def add_race_modifiers(self, char):
char.info.Race = self.race_name
char.stats.size.Race = self.size
char.stats.speed.Race = self.speed
char.proficiencies.languages.Race = self.languages
for trait in vars(self).keys():
if trait == 'attributes':
for attr in self.attributes:
char.attributes[attr[0]]['race'] = attr[1]
elif trait == 'skills':
for skill in self.skills:
char.skills[skill].prof += [self.race_name]
elif trait == 'feats':
from feats import get_feat
for feat in self.feats:
new_feat = get_feat(feat, self.race_name)
char.feats[feat] = new_feat
char.feats[feat].initial_effects(char)
elif trait == 'features':
from features import get_feature
for feature in self.features:
new_feature = get_feature(feature)
char.features[self.race_name][feature] = new_feature
new_feature.initial_effects(char)
elif trait not in ['race_name', 'size', 'speed', 'languages']:
raise Exception(f"{trait} included which hasn't been added.")
def predict(train_data,
train_labels,
test_path,
labels,
result_path,
n_estimators,
random_state,
fixed_size=DEF_SIZE):
clf = RandomForestClassifier(n_estimators=n_estimators,
random_state=random_state,
n_jobs=-1)
clf.fit(train_data, train_labels)
# loop through the test images
for ind, file in enumerate(glob.glob(test_path + "/*.*")):
# read the image
image = cv2.imread(file)
# ресайзим все к одному размеру
image = cv2.resize(image, fixed_size)
# get feature
global_feature = get_feature(image)
# predict label of test image
prediction = clf.predict(global_feature.reshape(1, -1))[0]
# display the output image
display(image, labels[prediction], f'{result_path}/{ind}.png')
def predict(self, fs, signal):
"""
return a label (name)
"""
try:
feat = get_feature(fs, signal)
except Exception as e:
print (e)
return self.gmmset.predict_one(feat)
def verify(self, fs, signal, personid):
"""
return a label (name)
"""
try:
feat = get_feature(fs, signal)
except Exception as e:
print(e)
return self.gmmset.verify(feat, personid)
def add_class_features(self, char):
if not hasattr(char, 'classes'):
char.classes = Dict()
origin = 'origin: ' + self.class_name
for trait in vars(self).keys():
if trait == 'saves':
for save in self.saves:
char.saving_throws[save].prof += [origin]
elif trait == 'skills':
for skill in self.skills:
char.skills[skill].prof += [origin]
elif trait == 'prof_weapons':
char.proficiencies.weapons[origin] = self.prof_weapons
elif trait == 'prof_armor':
char.proficiencies.armor[origin] = self.prof_armor
elif trait == 'prof_tools':
char.proficiencies.tools[origin] = self.prof_tools
elif trait == 'equipment':
self.equipment = f.choose_weapons(self.equipment)
for item in self.equipment:
if item[0] in char.equipment:
char.equipment[item[0]].add_number(item[1])
else:
char.equipment.update(
{item[0]: get_item(item[0], item[1])})
elif trait not in [
'class_name', 'hit_dice', 'hit_points', 'levels'
]:
raise Exception(
f"{trait} included which haven't been added.")
