Skip to content

P
pytensor
提交 35254935 authored 作者: Frédéric Bastien's avatar Frédéric Bastien
浏览文件
操作

Merge pull request #4145 from myme5261314/master

Improve code misleading tiny issues for Loop of Tutorial.
上级 0babc678 54a79d7f
隐藏空白字符变更
内嵌 并排
正在显示 包含 34 行增加 和 34 行删除
doc/tutorial/loop.txt
浏览文件 @ 35254935
...@@ -38,7 +38,7 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`. ...@@ -38,7 +38,7 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`.
b_sym = T.vector("b_sym") b_sym = T.vector("b_sym")
results, updates = theano.scan(lambda v: T.tanh(T.dot(v, W) + b_sym), sequences=X) results, updates = theano.scan(lambda v: T.tanh(T.dot(v, W) + b_sym), sequences=X)
compute_elementwise = theano.function(inputs=[X, W, b_sym], outputs=[results]) compute_elementwise = theano.function(inputs=[X, W, b_sym], outputs=results)
# test values # test values
x = np.eye(2, dtype=theano.config.floatX) x = np.eye(2, dtype=theano.config.floatX)
...@@ -46,7 +46,7 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`. ...@@ -46,7 +46,7 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`.
b = np.ones((2), dtype=theano.config.floatX) b = np.ones((2), dtype=theano.config.floatX)
b[1] = 2 b[1] = 2
print(compute_elementwise(x, w, b)[0]) print(compute_elementwise(x, w, b))
# comparison with numpy # comparison with numpy
print(np.tanh(x.dot(w) + b)) print(np.tanh(x.dot(w) + b))
...@@ -77,7 +77,7 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`. ...@@ -77,7 +77,7 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`.
results, updates = theano.scan(lambda y, p, x_tm1: T.tanh(T.dot(x_tm1, W) + T.dot(y, U) + T.dot(p, V)), results, updates = theano.scan(lambda y, p, x_tm1: T.tanh(T.dot(x_tm1, W) + T.dot(y, U) + T.dot(p, V)),
sequences=[Y, P[::-1]], outputs_info=[X]) sequences=[Y, P[::-1]], outputs_info=[X])
compute_seq = theano.function(inputs=[X, W, Y, U, P, V], outputs=[results]) compute_seq = theano.function(inputs=[X, W, Y, U, P, V], outputs=results)
# test values # test values
x = np.zeros((2), dtype=theano.config.floatX) x = np.zeros((2), dtype=theano.config.floatX)
...@@ -90,7 +90,7 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`. ...@@ -90,7 +90,7 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`.
p[0, :] = 3 p[0, :] = 3
v = np.ones((2, 2), dtype=theano.config.floatX) v = np.ones((2, 2), dtype=theano.config.floatX)
print(compute_seq(x, w, y, u, p, v)[0]) print(compute_seq(x, w, y, u, p, v))
# comparison with numpy # comparison with numpy
x_res = np.zeros((5, 2), dtype=theano.config.floatX) x_res = np.zeros((5, 2), dtype=theano.config.floatX)
...@@ -123,11 +123,11 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`. ...@@ -123,11 +123,11 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`.
# define tensor variable # define tensor variable
X = T.matrix("X") X = T.matrix("X")
results, updates = theano.scan(lambda x_i: T.sqrt((x_i ** 2).sum()), sequences=[X]) results, updates = theano.scan(lambda x_i: T.sqrt((x_i ** 2).sum()), sequences=[X])
compute_norm_lines = theano.function(inputs=[X], outputs=[results]) compute_norm_lines = theano.function(inputs=[X], outputs=results)
# test value # test value
x = np.diag(np.arange(1, 6, dtype=theano.config.floatX), 1) x = np.diag(np.arange(1, 6, dtype=theano.config.floatX), 1)
print(compute_norm_lines(x)[0]) print(compute_norm_lines(x))
# comparison with numpy # comparison with numpy
print(np.sqrt((x ** 2).sum(1))) print(np.sqrt((x ** 2).sum(1)))
...@@ -148,11 +148,11 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`. ...@@ -148,11 +148,11 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`.
