def create_single_dataset(database, path, adj_obj):
"""
Creates a pickle file dataset for each motions for all objects with a particular adjective
"""
adjective = adj_obj.adjective
#import pdb; pdb.set_trace()
# Test to see if any phase files already exist and only create the ones that don't, or skip all
all_phase_list = phases
phase_list = []
nobuild = 0
for phase in all_phase_list:
phase_obj = PhaseClass()
phase_obj.phase = phase;
dataset_file_name = "_".join(("hmm_feature", adjective,phase))+".pkl"
phase_obj.path_name = os.path.join(path, dataset_file_name)
if os.path.exists(phase_obj.path_name):
print "File %s already exists, skipping it." % phase_obj.path_name
phase_obj.build = False
nobuild = nobuild + 1
else:
phase_obj.build = True
print "Creating adjective %s and phase %s" % (adjective, phase)
phase_list.append(phase_obj)
if nobuild == 4:
print "All phases of adjective %s are already built. Moving on..." % adjective
return
# Open database and get train/test split
database = tables.openFile(database)
train_objs, test_objs = get_train_test_objects(database, adjective)
# Select the features from the feature objects
feature_train_dict_list = create_hmm_feature_set(database, train_objs, adj_obj, phase_list)
feature_test_dict_list = create_hmm_feature_set(database, test_objs, adj_obj, phase_list)
for i, phase_object in enumerate(phase_list):
if phase_object.build == True:
# Store the train/test in a dataset
#import pdb; pdb.set_trace()
dataset = defaultdict(dict)
dataset['train'] = feature_train_dict_list[i]
dataset['test'] = feature_test_dict_list[i]
if len(dataset) is 0:
print "Empty dataset for adj %s and phase %s" % (adjective, phase_object.phase)
continue
print "Saving dataset to file"
#import pdb; pdb.set_trace()
# Save the results in the folder
#Saves one file per motion. This needs to be a for loop
with open(phase_object.path_name, "w") as f:
print "Saving file: ", phase_object.path_name
cPickle.dump(dataset, f, protocol=cPickle.HIGHEST_PROTOCOL)
import pdb
pdb.set_trace()
# Initialize scores
f1s = 0
precs = 0
recalls = 0
total = 0
# Setup text file to store values to
adjective_report = open("adjective_score_report.txt", "w")
adjective_report.write("Adjective, precision, recall, f1\n")
for classifier in classifiers:
try:
# Pull out the objects that we want
train_objs, test_objs = get_train_test_objects(
database, classifier.adjective)
# Compute score for each adjective
p, r, f1 = test_adjective(classifier, database, test_objs,
adjective_report)
precs += p
recalls += r
f1s += f1
total += 1
except ValueError:
print "Skipping values"
continue
adjective_report.close()
def create_single_dataset(database, path, adj_obj):
"""
Creates a pickle file dataset for each motions for all objects with a particular adjective
"""
adjective = adj_obj.adjective
#import pdb; pdb.set_trace()
# Test to see if any phase files already exist and only create the ones that don't, or skip all
all_phase_list = phases
phase_list = []
nobuild = 0
for phase in all_phase_list:
phase_obj = PhaseClass()
phase_obj.phase = phase
dataset_file_name = "_".join(
("hmm_feature", adjective, phase)) + ".pkl"
phase_obj.path_name = os.path.join(path, dataset_file_name)
if os.path.exists(phase_obj.path_name):
print "File %s already exists, skipping it." % phase_obj.path_name
phase_obj.build = False
nobuild = nobuild + 1
else:
phase_obj.build = True
print "Creating adjective %s and phase %s" % (adjective, phase)
phase_list.append(phase_obj)
if nobuild == 4:
print "All phases of adjective %s are already built. Moving on..." % adjective
return
# Open database and get train/test split
database = tables.openFile(database)
train_objs, test_objs = get_train_test_objects(database, adjective)
# Select the features from the feature objects
feature_train_dict_list = create_hmm_feature_set(database, train_objs,
adj_obj, phase_list)
feature_test_dict_list = create_hmm_feature_set(database, test_objs,
adj_obj, phase_list)
for i, phase_object in enumerate(phase_list):
if phase_object.build == True:
# Store the train/test in a dataset
#import pdb; pdb.set_trace()
dataset = defaultdict(dict)
dataset['train'] = feature_train_dict_list[i]
dataset['test'] = feature_test_dict_list[i]
if len(dataset) is 0:
print "Empty dataset for adj %s and phase %s" % (
adjective, phase_object.phase)
continue
print "Saving dataset to file"
#import pdb; pdb.set_trace()
# Save the results in the folder
#Saves one file per motion. This needs to be a for loop
with open(phase_object.path_name, "w") as f:
print "Saving file: ", phase_object.path_name
cPickle.dump(dataset, f, protocol=cPickle.HIGHEST_PROTOCOL)
import pdb; pdb.set_trace()
# Initialize scores
f1s= 0
precs = 0
recalls = 0
total = 0
# Setup text file to store values to
adjective_report = open("adjective_score_report.txt", "w")
adjective_report.write("Adjective, precision, recall, f1\n")
for classifier in classifiers:
try:
# Pull out the objects that we want
train_objs, test_objs = get_train_test_objects(database, classifier.adjective)
# Compute score for each adjective
p, r, f1 = test_adjective(classifier, database, test_objs, adjective_report)
precs += p
recalls += r
f1s += f1
total += 1
except ValueError:
print "Skipping values"
continue
adjective_report.close()
print "Average f1s: ", f1s / total