char.classes[self.class_name] = character_class(
1, self.hit_dice, self.hit_points, True)
else:
char.classes[self.class_name] = character_class(
1, self.hit_dice, self.hit_points, False)
for feature in self.levels[1]:
new_feature = get_feature(feature)
char.features[f"Class: {self.class_name}"][feature] = new_feature
new_feature.initial_effects(char)
def task_predict(input_files, input_model):
# 把输入的多个模型目录字符串分离为目录列表
input_models = [os.path.expanduser(k) for k in input_model.strip().split()]
# 把各个目录下的模型列表解压出来组合成一个迭代器
models = itertools.chain(*(glob.glob(m) for m in input_models))
# 生成并加载包括所有模型文件的列表
models = [ModelInterface.load(m) for m in models]
# 定义统计准确率的变量
right = 0
wrong = 0
num = 0
# 对每个预测音频文件提取特征并与每个模型匹配得到TOP结果
for f in glob.glob(os.path.expanduser(input_files)):
fs, signal = read_wav(f)
print(f)
feat = get_feature(fs, signal)
predict_result = []
# 每个音频文件分别匹配每个模型组并得出分数放到总列表
for model in models:
#print(model)
#m = ModelInterface.load(model)
results = model.predict(feat)
for result in results:
predict_result.append(result)
#print("predict_result:",predict_result)
# 对预测结果按分数作高到底排序
predict_result = sorted(predict_result,
key=operator.itemgetter(1),
reverse=True)
#print("sort_predict_result:", predict_result)
# 微信语音数据集的label格式
label = os.path.basename(f).split('_')[0] #[6:11]
#label=os.path.basename(f).split('(')[0]#[6:11]
# AISHELL数据集的label格式
#label=os.path.basename(f)[6:11]
predict = predict_result[0][0]
#print('Top:',predict_result[:10])
# 统计准确率
if label in predict:
right += 1
print('label:', label, ' predict:', predict, ' right')
else:
wrong += 1
print('label:', label, ' predict:', predict, ' wrong')
num += 1
print('All:', num, ' right:', right, ' wrong:', wrong, ' acc:',
right / num)
def init_h5py(train_path: str,
train_labels: list[str],
h5_data_file: str,
h5_labels_file: str,
fixed_img_size=DEF_SIZE):
global_features = []
labels = []
# loop over the training data sub-folders
for training_name in train_labels:
current_dir = os.path.join(train_path, training_name)
images = [
f for f in os.listdir(current_dir)
if os.path.isfile(os.path.join(current_dir, f))
and f.endswith(".jpg") or f.endswith(".png")
]
for file in images:
file_path = os.path.join(current_dir, file)
# read the image and resize it to a fixed-size
image = cv2.imread(file_path)
image = cv2.resize(image, fixed_img_size)
global_feature = get_feature(image)
# update the list of labels and feature vectors
labels.append(training_name)
global_features.append(global_feature)
print(f"[STATUS] processed folder: {training_name}")
le = LabelEncoder()
target = le.fit_transform(labels)
# save the feature vector using HDF5
h5f_data = h5py.File(h5_data_file, 'w')
h5f_data.create_dataset('dataset_1', data=np.array(global_features))
h5f_label = h5py.File(h5_labels_file, 'w')
h5f_label.create_dataset('dataset_1', data=np.array(target))
h5f_data.close()
h5f_label.close()
print("[STATUS] end of training..")
def enroll(self, name, fs, signal):
feat = get_feature(fs, signal)
self.features[name].extend(feat)
def task_predict(input_files, input_model):
# 把输入的多个模型目录字符串分离为目录列表
input_models = [os.path.expanduser(k) for k in input_model.strip().split()]
# 把各个目录下的模型列表解压出来组合成一个迭代器
models = itertools.chain(*(glob.glob(m) for m in input_models))
# 生成并加载包括所有模型文件（skgmm.GMMSet object）的列表
models = [ModelInterface.load(m) for m in models]
if len(models) == 0:
print("No model file found in %s" % input_model)
sys.exit(1)
# 定义统计准确率的变量
right = 0
right1 = 0
wrong = 0
wrong1 = 0
num = 0
# 对每个预测音频文件提取特征并与每个模型匹配得到TOP结果
for f in glob.glob(os.path.expanduser(input_files)):
fs, signal = read_wav(f)
print(f)
feat = get_feature(fs, signal)
predict_result = []
# 每个音频文件分别匹配每个模型组并得出分数放到总列表
for model in models:
#print(model)
#m = ModelInterface.load(model)
results = model.predict_proba(feat)
print(results)
#print(results)
for result in results:
predict_result.append(result)
#print("predict_result:",predict_result)
# 对预测结果按分数作高到底排序
predict_result = sorted(predict_result,
key=operator.itemgetter(1),
reverse=True)
#print("sort_predict_result:", predict_result)
