=unused-variable
p = early_stop.MetricHistory.Params()
mh1 = early_stop.MetricHistory(p.Set(jobname='job1', metric='m1'))
mh2 = early_stop.MetricHistory(p.Set(jobname='job2', metric='m2'))
mh3 = early_stop.MetricHistory(p.Set(jobname='job1', metric='m1'))
m = early_stop.MetricHistory._metric_histories_map
self.assertEqual(len(m), 2)
self.assertEqual(m[early_stop.MetricHistory._Key('job1', 'm1')], mh3)
self.assertEqual(m[early_stop.MetricHistory._Key('job2', 'm2')], mh2)
def __init__(self, params):
super(AdaptiveScheduler, self).__init__(params)
if len(self.params.tasks) != 2 or len(self.params.expected) != 2:
raise ValueError('Only two tasks are supported by this scheduler.')
if self.params.epsilon < 0:
raise ValueError('Epsilon should be positive.')
self.tasks = self.params.tasks
self.last_scores = [0.0] * 2
self._metric_histories = [
early_stop.MetricHistory(self.params.mh_a),
early_stop.MetricHistory(self.params.mh_b)
]
def __init__(self, params):
super().__init__(params)
p = self.params
with tf.variable_scope(p.name):
wp = py_utils.WeightParams(
shape=[],
init=py_utils.WeightInit.Constant(1.0),
collections=['DevBasedSchedule_vars'],
dtype=tf.float32)
self._cur_factor = py_utils.CreateVariable(
'cur_factor', wp, trainable=False)
wp = py_utils.WeightParams(
shape=[],
init=py_utils.WeightInit.Constant(0),
collections=['DevBasedSchedule_vars'],
dtype=tf.int64)
self._ref_step = py_utils.CreateVariable('ref_step', wp, trainable=False)
self._metric_history = early_stop.MetricHistory(p.metric_history)
self._best_step = ops.best_step(self._metric_history.hist_file,
p.tolerance)
def __init__(self, params):
super().__init__(params)
p = self.params
self._metric_history = early_stop.MetricHistory(p.metric_history)
def __init__(self, params): super().__init__(params) self.SetVariableFree(False) p = self.params self._metric_history = early_stop.MetricHistory(p.metric_history)