Skip to content

P
pytensor
提交 60ad703d authored 作者: lamblin's avatar lamblin
浏览文件
操作

Merge pull request #1106 from nouiz/err_msg2

Add utt.assert_allclose that raise a more informative error.
上级 b7294365 96931fd1
隐藏空白字符变更
内嵌 并排
正在显示 包含 218 行增加 和 166 行删除
doc/tutorial/extending_theano.txt
浏览文件 @ 60ad703d
...@@ -211,8 +211,14 @@ returns the right answer. If you detect an error, you must raise an ...@@ -211,8 +211,14 @@ returns the right answer. If you detect an error, you must raise an
inp = numpy.asarray(numpy.random.rand(5, 4), dtype=config.floatX) inp = numpy.asarray(numpy.random.rand(5, 4), dtype=config.floatX)
out = f(inp) out = f(inp)
# Compare the result computed to the expected value. # Compare the result computed to the expected value.
assert numpy.allclose(inp * 2, out) utt.assert_allclose(inp * 2, out)
We call ``utt.assert_allclose(expected_value, value)`` to compare
NumPy ndarray.This raise an error message with more information. Also,
the default tolerance can be changed with the Theano flags
``config.tensor.cmp_sloppy`` that take values in 0, 1 and 2. The
defaul value do the most strict comparison, 1 and 2 make less strict
comparison.
Testing the infer_shape Testing the infer_shape
----------------------- -----------------------
......
theano/compile/debugmode.py
浏览文件 @ 60ad703d
...@@ -171,7 +171,8 @@ class BadThunkOutput(DebugModeError): ...@@ -171,7 +171,8 @@ class BadThunkOutput(DebugModeError):
of the exception""" of the exception"""
sio = StringIO() sio = StringIO()
print >> sio, "BadThunkOutput" print >> sio, "BadThunkOutput"
print >> sio, " variable :", self.r print >> sio, " Apply :", self.r.owner
print >> sio, " op :", self.offending_op()
print >> sio, " Outputs Type:", self.r.type print >> sio, " Outputs Type:", self.r.type
print >> sio, " Outputs Shape:", getattr(self.val1, 'shape', None) print >> sio, " Outputs Shape:", getattr(self.val1, 'shape', None)
print >> sio, " Outputs Strides:", getattr(self.val1, 'strides', None) print >> sio, " Outputs Strides:", getattr(self.val1, 'strides', None)
...@@ -180,60 +181,15 @@ class BadThunkOutput(DebugModeError): ...@@ -180,60 +181,15 @@ class BadThunkOutput(DebugModeError):
for val in self.inputs_val] for val in self.inputs_val]
print >> sio, " Inputs Strides:", [getattr(val, 'strides', None) print >> sio, " Inputs Strides:", [getattr(val, 'strides', None)
for val in self.inputs_val] for val in self.inputs_val]
print >> sio, " Apply :", self.r.owner print >> sio, " Bad Variable:", self.r
print >> sio, " thunk1 :", self.thunk1 print >> sio, " thunk1 :", self.thunk1
print >> sio, " thunk2 :", self.thunk2 print >> sio, " thunk2 :", self.thunk2
print >> sio, " val1 :", self.val1
print >> sio, " val2 :", self.val2 #Don't import it at the top of the file to prevent circular import.
print >> sio, " op :", self.offending_op() utt = theano.tests.unittest_tools
try: print >> sio, utt.str_diagnostic(self.val1, self.val2, None, None)
ssio = StringIO() ret = sio.getvalue()
print >> ssio, " Value 1 : shape, dtype, strides, min, max, n_inf, n_nan:", return ret
print >> ssio, self.val1.shape,
print >> ssio, self.val1.dtype,
print >> ssio, self.val1.strides,
print >> ssio, self.val1.min(),
print >> ssio, self.val1.max(),
print >> ssio, numpy.isinf(self.val1).sum(),
print >> ssio, numpy.isnan(self.val1).sum(),
# only if all succeeds to we add anything to sio
print >> sio, ssio.getvalue()
except Exception:
pass
try:
ssio = StringIO()
print >> ssio, " Value 2 : shape, dtype, strides, min, max, n_inf, n_nan:",
print >> ssio, self.val2.shape,
print >> ssio, self.val2.dtype,
print >> ssio, self.val2.strides,
print >> ssio, self.val2.min(),
print >> ssio, self.val2.max(),
print >> ssio, numpy.isinf(self.val2).sum(),
print >> ssio, numpy.isnan(self.val2).sum(),
# only if all succeeds to we add anything to sio
print >> sio, ssio.getvalue()
except Exception:
pass
try:
ov = numpy.asarray(self.val1)
