def get_dimensionality_reduction_data():
""" Get features for all models
This is used for dimensionality reduction, which is later used for
scatter plotting.
"""
# we don't have costs for all values, should do that.
data = []
experiment = 'Deterministic policy, regressed cost'
seed = 3
checkpoint = 25000
seeds = DataReader.find_option_values('seed', experiment)
for seed in seeds:
checkpoints = DataReader.find_option_values(
'checkpoint', experiment, seed)
for checkpoint in checkpoints:
try:
if data == []:
data = DimensionalityReduction.get_model_failing_features(
experiment, seed, checkpoint)
else:
data = np.concatenate([
data,
DimensionalityReduction.get_model_failing_features(
experiment, seed, checkpoint)
])
except Exception as e:
print(checkpoint, 'failed', e)
traceback.print_exc()
data = sklearn.preprocessing.scale(data)
return data
def seed_dropdown_change_callback(change):
if self.ignore_updates:
return
if change.name == 'value' and change.new is not None:
self.ignore_updates = True
self.checkpoint_dropdown.options = \
DataReader.find_option_values(
option='checkpoint',
experiment=self.experiment_dropdown.value,
seed=self.seed_dropdown.value
)
self.checkpoint_dropdown.value = None
self.checkpoint_dropdown.disabled = False
self.ignore_updates = False
def experiment_dropdown_change_callback(change):
if change.name == 'value' and change.new is not None:
self.ignore_updates = True
if level >= Picker.MODEL_LEVEL:
self.seed_dropdown.options = \
DataReader.find_option_values(
option='seed',
experiment=self.experiment_dropdown.value
)
self.seed_dropdown.value = None
self.seed_dropdown.disabled = False
self.checkpoint_dropdown.disabled = True
self.checkpoint_dropdown.value = None
self.ignore_updates = False
if level == Picker.EXPERIMENT_LEVEL:
self.call_callback(self.experiment_dropdown.value)
