def view_tongue_data_augment():
patient_tongue_dir = config['root_path'] + \
config['original_path'] + 'tongue_9585'
tongue_zhiliao_path = config['root_path'] + \
config['original_path'] + 'tongue_zhiliao.list'
yaopin_path = config['root_path'] + \
config['original_path'] + 'yaopin.vocab'
tongue_ids, tongue_image_arrays, tongue_yaofangs, tongue_image_shape = patient_tongue_generator.loadDatafromFile(
patient_tongue_dir, tongue_zhiliao_path, image_normal_size=(224, 224))
# fetch max(id) in yaopin.vocab as nb_yao
with open(yaopin_path, 'r') as yaopin_file:
nb_yao = max(
int(line.split(' ')[0]) for line in yaopin_file.readlines())
total_tongue_x, total_y = tongue2text_gen.data_tensorization(
tongue_image_arrays, tongue_yaofangs, tongue_image_shape, nb_yao)
datagen = image_augment.image_augment_gen()
augmented_x, augmented_y = image_augment.data_tensoration_augment(
datagen, total_tongue_x, total_y)
print(np.shape(augmented_x))
print(np.shape(augmented_y))
def tfidf_weights_test():
patient_tongue_dir = config['root_path'] + \
config['original_path'] + 'tongue_9585'
tongue_zhiliao_path = config['root_path'] + \
config['original_path'] + 'tongue_zhiliao.list'
yaopin_path = config['root_path'] + \
config['original_path'] + 'yaopin.vocab'
tongue_ids, tongue_image_arrays, tongue_yaofangs, tongue_image_shape = patient_tongue_generator.loadDatafromFile(
patient_tongue_dir, tongue_zhiliao_path, image_normal_size=(224, 224))
patient_tongue_dir = config['root_path'] + \
config['original_path'] + 'tongue_9585'
tongue_zhiliao_path = config['root_path'] + \
config['original_path'] + 'tongue_zhiliao.list'
yaofangs_corpus = tfidf.list2corpus(tongue_yaofangs)
word, weight = tfidf.get_tf_idf(yaofangs_corpus)
weight = list(weight)
min = 999999
for i in range(len(weight)):
for j in range(len(weight[i])):
if weight[i][j] > 0.0 and weight[i][j] < min:
min = weight[i][j]
print(min)
def train_predict_tongue2text_basic_gen(train_new=True):
'''
@param train_new: flag of train a new model model and replace the model on disk
'''
patient_tongue_dir = config['root_path'] + \
config['original_path'] + 'tongue_9585'
tongue_zhiliao_path = config['root_path'] + \
config['original_path'] + 'tongue_zhiliao.list'
yaopin_path = config['root_path'] + \
config['original_path'] + 'yaopin.vocab'
# tongue_ids: [01012045534615_1_4_7, ...]
# tongue_image_array: [np.array(pixels matrix of image), np.array(pixels matrix of image2), ...]
# tongue_yaofangs: [ [0,1,2,3], [4,5,6,7], ... ]
tongue_ids, tongue_image_arrays, tongue_yaofangs, tongue_image_shape = patient_tongue_generator.loadDatafromFile(
patient_tongue_dir, tongue_zhiliao_path, image_normal_size=(224, 224))
# fetch max(id) in yaopin.vocab as nb_yao
with open(yaopin_path, 'r') as yaopin_file:
nb_yao = max(
int(line.split(' ')[0]) for line in yaopin_file.readlines())
# _use_data_augment = True # set for use image data augment, can only be use on service 225 with big memory
_use_data_augment = False
'''
The part of train a new gen_model and storage it on disk,
the new one will cover the old one
'''
# store keras layers_framework(optional, gen_frame_path==None or not)
frame_name = 'tongue2text_cnnmlp_9585_act(bi)_t3_100it.json'
gen_frame_path = config['root_path'] + \
config['cache_path'] + 'keras/' + frame_name
train_on_batch = False # switch train_on_batch or not
if train_new == True:
_ = patient_tongue_generator.tongue_basic_gen_trainer(
tongue_image_arrays,
tongue_yaofangs,
tongue_image_shape,
nb_yao,
gen_model_path=gen_frame_path,
train_on_batch=train_on_batch,
