def restart():
# clear_db()
# clear_dirs()
# print("SCRAPING STORIES")
# scrape_stories()
print("CONSTRUCTING KEYWORD GRAPH")
construct_graph()
print("CREATING MODEL")
create_model()
print("CREATING TF-IDF")
tf_idf()
def __init__(self, model_name='model'):
try:
json_file = open('model/%s.json' % model_name,
'r') # open model structure
loaded_model_json = json_file.read() # read json
json_file.close() # close file
loaded_model = model_from_json(
loaded_model_json) # make json keras model
loaded_model.load_weights("model/%s.h5" %
model_name) # import the weights
self.model = loaded_model # store loaded model as self.model
except:
if model_name == 'model':
print('Model not found. Training model...'
) # if model is not found, then build model
self.model = create_model() # build model
print('Model trained.')
else:
json_file = open('model/%s.json' % model_name,
'r') # open model structure
loaded_model.load_weights("model/%s.h5" %
model_name) # import the weights
def plot_loss(lossList):
# xnew = np.linspace(0, len(lossList), len(lossList)*0.5)
# loss_smooth = spline(np.arange(len(lossList)), lossList, xnew)#损失函数平滑处理
plt.style.use("ggplot")
plt.figure()
# loss
# plt.plot(xnew, loss_smooth, 'g', label='loss_smo0th')
plt.plot(range(len(lossList)), lossList, 'r', label='train loss') #原版曲线
plt.grid(True)
plt.xlabel('batch')
plt.ylabel('loss')
plt.legend(loc="upper right")
plt.show()
model = create_model()
model.load_weights('my_model_weights.h5')
model.compile(optimizer=SGD(lr=0.0001, momentum=0.9))
HDF5 = h5py.File('singlesofa_data.h5', 'r')
anchor_data = HDF5['anchor']
positive_data = HDF5['positive']
negative_data = HDF5['negative']
BATCH_SIZE = 32
lossList = []
EPOCHS = 100
TOTAL_STEPS = len(anchor_data)
STEPS_PER_EPOCH = int(TOTAL_STEPS / BATCH_SIZE) + 1
# STEPS_PER_EPOCH = 100
for i in range(0, EPOCHS):
if i != 0: