def predict_y(x, params):
W1 = params["W1"]
b1 = params["b1"]
W2 = params["W2"]
b2 = params["b2"]
W3 = params["W3"]
b3 = params["b3"]
W4 = params["W4"]
b4 = params["b4"]
#测试用例样本数
test_m = x.shape[1]
prediction_Y = np.zeros((1, test_m))
#计算
A1, Z1 = model.l_layer_forward_model(x, W1, b1, acitvation="relu") #第1层
A2, Z2 = model.l_layer_forward_model(A1, W2, b2, acitvation="relu") #第2层
A3, Z3 = model.l_layer_forward_model(A2, W3, b3, acitvation="relu") #第3层
A4, Z4 = model.l_layer_forward_model(A3, W4, b4,
acitvation="sigmoid") #第4层
test_Y = A4
# print(test_Y.shape)
for i in range(test_Y.shape[1]):
if test_Y[0][i] > 0.5:
prediction_Y[0][i] = 1
else:
prediction_Y[0][i] = 0
return prediction_Y
def predict_y_with_load_params(x):
cur_filepath = os.path.basename(os.path.realpath(__file__))
cur_file, ext = os.path.splitext(cur_filepath)
filename = cur_file + ".h5"
f = h5py.File(filename, 'r')
params = {}
for k in f.keys():
params[k] = np.array(f[k][:])
W1 = params["W1"]
b1 = params["b1"]
W2 = params["W2"]
b2 = params["b2"]
W3 = params["W3"]
b3 = params["b3"]
W4 = params["W4"]
b4 = params["b4"]
#测试用例样本数
test_m = x.shape[1]
prediction_Y = np.zeros((1, test_m))
#计算
A1, Z1 = model.l_layer_forward_model(x, W1, b1, acitvation="relu") #第1层
A2, Z2 = model.l_layer_forward_model(A1, W2, b2, acitvation="relu") #第2层
A3, Z3 = model.l_layer_forward_model(A2, W3, b3, acitvation="relu") #第3层
A4, Z4 = model.l_layer_forward_model(A3, W4, b4,
acitvation="sigmoid") #第4层
test_Y = A4
# print(test_Y.shape)
for i in range(test_Y.shape[1]):
if test_Y[0][i] > 0.5:
prediction_Y[0][i] = 1
else:
prediction_Y[0][i] = 0
return prediction_Y
def predict_y(x, params):
W1 = params["W1"]
b1 = params["b1"]
W2 = params["W2"]
b2 = params["b2"]
#测试用例样本数
test_m = x.shape[1]
prediction_Y = np.zeros((1, test_m))
#计算
A1, Z1 = model.l_layer_forward_model(x, W1, b1, acitvation="relu") #第1层
A2, Z2 = model.l_layer_forward_model(A1, W2, b2,
acitvation="sigmoid") #第2层
test_Y = A2
# print(test_Y.shape)
for i in range(test_Y.shape[1]):
if test_Y[0][i] > 0.5:
prediction_Y[0][i] = 1
else:
prediction_Y[0][i] = 0
return prediction_Y
4, [20, 7, 5, 1],
init_method="He")
W1 = params["W1"]
b1 = params["b1"]
W2 = params["W2"]
b2 = params["b2"]
W3 = params["W3"]
b3 = params["b3"]
W4 = params["W4"]
b4 = params["b4"]
learning_rate = 0.0075
echo_num = 3000
for echo in range(echo_num):
#正向传播
A1, Z1 = model.l_layer_forward_model(train_x,
W1,
b1,
acitvation="relu") #第1层
A2, Z2 = model.l_layer_forward_model(A1, W2, b2,
acitvation="relu") #第2层
A3, Z3 = model.l_layer_forward_model(A2, W3, b3,
acitvation="relu") #第3层
A4, Z4 = model.l_layer_forward_model(A3, W4, b4,
acitvation="sigmoid") #第4层
#计算代价
J, dJ = model.cost_model(A4, train_y)
#反向传播
dA4_prew, dW4, db4 = model.l_layer_backward_model(
A3, W4, Z4, dJ, acitvation="sigmoid") #第4层
dA3_prew, dW3, db3 = model.l_layer_backward_model(
A2, W3, Z3, dA4_prew, acitvation="relu") #第3层
# print(train_X.shape)
test_x = test_x / 255
#参数初始化
params = model.params_init_model(train_X.shape[0],
4, [20, 7, 5, 1],
init_method="He",
seed=0)
v, s = model.init_adam(params)
learning_rate = 0.0075
# lambd=0.7
echo_num = 1000
grad = {}
for echo in range(echo_num):
#正向传播
A1, Z1 = model.l_layer_forward_model(train_X,
params["W1"],
params["b1"],
acitvation="relu") #第1层
A2, Z2 = model.l_layer_forward_model(A1,
params["W2"],
params["b2"],
acitvation="relu") #第2层
A3, Z3 = model.l_layer_forward_model(A2,
params["W3"],
params["b3"],
acitvation="relu") #第3层
A4, Z4 = model.l_layer_forward_model(A3,
params["W4"],
params["b4"],
acitvation="sigmoid") #第4层
#计算代价
train_x = train_x / 255
test_x = test_x / 255
#参数初始化
params = model.params_init_model(train_x.shape[0],
2, [5, 1],
init_method="He")
W1 = params["W1"]
b1 = params["b1"]
W2 = params["W2"]
b2 = params["b2"]
learning_rate = 0.01
echo_num = 5000
for echo in range(echo_num):
#正向传播
A1, Z1 = model.l_layer_forward_model(train_x,
W1,
b1,
acitvation="relu") #第1层
A2, Z2 = model.l_layer_forward_model(A1, W2, b2,
acitvation="sigmoid") #第2层
#计算代价
J, dJ = model.cost_model(A2, train_y)
#反向传播
dA2_prew, dW2, db2 = model.l_layer_backward_model(
A1, W2, Z2, dJ, acitvation="sigmoid") #第2层
dA1_prew, dW1, db1 = model.l_layer_backward_model(
train_x, W1, Z1, dA2_prew, acitvation="relu") #第1层
#更新参数权值
# print("w",W.shape)
# print("dw",dW.shape)
W1 = W1 - learning_rate * dW1
if __name__ == "__main__":
train_x, train_y, test_x, test_y, list_classes = load_dataset()
#归一化
train_x = train_x / 255
test_x = test_x / 255
#参数初始化
params = model.params_init_model(train_x.shape[0], 1, [1])
W1 = params["W1"]
b1 = params["b1"]
learning_rate = 0.001
echo_num = 1000
for echo in range(echo_num):
#正向传播
A1, Z1 = model.l_layer_forward_model(train_x,
W1,
b1,
acitvation="sigmoid") #第1层
#计算代价
J, dJ = model.cost_model(A1, train_y)
#反向传播
dA1_prew, dW1, db1 = model.l_layer_backward_model(
train_x, W1, Z1, dJ, acitvation="sigmoid") #第1层
#更新参数权值
# print("w",W.shape)
# print("dw",dW.shape)
W1 = W1 - learning_rate * dW1
b1 = b1 - learning_rate * db1
print("echo {0} cost:{1}".format(echo, J))
