self.mean_squared_error = []
def on_epoch_end(self, batch, logs={}):
self.losses.append(logs.get('loss'))
self.mean_squared_error.append(logs.get('mean_squared_error'))
history = LossHistory()
callbacks_list = [checkpoint, history]
#callbacks_list = [checkpoint]
###########################
### Start Training the network
###########################
subFields = lnn.loadBaseFNames(mapLoc)
base = [mapLoc + s for s in subFields]
dataset = tf.data.Dataset.from_tensor_slices(tf.convert_to_tensor(base))
dataset = dataset.shuffle(buffer_size=len(base))
dataset = dataset.map(lambda item: tuple(
tf.py_func(lnn.utf8FileToMapAndLum, [item, 'basic', True],
[tf.float64, tf.float64])))
dataset = dataset.repeat()
dataset = dataset.batch(batch_size)
history = model2.fit(dataset,
epochs=epochs,
steps_per_epoch=steps_per_epoch,
callbacks=callbacks_list,
verbose=1)
# where to store things data_loc = '../data/' predict_loc = data_loc + 'predictions/' # load in the config file config = configparser.ConfigParser() config.read(data_loc + config_name + '.ini') models = config.sections() # defaults defaults = lnn.get_config_info(config, models[0]) # start the work to load the validation maps cur_map_loc = config[models[0]]['map_loc'] subFields = lnn.loadBaseFNames(cur_map_loc) # set random seed so data is shuffeled the same way # every time and then make the seed random np.random.seed(1234) np.random.shuffle(subFields) np.random.seed() # get map locations valPer = 0.2 valPoint = int(len(subFields) * (1 - valPer)) base_val = [cur_map_loc + s for s in subFields[valPoint:]] # debug # debug_map_numbs = 2 # base_val = base_val[:debug_map_numbs]