use_data_augment=_use_data_augment)
'''
The part of load a trained gen_model from disk,
the trained gen_model will be reload and use to eval and predict directly,
without retraining which is for time saving
'''
trained_gen_model = tongue2text_gen.loadStoredModel(gen_frame_path,
gen_frame_path.replace(
'.json', '.h5'),
compile_info={
'recompile':
True,
'aux_output':
False,
'use_tfidf_tensor':
False
})
# test
# gen_output: [ [0.8, 0.4., ...], [...], [...], ... ]
gen_output = patient_tongue_generator.basic_gen_predictor_test(
tongue_image_arrays, tongue_yaofangs, tongue_image_shape, nb_yao,
trained_gen_model)
# print(gen_output[0])
# yaopin_dict: {0:'麻黄',1:'桂枝',...}
yaopin_dict = patient_tongue_generator.load_yaopin_dict(yaopin_path)
# print(yaopin_dict)
# test_tongue_ids = tongue_ids[: 500]
# test_yaofangs = tongue_yaofangs[: 500]
# test_tongue_ids = tongue_ids[2000 : 2500]
# test_yaofangs = tongue_yaofangs[2000 : 2500]
# test_tongue_ids = tongue_ids[4000 : 4500]
# test_yaofangs = tongue_yaofangs[4000 : 4500]
# test_tongue_ids = tongue_ids[6000 : 6500]
# test_yaofangs = tongue_yaofangs[6000 : 6500]
test_tongue_ids = tongue_ids[len(tongue_ids) - 500:]
test_yaofangs = tongue_yaofangs[len(tongue_yaofangs) - 500:]
'''the evaluation criterion '''
precisions = []
recalls = []
errors = []
for i, output in enumerate(gen_output):
# print test data label info:
print('%d. \npatient tongue id: %s' % (i, test_tongue_ids[i]))
print('label yaofang:')
yaofang_label = patient_tongue_generator.sample_yaofang(
test_yaofangs[i], yaopin_dict)
print(' '.join(yaofang_label))
# output_index = patient_tongue_generator.dynamic_threshold_outputfilter(output)
output_index = patient_tongue_generator.threshold_outputfilter(output)
# print('predicted yaofang ids: {0}'.format(output_index))
yaofang_output = patient_tongue_generator.sample_yaofang(
output_index, yaopin_dict)
print('predicted yaofang:')
print(' '.join(yaofang_output) + '\n')
precision, recall, error = generator_eval.evaluator(
test_yaofangs[i], output_index)
precisions.append(precision)
recalls.append(recall)
errors.append(error)
print('------Score: precision: %f, recall: %f, error: %f' %
(precision, recall, error))
print(
'------Average Score: average precision: %f, average recall: %f, error: %f'
% (np.average(precisions), np.average(recalls), np.average(errors)))
def train_predict_tongue2text_cnn2_withlda_gen(train_new=True):
'''
@param train_new: flag of train a new model model and replace the model on disk
'''
patient_tongue_dir = config['root_path'] + \
config['original_path'] + 'tongue_9585'
tongue_zhiliao_path = config['root_path'] + \
config['original_path'] + 'tongue_zhiliao.list'
yaopin_path = config['root_path'] + \
config['original_path'] + 'yaopin.vocab'
tongue_ids, tongue_image_arrays, tongue_yaofangs, tongue_image_shape = patient_tongue_generator.loadDatafromFile(
patient_tongue_dir, tongue_zhiliao_path, image_normal_size=(224, 224))
# fetch max(id) in yaopin.vocab as nb_yao
with open(yaopin_path, 'r') as yaopin_file:
nb_yao = max(
int(line.split(' ')[0]) for line in yaopin_file.readlines())
# _use_tfidf_tensor = True # set for use tfidf_tensor
_use_tfidf_tensor = False
# _use_data_augment = True # set for use image data augment, can only be use on service 225 with big memory
_use_data_augment = False
'''
TODO: storage model and load it from disk
The part of train a new gen_model with lda and storage it on disk,
the new one will cover the old one
'''
lda_model_name = 'tongue_9585_gensim_lda.topic'
lda_model_path = config['root_path'] + \
