def model_coor(model_statis: list, label, thresholds, val_index):
_, val_files = data_loader.get_k_fold_files("1.txt", val_index, [config.DATA_TYPE_ORIGINAL])
y = data_loader.get_labels(val_files)
y = np.array(y, np.int8).reshape((-1, 13))
model_names = []
model_predicts = []
df = pd.DataFrame()
for i in model_statis:
weight_path = i[0]
name, info = path_2_model_name(weight_path)
if not os.path.exists(weight_path + ".predict.npy"):
continue
predict = np.load(weight_path + ".predict.npy")
if thresholds is not None:
for l in range(len(thresholds)):
threshold = thresholds[l][weight_path]
predict[:, l] = predict[:, l] > threshold
predict = predict.astype(np.int8)
try:
if label is not None:
predict = predict[:, label]
f2 = fbeta_score(y[:, label], predict, beta=2)
else:
f2 = fbeta_score(y, predict, beta=2, average='macro')
assert (f2 - i[1]) / i[1] < 0.01
model_predicts.append(predict)
model_names.append(name)
df[name] = predict.flatten()
except:
print(weight_path)
return df.corr()
def get_label_dic():
data_list1 = data_loader.get_labels(data_loader.list_image_name(DATA_FILE))
data_list2 = data_loader.get_labels(
data_loader.list_image_name(DATA_COLOR_FILE))
data_list = data_list1 + data_list2
data_dic = {}
for _data in data_list:
data_ = str(_data)
if data_ not in list(data_dic):
data_dic[data_] = 1
else:
data_dic[data_] += 1
with open("data_dic.txt", 'w') as f:
for k, v in data_dic.items():
log = k + " : " + str(v) + '\n'
f.write(log)
def do_statistic():
images = data_loader.list_image_dir(path.COLOR_AUGMENTED_PATH)
augment_images = []
for image in images:
if "original" not in image:
augment_images.append(image)
labels = data_loader.get_labels(augment_images)
labels = np.array(labels)
data_metric.label_analysis(labels)
def show_label_calss_bar_per_epoch(train_file, record_dir):
labels = data_loader.get_labels(train_file)
labels = np.array(labels)
df = pd.DataFrame(labels, columns=get_columns())
_x = df.sum(axis=0).tolist()
_y = label_class_name.copy()
for i in range(len(_y)):
_y[i] += '(%d)' % _x[i]
plt.figure(figsize=(15, 6))
sns.barplot(y=_y, x=_x, orient='h')
record_save = record_dir + '_' + 'label_calss_bar'
plt.savefig(record_save)
def get_frame(file):
data = data_loader.get_labels(data_loader.list_image_name(file))
data = np.array(data)
df = pd.DataFrame(data, columns=get_columns())
return df
import sys
import numpy as np
import pathlib
sys.path.append(os.path.abspath("../"))
import config
from util import data_loader
from util import data_visualization as dv
from util import keras_util
from util import path
y_val = {}
for i in range(1, 6):
val1_train_files, val1_val_files = data_loader.get_k_fold_files(
"1.txt", i, [config.DATA_TYPE_SEGMENTED], shuffle=False)
y_val[i] = np.array(data_loader.get_labels(val1_val_files), np.bool)
def check_invalid_model(path):
for root, dirs, files in os.walk(path):
for file in files:
if "model" in file and "val" in file and ".py" in file and "pyc" not in file:
record_dir = os.path.join(os.path.join(root, "record"),
file.split("_")[0])
record_dir = os.path.join(record_dir,
file.split("_")[1].split(".")[0])
weight_path = os.path.join(
record_dir,
"%sweights.%03d.hdf5" % (str([str(j)
for j in range(13)]), 3))