本文整理汇总了Python中calculator.Calculator.normal方法的典型用法代码示例。如果您正苦于以下问题：Python Calculator.normal方法的具体用法？Python Calculator.normal怎么用？Python Calculator.normal使用的例子？那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类calculator.Calculator的用法示例。
在下文中一共展示了Calculator.normal方法的1个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: Model
# 需要导入模块: from calculator import Calculator [as 别名]
# 或者: from calculator.Calculator import normal [as 别名]
class Model(object):
def __init__(self, I, K):
# num(X), dim(X), dim(Z), dim(Y)
self.I = I
self.H = None
self.K = K
self.best_w = None
self.w = None
self.mean = None
self.var = None
self.layer = 1
self.hidden_type = HIDDEN_TYPE["DIRECT"]
self.act_type = ACTIVATION_TYPE["SOFTMAX"]
self.indices = None
# Optimizer
self.optimizer = None
self.logger = None
self.calc = Calculator()
def set_hidden(self, opt=HIDDEN_TYPE["DIRECT"]):
self.hidden_type = opt
self.H = HIDDEN_NUM[opt]
self.w = np.zeros((self.H,self.K))
def set_activation(self, opt=ACTIVATION_TYPE["SOFTMAX"]):
self.act_type = opt
def set_optimizer(self, optimizer):
self.optimizer = optimizer
self.layer = optimizer.layer
self.set_logger(optimizer.loss, optimizer.lr, optimizer.batch_size)
def set_logger(self, loss, lr, batch_size):
self.logger = Logger(loss, lr, batch_size)
def learn(self, data_x, data_y, time_limit):
N = data_x.shape[0]
self.logger.start_learn()
# init 1st layer : Mean, Variance
self.mean, self.var = self.calc.normal_parameters(self, data_x, data_y)
# init 2nd layer : Weights (H+1)*K
self.w = self.optimizer.init_weights(self, WEIGHT_INIT["UNIFORM"])
min_error = float("inf")
batch_amount = int(10000*4/5.)/self.optimizer.batch_size
t_errs = np.zeros(batch_amount)
t_accs = np.zeros(batch_amount)
v_errs = np.zeros(batch_amount)
v_accs = np.zeros(batch_amount)
for epoch in range(self.optimizer.max_epoch):
idx_t, idx_v = self.cross_validation(epoch, N)
for b_cnt in range(batch_amount):
batch = self.get_batch(b_cnt, idx_t)
x = data_x[batch]
y = data_y[batch]
vx = data_x[idx_v]
vy = data_y[idx_v]
# Forward Propagation
z = self.calc.normal(self, x) # batch * (H+1)
a = z.dot(self.w) # batch * K
f = self.calc.nonlinear(self, a)
vf = self.predict(vx, self.w)
t_errs[b_cnt], t_accs[b_cnt], v_errs[b_cnt], v_accs[b_cnt] = \
self.evaluate(f, y, vf, vy)
# Backward Propagation > grads = [dW, dMean, dVariance]
grads = self.optimizer.get_gradients(self, x, z, f, y)
self.w += grads[0]
self.mean += grads[1]
self.var += grads[2]
if v_errs.mean() < min_error:
min_error = v_errs.mean()
self.best_w = self.w
# Logging
self.optimizer.update(epoch, batch_amount)
self.logger.update(t_errs.mean(), t_accs.mean(), v_errs.mean(), v_accs.mean())
if self.logger.is_finish(time_limit):
break
# self.logger.save_log(self, vf, vy)
best_e = np.array(self.logger.v_acc).argmax()
return self.logger.t_err[best_e], self.logger.t_acc[best_e], self.logger.v_err[best_e], self.logger.v_acc[best_e]
def predict(self, x, w=None):
if w is None:
w = self.best_w
#.........这里部分代码省略.........