config['cache_path'] + 'nlp/' + lda_model_name
# _lda_replace = True # first time is True, other is False if not needed
_lda_replace = False
if _use_tfidf_tensor == True:
frame_name = 'tongue2text_cnn2passmlp_lda_9585_act(tfidf)_t3_100it.json'
else:
frame_name = 'tongue2text_cnn2passmlp_lda_9585_act(bi)_t3_100it.json'
gen_frame_path = config['root_path'] + \
config['cache_path'] + 'keras/' + frame_name
if train_new == True:
_ = patient_tongue_generator.tongue_gen_withlda_trainer(
tongue_image_arrays,
tongue_yaofangs,
tongue_image_shape,
nb_yao,
lda_model_path,
gen_model_path=gen_frame_path,
lda_replace=_lda_replace,
use_tfidf_tensor=_use_tfidf_tensor,
use_data_augment=_use_data_augment)
'''
The part of load a trained gen_model from disk,
the trained gen_model will be reload and use to eval and predict directly,
without retraining which is for time saving
'''
trained_gen_model = tongue2text_gen.loadStoredModel(gen_frame_path,
gen_frame_path.replace(
'.json', '.h5'),
compile_info={
'recompile':
True,
'aux_output':
True,
'use_tfidf_tensor':
_use_tfidf_tensor
})
# test
# gen_output: [ [0.8, 0.4., ...], [...], [...], ... ]
gen_output_list = patient_tongue_generator.gen_withlda_predictor_test(
tongue_image_arrays,
tongue_yaofangs,
tongue_image_shape,
nb_yao,
trained_gen_model,
lda_model_path,
use_tfidf_tensor=_use_tfidf_tensor)
# just get the gen_output, dropout the aux_output
gen_output = gen_output_list[0]
del (gen_output_list)
# print(gen_output[0])
# yaopin_dict: {0:'麻黄',1:'桂枝',...}
yaopin_dict = patient_tongue_generator.load_yaopin_dict(yaopin_path)
# print(yaopin_dict)
# test_tongue_ids = tongue_ids[: 500]
# test_yaofangs = tongue_yaofangs[: 500]
# test_tongue_ids = tongue_ids[2000 : 2500]
# test_yaofangs = tongue_yaofangs[2000 : 2500]
# test_tongue_ids = tongue_ids[4000 : 4500]
# test_yaofangs = tongue_yaofangs[4000 : 4500]
# test_tongue_ids = tongue_ids[6000 : 6500]
# test_yaofangs = tongue_yaofangs[6000 : 6500]
test_tongue_ids = tongue_ids[len(tongue_ids) - 500:]
test_yaofangs = tongue_yaofangs[len(tongue_yaofangs) - 500:]
'''the evaluation criterion '''
precisions = []
recalls = []
errors = []
for i, output in enumerate(gen_output):
# print test data label info:
print('%d. \npatient tongue id: %s' % (i, test_tongue_ids[i]))
print('label yaofang:')
yaofang_label = patient_tongue_generator.sample_yaofang(
test_yaofangs[i], yaopin_dict)
print(' '.join(yaofang_label))
# output_index = patient_tongue_generator.dynamic_threshold_outputfilter(output)
output_index = patient_tongue_generator.threshold_outputfilter(output)
# print('predicted yaofang ids: {0}'.format(output_index))
yaofang_output = patient_tongue_generator.sample_yaofang(
output_index, yaopin_dict)
print('predicted yaofang:')
print(' '.join(yaofang_output) + '\n')
precision, recall, error = generator_eval.evaluator(
test_yaofangs[i], output_index)
precisions.append(precision)
recalls.append(recall)
errors.append(error)
print('------Score: precision: %f, recall: %f, error: %f' %
(precision, recall, error))
print(
'------Average Score: average precision: %f, average recall: %f, error: %f'
% (np.average(precisions), np.average(recalls), np.average(errors)))
def train_predict_tongue2text_sklearn_gen(step=0):
'''
@param step: 0: keras train, 1: load keras model train sk_classifier and test
'''
patient_tongue_dir = config['root_path'] + \
config['original_path'] + 'tongue_9585'
tongue_zhiliao_path = config['root_path'] + \
config['original_path'] + 'tongue_zhiliao.list'
yaopin_path = config['root_path'] + \
config['original_path'] + 'yaopin.vocab'