# define tensor variable # define tensor variable
X = T.matrix("X") X = T.matrix("X")
results, updates = theano.scan(lambda x_i: T.sqrt((x_i ** 2).sum()), sequences=[X.T]) results, updates = theano.scan(lambda x_i: T.sqrt((x_i ** 2).sum()), sequences=[X.T])
compute_norm_cols = theano.function(inputs=[X], outputs=[results]) compute_norm_cols = theano.function(inputs=[X], outputs=results)
# test value # test value
x = np.diag(np.arange(1, 6, dtype=theano.config.floatX), 1) x = np.diag(np.arange(1, 6, dtype=theano.config.floatX), 1)
print(compute_norm_cols(x)[0]) print(compute_norm_cols(x))
# comparison with numpy # comparison with numpy
print(np.sqrt((x ** 2).sum(0))) print(np.sqrt((x ** 2).sum(0)))
...@@ -177,12 +177,12 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`. ...@@ -177,12 +177,12 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`.
sequences=[T.arange(X.shape[0]), T.arange(X.shape[1])], sequences=[T.arange(X.shape[0]), T.arange(X.shape[1])],
outputs_info=np.asarray(0., dtype=floatX)) outputs_info=np.asarray(0., dtype=floatX))
result = results[-1] result = results[-1]
compute_trace = theano.function(inputs=[X], outputs=[result]) compute_trace = theano.function(inputs=[X], outputs=result)
# test value # test value
x = np.eye(5, dtype=theano.config.floatX) x = np.eye(5, dtype=theano.config.floatX)
x[0] = np.arange(5, dtype=theano.config.floatX) x[0] = np.arange(5, dtype=theano.config.floatX)
print(compute_trace(x)[0]) print(compute_trace(x))
# comparison with numpy # comparison with numpy
print(np.diagonal(x).sum()) print(np.diagonal(x).sum())
...@@ -211,7 +211,7 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`. ...@@ -211,7 +211,7 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`.
results, updates = theano.scan(lambda x_tm2, x_tm1: T.dot(x_tm2, U) + T.dot(x_tm1, V) + T.tanh(T.dot(x_tm1, W) + b_sym), results, updates = theano.scan(lambda x_tm2, x_tm1: T.dot(x_tm2, U) + T.dot(x_tm1, V) + T.tanh(T.dot(x_tm1, W) + b_sym),
n_steps=n_sym, outputs_info=[dict(initial=X, taps=[-2, -1])]) n_steps=n_sym, outputs_info=[dict(initial=X, taps=[-2, -1])])
compute_seq2 = theano.function(inputs=[X, U, V, W, b_sym, n_sym], outputs=[results]) compute_seq2 = theano.function(inputs=[X, U, V, W, b_sym, n_sym], outputs=results)
# test values # test values
x = np.zeros((2, 2), dtype=theano.config.floatX) # the initial value must be able to return x[-2] x = np.zeros((2, 2), dtype=theano.config.floatX) # the initial value must be able to return x[-2]
...@@ -236,16 +236,16 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`. ...@@ -236,16 +236,16 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`.
.. testoutput:: .. testoutput::
[array([[ 1.40514825, 1.40514825], [[ 1.40514825 1.40514825]
[ 2.88898899, 2.38898899], [ 2.88898899 2.38898899]
[ 4.34018291, 4.34018291], [ 4.34018291 4.34018291]
[ 6.53463142, 6.78463142], [ 6.53463142 6.78463142]
[ 9.82972243, 9.82972243], [ 9.82972243 9.82972243]
[ 14.22203814, 14.09703814], [ 14.22203814 14.09703814]
[ 20.07439936, 20.07439936], [ 20.07439936 20.07439936]
[ 28.12291843, 28.18541843], [ 28.12291843 28.18541843]
[ 39.1913681 , 39.1913681 ], [ 39.1913681 39.1913681 ]
[ 54.28407732, 54.25282732]])] [ 54.28407732 54.25282732]]
[[ 1.40514825 1.40514825] [[ 1.40514825 1.40514825]
[ 2.88898899 2.38898899] [ 2.88898899 2.38898899]
[ 4.34018291 4.34018291] [ 4.34018291 4.34018291]
...@@ -271,13 +271,13 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`. ...@@ -271,13 +271,13 @@ The full documentation can be found in the library: :ref:`Scan <lib_scan>`.