# 微信语音数据集的label格式
label = os.path.basename(f).split('_')[0] #[6:11]
#label=os.path.basename(f).split('(')[0]#[6:11]
# AISHELL数据集的label格式
# label=os.path.basename(f)[6:11]
predict = predict_result[0][0]
predict_score = predict_result[0][1]
# #print('Top:',predict_result[:10])
# 统计top1准确率
if label in predict:
right1 += 1
print('label:', label, ' predict:', predict, ' score:',
predict_score, ' top1 right')
else:
wrong1 += 1
print('label:', label, ' predict:', predict, ' score:',
predict_score, ' top1 wrong')
# 统计Top10准确率
predicts = []
predict_scores = []
for pre in predict_result[:10]:
predicts.append(pre[0])
predict_scores.append(pre[1])
if label in predicts:
right += 1
print('label:', label, ' predicts:', predicts, ' scores:',
predict_scores, ' top10 Right')
else:
wrong += 1
print('label:', label, ' predicts:', predicts, ' scores:',
predict_scores, ' top10 Wrong')
num += 1
print('top1:', num, ' right:', right1, ' wrong:', wrong1, ' top1 acc:',
right1 / num)
print('top10:', num, ' right:', right, ' wrong:', wrong, ' top10 acc:',
right / num)
def enroll(self, name, fs, signal):
feat = get_feature(fs, signal)
#print("feat:",feat)
#print(len(feat))
self.features[name].extend(feat)
def enroll(self, name, fs, signal):
print("enroll "+name)
feat = get_feature(fs, signal)
self.features[name].extend(feat)
def task_predict(input_files, input_model):
# 把输入的多个模型目录字符串分离为目录列表
input_models = [os.path.expanduser(k) for k in input_model.strip().split()]
# 把各个目录下的模型列表解压出来组合成一个迭代器
models = itertools.chain(*(glob.glob(m) for m in input_models))
# 生成并加载包括所有模型文件（skgmm.GMMSet object）的列表
models = [ModelInterface.load(m) for m in models]
if len(models) == 0:
print("No model file found in %s" % input_model)
sys.exit(1)
# 定义统计准确率的变量
right = 0
right1 = 0
wrong = 0
wrong1 = 0
num = 0
# 对每个预测音频文件提取特征并与每个模型匹配得到TOP结果
for f in glob.glob(os.path.expanduser(input_files)):
start_time = time.time()
fs, signal = read_wav(f)
print(f)
feat = get_feature(fs, signal)
#print("Get feature ", time.time() - start_time, " seconds")
predict_result = []
f_models = [(feat, m) for m in models]
#print(models)
# 每个音频文件分别匹配每个模型组并得出分数放到总列表
# for model in models:
# #start_time1 = time.time()
# #print(model)
# # 模型文件是一个元组：(label,gmm)
# score = model[1].score(feat)
# label=model[0]
# result=(label,score)
# #print(results)
# predict_result.append(result)
#print("Get one score ", time.time() - start_time1, " seconds")
pool = ThreadPool(2)
predict_result = pool.map(get_score, f_models)
pool.close()
pool.join()
#print(results)
#print("Get score ", time.time() - start_time, " seconds")
proba = GMMSet.softmax([i[1] for i in predict_result])
predict_result = [(predict_result[i][0], proba[i])
for i in range(len(proba))]
#print("predict_result:",predict_result)
# 对预测结果按分数作高到底排序
predict_result = sorted(predict_result,
key=operator.itemgetter(1),
reverse=True)
#print("sort_predict_result:", predict_result)
# 微信语音数据集的label格式
label = os.path.basename(f).split('_')[0] #[6:11]
#label=os.path.basename(f).split('(')[0]#[6:11]
# AISHELL数据集的label格式
# label=os.path.basename(f)[6:11]
predict = predict_result[0][0]
predict_score = predict_result[0][1]
print("Predict ", time.time() - start_time, " seconds")
# #print('Top:',predict_result[:10])
# 统计top1准确率
if label in predict:
right1 += 1
print('label:', label, ' predict:', predict, ' score:',
predict_score, ' top1 right')
else:
wrong1 += 1
print('label:', label, ' predict:', predict, ' score:',
predict_score, ' top1 wrong')
# 统计Top10准确率
predicts = []
predict_scores = []
for pre in predict_result[:10]:
predicts.append(pre[0])
predict_scores.append(pre[1])
if label in predicts:
right += 1
print('label:', label, ' predicts:', predicts, ' scores:',
predict_scores, ' top10 Right')
else:
wrong += 1
print('label:', label, ' predicts:', predicts, ' scores:',
predict_scores, ' top10 Wrong')
num += 1
print('top1:', num, ' right:', right1, ' wrong:', wrong1, ' top1 acc:',
right1 / num)
print('top10:', num, ' right:', right, ' wrong:', wrong, ' top10 acc:',
right / num)
