if (os.path.exists(model_save_path + "\checkpoint")):
Global_Model.load_model_weights()
step_model.load_model_weights()
for env in envs:
workers.append(Worker(step_model, env, batch_size=batch_size, render=False))
print("Model restored.")
else:
for env in envs:
workers.append(Worker(step_model, env, batch_size=batch_size, render=False))
print("Creating new model.")
except:
print("ERROR: There was an issue loading the model!")
raise
coordinator = Coordinator(Global_Model, step_model, workers, plot, num_envs, num_epocs, batches_per_epoch, batch_size, gamma, model_save_path, anneling_steps)
# Train and save
if coordinator.run():
try:
Global_Model.save_model_weights()
print("Model saved.")
print("Now testing results....")
except:
print("ERROR: There was an issue saving the model!")
def start(self):
workers = []
network_params = (self.NUM_STATE,
self._config['Max steps taken per batch'],
self.NUM_ACTIONS, self.ACTION_SPACE)
# Init Global and Local networks. Generate Weights for them as well.
if self._config['CNN type'] == '':
self._global_model = AC_Model_Large(self.NUM_STATE,
self.NUM_ACTIONS,
self._config,
is_training=True)
self._global_model(
tf.convert_to_tensor(np.random.random((1, *self.NUM_STATE)),
dtype='float64'))
self._step_model = AC_Model_Large(self.NUM_STATE,
self.NUM_ACTIONS,
self._config,
is_training=True)
self._step_model(
tf.convert_to_tensor(np.random.random((1, *self.NUM_STATE)),
dtype='float64'))
else:
self._global_model = CNN_class_import(
self._config['CNN type'],
(self.NUM_STATE, self.NUM_ACTIONS, self._config, True))
self._global_model(
tf.convert_to_tensor(np.random.random((1, *self.NUM_STATE)),
dtype='float64'))
self._step_model = CNN_class_import(
self._config['CNN type'],
(self.NUM_STATE, self.NUM_ACTIONS, self._config, True))
self._step_model(
tf.convert_to_tensor(np.random.random((1, *self.NUM_STATE)),
dtype='float64'))
# Load model if exists
if not os.path.exists(self._model_save_path):
os.makedirs(self._model_save_path)
else:
try:
if (os.path.exists(self._model_save_path + "\checkpoint")):
self._global_model.load_model_weights()
self._step_model.load_model_weights()
for env in self._envs:
workers.append(
Worker(self._step_model,
env,
batch_size=self.
_config['Max steps taken per batch'],
render=False))
print("Model restored.")
else:
for env in self._envs:
workers.append(
Worker(self._step_model,
env,
batch_size=self.
_config['Max steps taken per batch'],
render=False))
print("Creating new model.")
except:
print("ERROR: There was an issue loading the model!")
raise
coordinator = Coordinator(self._global_model, self._step_model,
workers, self._plot, self._model_save_path,
self._config)
# Train and save
try:
if coordinator.run():
try:
self._global_model.save_model_weights()
print("Model saved.")
return True
except:
print("ERROR: There was an issue saving the model!")
raise
except:
print("ERROR: There was an issues during training!")
raise
with open(path, action, encoding="utf-8") as f: # open the file to write
f.write(data)
# function to pass dictionary to text
def dictToStr(dict):
strDict = ''
for k, v in dict.items():
strDict += f'{str(k)} -> {str(v)}\n'
return strDict
# Join dataSets and normalize
coord = Coordinator()
# joinedDS = coord.join(pathA, pathB)
# normalize joined Dataset
initPf = pd.read_csv(initPath, delimiter=';', encoding="ISO-8859-1")
initPf[initPf < 0] = 0
# initPf[(np.abs(stats.zscore(initPf)) < 3).all(axis=1)]
normPf = coord.normalize(initPf)
print('normPf ', str(normPf))
writeToFile(str(normPf), initNormed, 'w')
# # Run Spectral clustering for k 2 to k 10
labelsList = coord.runConfig(normPf.head(100))
# Run best k spectral clustering
resultDict = coord.run(normPf.head(100), 3, labelsList)
strResDict = dictToStr(resultDict)
writeToFile(strResDict, resultsPath, 'w')