A = T.matrix() A = T.matrix()
y = T.tanh(T.dot(v, A)) y = T.tanh(T.dot(v, A))
results, updates = theano.scan(lambda i: T.grad(y[i], v), sequences=[T.arange(y.shape[0])]) results, updates = theano.scan(lambda i: T.grad(y[i], v), sequences=[T.arange(y.shape[0])])
compute_jac_t = theano.function([A, v], [results], allow_input_downcast=True) # shape (d_out, d_in) compute_jac_t = theano.function([A, v], results, allow_input_downcast=True) # shape (d_out, d_in)
# test values # test values
x = np.eye(5, dtype=theano.config.floatX)[0] x = np.eye(5, dtype=theano.config.floatX)[0]
w = np.eye(5, 3, dtype=theano.config.floatX) w = np.eye(5, 3, dtype=theano.config.floatX)
w[2] = np.ones((3), dtype=theano.config.floatX) w[2] = np.ones((3), dtype=theano.config.floatX)
print(compute_jac_t(w, x)[0]) print(compute_jac_t(w, x))
# compare with numpy # compare with numpy
print(((1 - np.tanh(x.dot(w)) ** 2) * w).T) print(((1 - np.tanh(x.dot(w)) ** 2) * w).T)
...@@ -330,7 +330,7 @@ Note that we need to iterate over the indices of ``y`` and not over the elements ...@@ -330,7 +330,7 @@ Note that we need to iterate over the indices of ``y`` and not over the elements
d=trng.binomial(size=W[1].shape) d=trng.binomial(size=W[1].shape)
results, updates = theano.scan(lambda v: T.tanh(T.dot(v, W) + b_sym) * d, sequences=X) results, updates = theano.scan(lambda v: T.tanh(T.dot(v, W) + b_sym) * d, sequences=X)
compute_with_bnoise = theano.function(inputs=[X, W, b_sym], outputs=[results], compute_with_bnoise = theano.function(inputs=[X, W, b_sym], outputs=results,
updates=updates, allow_input_downcast=True) updates=updates, allow_input_downcast=True)
x = np.eye(10, 2, dtype=theano.config.floatX) x = np.eye(10, 2, dtype=theano.config.floatX)
w = np.ones((2, 2), dtype=theano.config.floatX) w = np.ones((2, 2), dtype=theano.config.floatX)
...@@ -340,16 +340,16 @@ Note that we need to iterate over the indices of ``y`` and not over the elements ...@@ -340,16 +340,16 @@ Note that we need to iterate over the indices of ``y`` and not over the elements
.. testoutput:: .. testoutput::
[array([[ 0.96402758, 0. ], [[ 0.96402758 0. ]
[ 0. , 0.96402758], [ 0. 0.96402758]
[ 0. , 0. ], [ 0. 0. ]
[ 0.76159416, 0.76159416], [ 0.76159416 0.76159416]
[ 0.76159416, 0. ], [ 0.76159416 0. ]
[ 0. , 0.76159416], [ 0. 0.76159416]
[ 0. , 0.76159416], [ 0. 0.76159416]
[ 0. , 0.76159416], [ 0. 0.76159416]
[ 0. , 0. ], [ 0. 0. ]
[ 0.76159416, 0.76159416]])] [ 0.76159416 0.76159416]]
Note that if you want to use a random variable ``d`` that will not be updated through scan loops, you should pass this variable as a ``non_sequences`` arguments. Note that if you want to use a random variable ``d`` that will not be updated through scan loops, you should pass this variable as a ``non_sequences`` arguments.
......
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论