for start_month in range(file_num, file_num + month - interval, interval):
predict_data[0] = file_helper_unformatted.read_data_sstaha(start_month)
nino34 = [index_calculation.get_nino34(predict_data[0])]
# data preprocess z-zero
if data_preprocess_method == 'preprocess_Z':
predict_data = data_preprocess.preprocess_Z(predict_data, 0)
# data preprocess dimensionless
if data_preprocess_method == 'dimensionless':
redict_data = data_preprocess.dimensionless(predict_data, 0)
# data preprocess 0-1
if data_preprocess_method == 'preprocess_01':
predict_data = data_preprocess.preprocess_01(predict_data, 0)
# data preprocess no month mean
if data_preprocess_method == 'nomonthmean':
predict_data = data_preprocess.no_month_mean(predict_data, 0)
if model_type == 'conv':
data_x[0] = predict_data[0]
elif model_type == 'dense':
data_x[0] = np.reshape(predict_data[0], (1, 1080))
for i in range(prediction_month):
if is_seasonal_circle:
data_sc[0] = [(start_month+i) % 12]
data_x = model.predict([data_x, data_sc])
else:
data_x = model.predict(data_x)
# data_x = model.predict(data_x)
if model_type == 'conv':
# the data for training is ssta and ha
training_data, testing_data = file_helper_unformatted.load_sstha_for_conv2d(
training_start, training_num)
# data preprocess z-zero
if data_preprocess_method == 'preprocess_Z':
training_data = data_preprocess.preprocess_Z(training_data, 0)
# data preprocess dimensionless
if data_preprocess_method == 'dimensionless':
training_data = data_preprocess.dimensionless(training_data, 0)
# data preprocess 0-1
if data_preprocess_method == 'preprocess_01':
training_data = data_preprocess.preprocess_01(training_data, 0)
# data preprocess no month mean
if data_preprocess_method == 'nomonthmean':
training_data = data_preprocess.no_month_mean(training_data, 0)
data_x = training_data[:-1]
data_y = training_data[1:]
# tesorboard = TensorBoard('..\..\model\\tensorboard\\' + model_name)
save_best = ModelCheckpoint('..\..\model\\best\\' + model_name + '@' +
'best.h5',
monitor='val_root_mean_squared_error',
verbose=1,
save_best_only=True,
mode='min',
period=1)
train_hist = model.fit(data_x,
data_y,
batch_size=batch_size,
epochs=epochs,
sc = np.linspace(0, 11, 12, dtype='int32')
sc = np.tile(sc, int((all_num - training_start) / 12 + 1))
data_sc = sc[:(all_num - training_start)]
# data preprocess z-zero
if data_preprocess_method == 'preprocess_Z':
all_data = data_preprocess.preprocess_Z(all_data, 0)
# data preprocess dimensionless
if data_preprocess_method == 'dimensionless':
all_data = data_preprocess.dimensionless(all_data, 0)
# data preprocess 0-1
if data_preprocess_method == 'preprocess_01':
all_data = data_preprocess.preprocess_01(all_data, 0)
# data preprocess no month mean
if data_preprocess_method == 'nomonthmean':
all_data = data_preprocess.no_month_mean(all_data, 0)
if model_type == 'conv':
data_x = all_data[:-1]
data_y = all_data[1:]
elif model_type == 'dense':
data_x = np.reshape(all_data[:-1], (all_num, 1080))
data_y = np.reshape(all_data[1:], (all_num, 1080))
# for layer in model.layers[:]:
# layer.trainable = False
# for layer in model.layers[6:20]:
# layer.trainable = True
for layer in model.layers:
print(layer.trainable)
save_best = ModelCheckpoint('..\model\\best\\' + continue_model_name +
def calculate_12_month(start_month):
# for dense_model ssta and ha
predict_data = np.empty([1, 20, 27, 2])
data_y = np.empty([1, 20, 27, 2])
data_realistic = np.empty([1, 20, 27, 2])
if is_seasonal_circle:
data_sc = np.empty([1, 1], dtype='int32')
if model_type == 'conv':
data_x = np.empty([1, 20, 27, 2])
else:
data_x = np.empty([1, 1080])
predict_data[0] = file_helper_unformatted.read_data_sstaha(start_month)
if is_retrain:
predict_data = file_helper_unformatted.exchange_rows(predict_data)
# data preprocess z-zero
if data_preprocess_method == 'preprocess_Z':
predict_data = data_preprocess.preprocess_Z(predict_data, 0)
# data preprocess dimensionless
if data_preprocess_method == 'dimensionless':
redict_data = data_preprocess.dimensionless(predict_data, 0)
# data preprocess 0-1
if data_preprocess_method == 'preprocess_01':
predict_data = data_preprocess.preprocess_01(predict_data, 0)
# data preprocess no month mean
if data_preprocess_method == 'nomonthmean':
predict_data = data_preprocess.no_month_mean(predict_data, 0)
if model_type == 'conv':
data_x[0] = predict_data[0]
else:
data_x[0] = np.reshape(predict_data[0], (1, 1080))
ssta_error = np.empty([directly_month])
for i in range(directly_month):
if is_seasonal_circle:
data_sc[0] = [(start_month + i) % 12]
data_x = model.predict([data_x, data_sc])
else:
data_x = model.predict(data_x)
if model_type == 'conv':
data_y[0] = data_x[0]
else:
data_y[0] = np.reshape(data_x[0], (20, 27, 2))
# data preprocess z-zero
if data_preprocess_method == 'preprocess_Z':
data_y = data_preprocess.preprocess_Z(data_y, 1)
# data preprocess dimensionless
if data_preprocess_method == 'dimensionless':
data_y = data_preprocess.dimensionless(data_y, 1)
# data preprocess 0-1
if data_preprocess_method == 'preprocess_01':
data_y = data_preprocess.preprocess_01(data_y, 1)
# data preprocess no month mean
if data_preprocess_method == 'nomonthmean':
data_y = data_preprocess.no_month_mean(data_y, 1)
data_realistic[0] = file_helper_unformatted.read_data_sstaha(
start_month + i + 1)
if is_retrain:
data_realistic = file_helper_unformatted.exchange_rows(
data_realistic)
# ssta_error[i] = calculate_error(data_y[0], data_realistic[0])
ssta_error[i] = math_tool.calculate_rmse(data_y[0], data_realistic[0])
print(ssta_error)
ssta_error2 = ssta_error.copy()
for i in range(1, directly_month):
ssta_error2[i] = (ssta_error[i] - ssta_error[i - 1]) / 10000
ssta_error2[0] = ssta_error2[0] / 10000
return ssta_error2
