def one_classification_prediction(directory, log_file, user_id, class_counts,
verbose):
data_directory = bias_util.data_directory
data_file_name = bias_util.data_file_name
all_logs, attr_logs, item_logs, help_logs, cat_logs = bias_util.recreate_logs(
directory, log_file)
dataset, attr_map = bias_util.read_data(data_directory, data_file_name)
classification, decisions_labels, decisions_cat, decisions_help = bias_util.get_classifications_and_decisions(
all_logs, dataset)
X = []
Y = []
# decisions_labels of the form (index, user classification, actual classification, player id)
prev = -1
cur = -1
for i in range(0, len(decisions_labels)):
prev = cur
cur = decisions_labels[i][1]
if (not cur in class_counts):
class_counts[cur] = 1
else:
class_counts[cur] += 1
if (prev != -1 and
cur != -1): # and prev != cur): # create a training instance
# comment prev != cur condition to allow repetitions
X.append([bias_util.pos_to_num_map[prev]])
Y.append(bias_util.pos_to_num_map[cur])
return X, Y, class_counts
def __init__(self, directory, in_file_name, out_file_name, data_directory, data_file_name):
self.directory = directory
self.in_file_name = in_file_name
self.out_file_name = out_file_name
self.data_directory = data_directory
self.data_file_name = data_file_name
self.dpc_logs = []
self.dpd_logs = []
self.ac_logs = []
self.ad_logs = []
self.awc_logs = []
self.awd_logs = []
self.dataset, self.attr_value_map = bias_util.read_data(data_directory, data_file_name)
self.all_logs, self.attr_logs, self.item_logs, self.help_logs, self.cat_logs = bias_util.recreate_logs(directory, in_file_name)
def write_svm_results(directory, file_name, log_file, to_plot, fig_num,
verbose):
if (verbose):
print 'Writing and Plotting SVM Data: ', file_name
data_directory = bias_util.data_directory
data_file_name = bias_util.data_file_name
all_logs, attr_logs, item_logs, help_logs, cat_logs = bias_util.recreate_logs(
directory, log_file)
logs = item_logs
dataset, attr_map = bias_util.read_data(data_directory, data_file_name)
classification, decisions_labels, decisions_cat, decisions_help = bias_util.get_classifications_and_decisions(
logs, dataset)
all_data = dict()
all_data['classifications'] = classification
x_data = []
y_data = []
data = []
features = bias_util.get_bball_player(
dataset,
list(classification.keys())[0]).get_map().keys()
features.remove('Name')
features = sorted(features)
for key in classification.keys():
cur_player = bias_util.get_bball_player(dataset, key)
cur_map = cur_player.get_map()
cur_map['*Classification'] = classification[key]
data.append(cur_map)
cur_x = []
for i in range(0, len(features)):
cur_x.append(cur_map[features[i]])
cur_x = [float(x) for x in cur_x]
x_data.append(cur_x)
y_data.append(bias_util.pos_to_num_map[classification[key]])
svm_weights, svm_classes = get_svm_weights(x_data, y_data)
weights_map = dict()
i = 0
for j in range(0, len(svm_classes)):
for k in range(j + 1, len(svm_classes)):
key = bias_util.num_to_pos_map[
j] + ' - ' + bias_util.num_to_pos_map[k]
value = svm_weights[i]
weights_map[key] = value
i += 1
all_data['features'] = features
all_data['weights'] = weights_map
all_data['classifications'] = data
if not os.path.exists(directory):
os.makedirs(directory)
f_out = open(directory + file_name, 'w+')
f_out.write('{')
f_out.write('"features":' + json.dumps(all_data['features']) + ',')
f_out.write('"weights":' + json.dumps(all_data['weights']) + ',')
f_out.write('"classifications":' + json.dumps(all_data['classifications']))
f_out.write('}')
f_out.close()
if (to_plot == True):
for key in weights_map.keys():
plot_svm(
features, weights_map[key], 'SVM Feature Weights: ' + key,
'Feature', 'Weight', directory.replace('/logs/', '/plots/'),
file_name.replace('.json',
'.png').replace('svm', 'svm_' + key),
fig_num, verbose)
fig_num += 1
return svm_weights
def write_classification_accuracy(directory, file_name, log_file, fig_num,
verbose):
print 'Writing and Plotting Accuracy Over Time: ', file_name
data_directory = bias_util.data_directory
data_file_name = bias_util.data_file_name
all_logs, attr_logs, item_logs, help_logs, cat_logs = bias_util.recreate_logs(
directory, log_file)
logs = item_logs
dataset, attr_map = bias_util.read_data(data_directory, data_file_name)
classification, decisions_labels, decisions_cat, decisions_help = bias_util.get_classifications_and_decisions(
logs, dataset)
total_labeled = 0
total_correct = 0
decision_points = np.arange(1, len(all_logs) + 1)
accuracy = [-1] * len(all_logs)
current_labels = dict()
correct_labels = dict()
for i in range(0, len(decisions_labels)):
cur = decisions_labels[i]
cur_id = cur[3]
correct_labels[cur_id] = cur[2]
if ((cur_id not in current_labels and cur[1] != 'Un-Assign') or
(cur_id in current_labels and current_labels[cur_id] == 'Un-Assign'
and cur[1] != 'Un-Assign')):
total_labeled += 1
elif (cur_id in current_labels and cur[1] == 'Un-Assign'
and current_labels[cur_id] != 'Un-Assign'):
total_labeled -= 1
if (cur_id not in current_labels and cur[1] == correct_labels[cur_id]):
total_correct += 1
elif (cur_id in current_labels
and current_labels[cur_id] != correct_labels[cur_id]
and cur[1] == correct_labels[cur_id]):
total_correct += 1
if (cur_id in current_labels
and current_labels[cur_id] == correct_labels[cur_id]
and cur[1] != correct_labels[cur_id]):
total_correct -= 1
if (total_labeled != 0):
accuracy[cur[0]] = total_correct / float(total_labeled)
else:
accuracy[cur[0]] = 0
current_labels[cur_id] = cur[1]
if (len(decisions_labels) < 1):
first_decision = -1
else:
first_decision = decisions_labels[0][0]
accuracy = bias_util.remove_defaults(accuracy, first_decision)
accuracy = bias_util.forward_fill(accuracy)
if not os.path.exists(directory):
os.makedirs(directory)
f_out = open(directory + file_name, 'w+')
f_out.write('[')
for i in range(0, len(decisions_labels)):
f_out.write('{')
f_out.write('"interaction_number":"' + str(decisions_labels[i][0]) +
'",')
f_out.write('"data_point":"' + str(decisions_labels[i][3]) + '",')
f_out.write('"actual_class":"' + str(decisions_labels[i][2]) + '",')
f_out.write('"user_class":"' + str(decisions_labels[i][1]) + '",')
f_out.write('"current_accuracy":"' +
str(accuracy[decisions_labels[i][0]]) + '"')
f_out.write('}')
if (i != len(decisions_labels) - 1):
f_out.write(',')
f_out.write(']')
f_out.close()
plot_classification_accuracy(decision_points, accuracy,
'Accuracy Over Time', 'Interactions',
'Accuracy',
directory.replace('/logs/', '/plots/'),
file_name.replace('.json', '.png'),
decisions_labels, fig_num, verbose)
def write_id_confusion_matrix(directory, file_name, log_file, fig_num,
verbose):
if (verbose):
print 'Writing and Plotting ID-Confusion Matrix Data: ', file_name
data_directory = bias_util.data_directory
data_file_name = bias_util.data_file_name
all_logs, attr_logs, item_logs, help_logs, cat_logs = bias_util.recreate_logs(
directory, log_file)
logs = item_logs
dataset, attr_map = bias_util.read_data(data_directory, data_file_name)
id_confusion, pos_to_num_map, all_data = get_id_confusion_matrix(
logs, dataset)
if not os.path.exists(directory):
os.makedirs(directory)
f_out = open(directory + file_name, 'w+')
summary = dict()
summary['rows (y)'] = 'user'
summary['cols (x)'] = 'actual'
summary['position_indices'] = pos_to_num_map
summary['centroids'] = dict()
summary['centroids']['user_centroids'] = dict()
summary['centroids']['actual_centroids'] = dict()
# separate out user labels and actual labels
user_labels = dict()
actual_labels = dict()
for key in all_data.keys():
cur_user_label = key[5:key.index(',')]
cur_actual_label = key[key.index('actual') + 7:]
cur_data_point = copy.deepcopy(all_data[key])
if (cur_user_label in user_labels.keys()):
user_labels[cur_user_label] += cur_data_point
else:
user_labels[cur_user_label] = cur_data_point
if (cur_actual_label in actual_labels.keys()):
actual_labels[cur_actual_label] += cur_data_point
else:
actual_labels[cur_actual_label] = cur_data_point
# compute centroids
for key in user_labels.keys():
summary['centroids']['user_centroids'][key] = compute_centroid(
user_labels[key])
#if (verbose):
# print 'User Centroid', key, summary['centroids']['user_centroids'][key]
for key in actual_labels.keys():
summary['centroids']['actual_centroids'][key] = compute_centroid(
actual_labels[key])
#if (verbose):
# print 'Actual Centroid', key, summary['centroids']['actual_centroids'][key]
# get total accuracy
num_correct = 0
total_classifications = 0
for i in range(0, len(id_confusion)):
for j in range(0, len(id_confusion[i])):
total_classifications += id_confusion[i][j]
if (i == j):
num_correct += id_confusion[i][j]
summary['total_accuracy'] = str(num_correct) + '/' + str(
total_classifications)
summary['matrix'] = id_confusion.tolist()
f_out.write('{')
f_out.write('"summary":' + json.dumps(summary) + ',')
f_out.write('"all_data":' + json.dumps(all_data))
f_out.write('}')
f_out.close()
# plot the matrix
labels = [
bias_util.num_to_pos_map[0], bias_util.num_to_pos_map[1],
bias_util.num_to_pos_map[2], bias_util.num_to_pos_map[3],
bias_util.num_to_pos_map[4]
]
plot_id_conf_matrix(id_confusion.tolist(), labels,
directory.replace('/logs/', '/plots/'),
file_name.replace('.json', '.png'), fig_num)