# tongue_ids: [01012045534615_1_4_7, ...]
# tongue_image_array: [np.array(pixels matrix of image), np.array(pixels matrix of image2), ...]
# tongue_yaofangs: [ [0,1,2,3], [4,5,6,7], ... ]
tongue_ids, tongue_image_arrays, tongue_yaofangs, tongue_image_shape = patient_tongue_generator.loadDatafromFile(
patient_tongue_dir, tongue_zhiliao_path, image_normal_size=(224, 224))
# fetch max(id) in yaopin.vocab as nb_yao
with open(yaopin_path, 'r') as yaopin_file:
nb_yao = max(int(line.split(' ')[0])
for line in yaopin_file.readlines())
frame_name = 'test_tongue2text_cnn2mlp_9585_act(bi)_t3_100it.json'
if step == 0:
'''
The part of train a new sklearn_gen_model
@todo: need to storage keras feature scratch and sklearn generator
on disk together(with same type names)
'''
trained_gen_model = patient_tongue_generator.tongue_sklearn_gen_keras_trainer(
tongue_image_arrays, tongue_yaofangs, tongue_image_shape, nb_yao)
# store keras layers_framework(optional)
gen_frame_path = config['root_path'] + \
config['cache_path'] + 'keras/' + frame_name
tongue2text_sklearn_gen.storageKerasModel(
model=trained_gen_model, frame_path=gen_frame_path)
if step == 1:
'''
The part of load keras model from disk first and train the sklearn classifier
'''
gen_frame_path = config['root_path'] + \
config['cache_path'] + 'keras/' + frame_name
gen_record_path = gen_frame_path.replace('.json', '.h5')
print('load keras model from disk...')
trained_tongue_gen_model = tongue2text_sklearn_gen.loadStoredKerasModel(
gen_frame_path, gen_record_path, recompile=True)
trained_tongue_gen_classifier = patient_tongue_generator.tongue_sklearn_gen_sk_trainer(
tongue_image_arrays, tongue_yaofangs, tongue_image_shape, nb_yao, trained_tongue_gen_model)
'''
The part of test trained sklearn classifier-generator
'''
_proba_predict = True # set the output_type of sklearn classifier-generator(proba or not)
gen_output = patient_tongue_generator.sklearn_gen_predictor_test(
tongue_image_arrays, tongue_yaofangs, tongue_image_shape,
nb_yao, trained_tongue_gen_model, trained_tongue_gen_classifier,
proba_predict=_proba_predict)
print(gen_output[0])
# yaopin_dict: {0:'麻黄',1:'桂枝',...}
yaopin_dict = patient_tongue_generator.load_yaopin_dict(yaopin_path)
# print(yaopin_dict)
test_tongue_ids = tongue_ids[:200]
test_yaofangs = tongue_yaofangs[:200]
'''the evaluation criterion '''
precisions = []
recalls = []
errors = []
for i, output in enumerate(gen_output):
# print test data label info:
print('%d. \npatient tongue id: %s' % (i, test_tongue_ids[i]))
print('label yaofang:')
yaofang_label = patient_tongue_generator.sample_yaofang(
test_yaofangs[i], yaopin_dict)
print(' '.join(yaofang_label))
if _proba_predict == False:
output_index = patient_tongue_generator.label_outputfilter(
output)
else:
output_index = patient_tongue_generator.threshold_outputfilter(
output)
# print('predicted yaofang ids: {0}'.format(output_index))
yaofang_output = patient_tongue_generator.sample_yaofang(
output_index, yaopin_dict)
print('predicted yaofang:')
print(' '.join(yaofang_output) + '\n')
precision, recall, error = generator_eval.evaluator(
test_yaofangs[i], output_index)
precisions.append(precision)
recalls.append(recall)
errors.append(error)
print('------Score: precision: %f, recall: %f, error: %f' %
(precision, recall, error))
print('------Average Score: average precision: %f, average recall: %f, error: %f' %
(np.average(precisions), np.average(recalls), np.average(errors)))