2

我是 Theano/RNN 编程的初学者，我根据这里的代码做了第一个简单的实现

在我的代码中，我创建了一个简单的自回归过程作为数据集，其中输出系列的每一项都是输入系列的前 4 个值的总和（正态分布值）

完整的代码是

import numpy as np
import theano
import theano.tensor as TT
#theano.config.exception_verbosity = "high"
import matplotlib.pyplot as plt

SERIES_LEN = 1000;
REGR = 4;
LEARN_RATE = 0.01;
x_train = np.random.randn(SERIES_LEN)
y_train = np.asarray([sum(x_train[range(max(i-REGR,0),i+1)]) for i in range(SERIES_LEN)])
mval = max(y_train)
y_train /= mval
x_train /= mval

x_test = np.random.randn(SERIES_LEN)
y_test = np.asarray([sum(x_test[range(max(i-REGR,0),i+1)]) for i in range(SERIES_LEN)])
mval = max(y_test)
y_test /= mval
x_test /= mval


# number of hidden units
n = 10
# number of input units
nin = 1
# number of output units
nout = 1

# input (where first dimension is time)
u = TT.matrix()
# target (where first dimension is time)
t = TT.matrix()
# initial hidden state of the RNN
h0 = TT.vector()
# learning rate
lr = TT.scalar()
# recurrent weights as a shared variable
W = theano.shared(np.random.uniform(size=(n, n), low=-.01, high=.01))
# input to hidden layer weights
W_in = theano.shared(np.random.uniform(size=(nin, n), low=-.01, high=.01))
# hidden to output layer weights
W_out = theano.shared(np.random.uniform(size=(n, nout), low=-.01, high=.01))


# recurrent function (using tanh activation function) and linear output
# activation function
def step(u_t, h_tm1, W, W_in, W_out):
    h_t = TT.tanh(TT.dot(u_t, W_in) + TT.dot(h_tm1, W))
    y_t = TT.dot(h_t, W_out)
    return h_t, y_t

# the hidden state `h` for the entire sequence, and the output for the
# entrie sequence `y` (first dimension is always time)
[h, y], _ = theano.scan(step,
                        sequences=u,
                        outputs_info=[dict(initial=TT.zeros(n)), None],
                        non_sequences=[W, W_in, W_out],
                        truncate_gradient=REGR)
# error between output and target
error = ((y - t) ** 2).sum()
# gradients on the weights using BPTT
gW, gW_in, gW_out = TT.grad(error, [W, W_in, W_out])
# training function, that computes the error and updates the weights using
# SGD.
fn = theano.function([u, t, lr],
                     error,
                     updates=((W, W - lr * gW),
                              (W_in, W_in - lr * gW_in),
                              (W_out, W_out - lr * gW_out)))


fn(np.atleast_2d(x_train).T, np.atleast_2d(y_train).T, LEARN_RATE)

eval_output = theano.function([u],y)
pred_out = eval_output(x_test)

print pred_out
plt.plot(pred_out)
plt.plot(y_test)
plt.show()

对链接页面中定义的基本行为几乎没有修改。我收到一个奇怪的错误，阻止程序在该行之后运行进一步的指令

fn(np.atleast_2d(x_train).T, np.atleast_2d(y_train).T, LEARN_RATE)

我得到的错误是

/Library/Python/2.7/site-packages/theano/scan_module/scan_perform_ext.py:133：RuntimeWarning：numpy.ndarray 大小已更改，可能表示 scan_perform.scan_perform 导入的二进制不兼容 * BLAS 错误：传递给 cblas_dgemv 的参数 incX 为 0，这是无效的。

我真的不明白如何解决这个问题，也没有发现这个错误

4

0 回答 0