| import os |
| import pickle |
| from _bootstrap import DATA_DIR, OUTPUT_DIR |
| from model import get_model |
| from tf_keras.callbacks import Callback |
| from datetime import datetime
|
| from sklearn.metrics import roc_auc_score,average_precision_score
|
|
|
| cdr_kmer = 3
|
| ag_kmer = 1
|
| epochs=100
|
| batch_size=64
|
| data_path = DATA_DIR / 'features' / ('cdr_kmer' + str(cdr_kmer) + '_ag_kmer' + str(ag_kmer)) |
| directory_path = OUTPUT_DIR / 'checkpoints' / ('cdr_kmer' + str(cdr_kmer) + '_ag_kmer' + str(ag_kmer)) |
|
|
|
|
| def create_directory_if_not_exists(directory_path):
|
| if not os.path.exists(directory_path):
|
| os.makedirs(directory_path)
|
| print(f"Directory '{directory_path}' created.")
|
| else:
|
| print(f"Directory '{directory_path}' already exists.")
|
|
|
| class roc_callback(Callback):
|
| def __init__(self, val_data):
|
| self.cdr_ids = val_data[0]
|
| self.cdr_number_ids = val_data[1]
|
| self.ag_ids = val_data[2]
|
| self.labels = val_data[3]
|
|
|
| def on_train_begin(self, logs={}):
|
| return
|
|
|
| def on_train_end(self, logs={}):
|
| return
|
|
|
| def on_epoch_begin(self, epoch, logs={}):
|
| return
|
|
|
| def on_epoch_end(self, epoch, logs={}):
|
| labels_pred = self.model.predict([self.cdr_ids, self.cdr_number_ids, self.ag_ids])
|
| auc_val = roc_auc_score(self.labels, labels_pred)
|
| aupr_val = average_precision_score(self.labels, labels_pred)
|
|
|
| create_directory_if_not_exists(directory_path)
|
| self.model.save_weights(str(directory_path / ("Model%d.weights.h5" % epoch))) |
|
|
| print('\r auc_val: %s ' %str(round(auc_val, 4)), end=100 * ' ' + '\n')
|
| print('\r aupr_val: %s ' % str(round(aupr_val, 4)), end=100 * ' ' + '\n')
|
| return
|
|
|
| def on_batch_begin(self, batch, logs={}):
|
| return
|
|
|
| def on_batch_end(self, batch, logs={}):
|
| return
|
|
|
| t1 = datetime.now().strftime('%Y-%m-%d-%H:%M:%S')
|
|
|
|
|
| with (data_path / 'cdr_features_tr.pickle').open('rb') as binary_reader: |
| cdr_features_tr = pickle.load(binary_reader)
|
| with (data_path / 'ag_features_tr.pickle').open('rb') as binary_reader: |
| ag_features_tr = pickle.load(binary_reader)
|
|
|
| with (data_path / 'cdr_features_val.pickle').open('rb') as binary_reader: |
| cdr_features_val = pickle.load(binary_reader)
|
| with (data_path / 'ag_features_val.pickle').open('rb') as binary_reader: |
| ag_features_val = pickle.load(binary_reader)
|
|
|
|
|
| dtrain_cdr_ids = []
|
| dtrain_cdr_number_ids = []
|
| dtrain_ag_ids = []
|
| dtrain_labels = []
|
| dtrain_labels_pos = 0
|
| dtrain_labels_neg = 0
|
|
|
| for feature in cdr_features_tr:
|
| dtrain_cdr_ids.append(feature.input_ids)
|
| dtrain_cdr_number_ids.append(feature.cdr_number_ids)
|
| dtrain_labels.append(feature.label_id)
|
| dtrain_labels_pos = dtrain_labels_pos + feature.label_id
|
| dtrain_labels_neg = len(dtrain_labels) - dtrain_labels_pos
|
|
|
| for feature in ag_features_tr:
|
| dtrain_ag_ids.append(feature.input_ids)
|
|
|
|
|
|
|
| dval_cdr_ids = []
|
| dval_cdr_number_ids = []
|
| dval_ag_ids = []
|
| dval_labels = []
|
| dval_labels_pos = 0
|
| dval_labels_neg = 0
|
|
|
| for feature in cdr_features_val:
|
| dval_cdr_ids.append(feature.input_ids)
|
| dval_cdr_number_ids.append(feature.cdr_number_ids)
|
| dval_labels.append(feature.label_id)
|
| dval_labels_pos = dval_labels_pos + feature.label_id
|
| dval_labels_neg = len(dval_labels) - dval_labels_pos
|
|
|
| for feature in ag_features_val:
|
| dval_ag_ids.append(feature.input_ids)
|
|
|
|
|
|
|
| model=None
|
| model=get_model()
|
| model.summary()
|
|
|
| print ('Training the model')
|
|
|
| back = roc_callback(val_data=[dval_cdr_ids, dval_cdr_number_ids, dval_ag_ids, dval_labels])
|
|
|
| history=model.fit([dtrain_cdr_ids, dtrain_cdr_number_ids, dtrain_ag_ids], dtrain_labels,
|
| validation_data=([dval_cdr_ids, dval_cdr_number_ids, dval_ag_ids], dval_labels),
|
| epochs=epochs,
|
| batch_size=batch_size,
|
| callbacks=[back])
|
|
|
| t2 = datetime.now().strftime('%Y-%m-%d-%H:%M:%S')
|
| print("开始时间:"+t1+"结束时间："+t2)
|
|
|
