def init_hidden_layers(self):
if self.num_of_hidden_layers != 0:
# First hidden layer
self.hidden_layers.append(
nn(self.hidden_layer_neurons_num[0], self.num_of_inputs))
for i in range(self.num_of_hidden_layers - 1):
# Other hidden layers
self.hidden_layers.append(
nn(self.hidden_layer_neurons_num[i + 1],
self.hidden_layer_neurons_num[i]))
def __init__(self, alpha=0.3, gamma=1):
"""
Args:
alpha: <float> learning rate (default = 0.3)
gamma: <float> value decay rate (default = 1)
n: <int> number of back steps to update (default = 1)
"""
self._counter = 0
# Possibility of taking a random action instead of the best one
self.epsilon = 0.30
# Store a neural network attribute to the agent for its Q-value predictions
self.gamma = gamma
self.nn = nn(gamma)
self.alpha = alpha
# Collections of the previous state's information
self.previous_life = 0.0
self.previous_possable_actions = []
self.previous_state = None
self.previous_action = None
self.previous_two_states = None
self.previous_closest_enemy = 0
self.previous_closest_wall = 0
self.action_map = {'left': 0, 'right': 1, 'up': 2, 'down': 3}
self.mse = []
self.predict_value = 0.0
def post(self):
args = parser.parse_args()
data = args['data']
data = eval(data)
dataset = data['value']
grey_model = gy() # 实例化灰色预测类
grey_result = grey_model.start(data=dataset) # 启动灰色预测进行预测
dataset = np.asarray(dataset) # ? 为什么不能 ndarray
error = dataset - grey_result # 得到残差
neuralnetwork_model = nn() # 神经网络进行实例化
correction_error, error_min, error_max = neuralnetwork_model.launch(
1, 10, dataset, error) # 启动神经网络残差补偿，得到最终预测结果
predict = dataset - (error * (error_max - error_min) + error_min)
my_dict = {
"message": "success",
"result": predict.tolist(),
"date": datetime.now()
}
return jsonify(my_dict)
def init_output_layer(self):
if self.num_of_hidden_layers != 0:
self.output_layer = nn(self.num_of_outputs,
self.hidden_layer_neurons_num[-1])
else:
self.output_layer = nn(self.num_of_outputs, self.num_of_inputs)
def orient(name, filename, model_file, model):
if name == 'train':
if model == 'nearest' or model == 'best':
train = pd.read_csv(filename, sep=' ', header=None)
filename_knn = model_file
file = open(filename_knn, 'wb')
pickle.dump(train, file)
if model == 'nnet':
train = pd.read_csv(filename, sep=' ', header=None)
x_train = train.drop(columns=[0, 1], axis=1)
y_train = train[1]
y_train = pd.get_dummies(y_train)
y_columns = y_train.columns
x_train = x_train.to_numpy()
y_train = y_train.to_numpy()
print(x_train.shape[0], 'train samples')
a = nn(25, 0.001, 0.9)
(w1, w2, w3, b1, b2, b3) = a.fit(x_train, y_train)
weights = {
'w1': w1,
'w2': w2,
'w3': w3,
'b1': b1,
'b2': b2,
'b3': b3,
'y_columns': y_columns
}
filename_nn = model_file
file = open(filename_nn, 'wb')
pickle.dump(weights, file)
if model == 'tree':
dtreemain(name, filename, model_file)
if name == 'test':
if model == 'nearest' or model == 'best':
file = open(model_file, 'rb')
train = pickle.load(file)
test = pd.read_csv(filename, sep=' ', header=None)
X_test = test.drop(columns=[0, 1], axis=1)
y_filenames = test[0]
y_test = test[1]
X_test = X_test.to_numpy()
y_test = y_test.to_numpy()
obj = knn(10)
ypred = obj.predict(train, X_test)
f = open("output.txt", "w")
for i in range(len(X_test)):
with open('output.txt', 'a') as f:
f.write(str(y_filenames[i]) + ' ' + str(ypred[i]) + '\n')
if model == 'tree':
dtreemain(name, filename, model_file)
if model == 'nnet':
test = pd.read_csv(filename, sep=' ', header=None)
x_test = test.drop(columns=[0, 1], axis=1)
y_test = test[1]
y_filenames = test[0]
y_test = pd.get_dummies(y_test)
x_test = x_test.to_numpy()
y_test = y_test.to_numpy()
print(x_test.shape[0], 'test samples')
file = open(model_file, 'rb')
new_weights = pickle.load(file)
w1f = new_weights['w1']
w2f = new_weights['w2']
w3f = new_weights['w3']
b1f = new_weights['b1']
b2f = new_weights['b2']
b3f = new_weights['b3']
y_columns = new_weights['y_columns']
a = nn(25, 0.001, 0.9)
y_test_predicted = a.predict(x_test, w1f, w2f, w3f, b1f, b2f, b3f)
zero_one = (y_test_predicted == y_test_predicted.max(
axis=1)[:, None]).astype(int)
diff = (y_test == zero_one).sum(axis=1)
accuracy = np.count_nonzero(diff == y_test.shape[1])
print('accuracy ', accuracy / diff.shape[0] * 100)
f = open("Output.txt", "w")
for i in range(len(x_test)):
with open('Output.txt', 'a') as f:
f.write(
str(y_filenames[i]) + ' ' +
str(y_columns[np.argmax(y_test_predicted[i])]) + '\n')