nv = numpy.asarray(self.val2)
ssio = StringIO()
absdiff = numpy.absolute(nv - ov)
print >> ssio, " Max Abs Diff: ", numpy.max(absdiff)
print >> ssio, " Mean Abs Diff: ", numpy.mean(absdiff)
print >> ssio, " Median Abs Diff: ", numpy.median(absdiff)
print >> ssio, " Std Abs Diff: ", numpy.std(absdiff)
reldiff = numpy.absolute(nv - ov) / (numpy.absolute(nv) +
numpy.absolute(ov))
print >> ssio, " Max Rel Diff: ", numpy.max(reldiff)
print >> ssio, " Mean Rel Diff: ", numpy.mean(reldiff)
print >> ssio, " Median Rel Diff: ", numpy.median(reldiff)
print >> ssio, " Std Rel Diff: ", numpy.std(reldiff)
# only if all succeeds to we add anything to sio
print >> sio, ssio.getvalue()
except Exception:
pass
return sio.getvalue()
class BadOptimization(DebugModeError): class BadOptimization(DebugModeError):
......
theano/scan_module/tests/test_scan.py
浏览文件 @ 60ad703d
...@@ -258,7 +258,7 @@ class T_Scan(unittest.TestCase): ...@@ -258,7 +258,7 @@ class T_Scan(unittest.TestCase):
numpy_values = numpy.array([state * (2 ** (k + 1)) for k numpy_values = numpy.array([state * (2 ** (k + 1)) for k
in xrange(steps)]) in xrange(steps)])
theano_values = my_f(state, steps) theano_values = my_f(state, steps)
assert numpy.allclose(numpy_values, theano_values) utt.assert_allclose(numpy_values, theano_values)
# generator network, only one output , type scalar ; no sequence or # generator network, only one output , type scalar ; no sequence or
# non sequence arguments # non sequence arguments
...@@ -287,7 +287,7 @@ class T_Scan(unittest.TestCase): ...@@ -287,7 +287,7 @@ class T_Scan(unittest.TestCase):
numpy_values = numpy.array([state * (2 ** (k + 1)) for k numpy_values = numpy.array([state * (2 ** (k + 1)) for k
in xrange(steps)]) in xrange(steps)])
theano_values = my_f(state, steps) theano_values = my_f(state, steps)
assert numpy.allclose(numpy_values, theano_values) utt.assert_allclose(numpy_values, theano_values)
def test_subtensor_multiple_slices(self): def test_subtensor_multiple_slices(self):
# This addresses a bug reported by Matthias Zoehrer # This addresses a bug reported by Matthias Zoehrer
...@@ -359,7 +359,7 @@ class T_Scan(unittest.TestCase): ...@@ -359,7 +359,7 @@ class T_Scan(unittest.TestCase):
for step in xrange(1, 4): for step in xrange(1, 4):
v_out[step] = v_u[step] * W_in + v_out[step - 1] * W v_out[step] = v_u[step] * W_in + v_out[step - 1] * W
theano_values = f2(v_u, v_x0, W_in, W) theano_values = f2(v_u, v_x0, W_in, W)
assert numpy.allclose(theano_values, v_out) utt.assert_allclose(theano_values, v_out)
# as test_one_sequence_one_output_weights, but on the gpu # as test_one_sequence_one_output_weights, but on the gpu
# This first version test the first case in the optimizer to the gpu. # This first version test the first case in the optimizer to the gpu.
...@@ -413,8 +413,7 @@ class T_Scan(unittest.TestCase): ...@@ -413,8 +413,7 @@ class T_Scan(unittest.TestCase):
for step in xrange(1, 4): for step in xrange(1, 4):
v_out[step] = v_u[step] * W_in + v_out[step - 1] * W v_out[step] = v_u[step] * W_in + v_out[step - 1] * W
theano_values = f2(v_u, v_x0, W_in, W) theano_values = f2(v_u, v_x0, W_in, W)
assert numpy.allclose(theano_values, v_out), (theano_values, v_out, utt.assert_allclose(theano_values, v_out)
theano_values - v_out)
# TO DEL # TO DEL
topo = f2.maker.fgraph.toposort() topo = f2.maker.fgraph.toposort()
...@@ -484,7 +483,7 @@ class T_Scan(unittest.TestCase): ...@@ -484,7 +483,7 @@ class T_Scan(unittest.TestCase):
for step in xrange(1, 4): for step in xrange(1, 4):
v_out[step] = v_u[step] * W_in + v_out[step - 1] * W v_out[step] = v_u[step] * W_in + v_out[step - 1] * W
theano_values = f2(v_u, v_x0, W_in, W) theano_values = f2(v_u, v_x0, W_in, W)
assert numpy.allclose(theano_values, v_out) utt.assert_allclose(theano_values, v_out)
topo = f2.maker.fgraph.toposort() topo = f2.maker.fgraph.toposort()
assert sum([isinstance(node.op, theano.sandbox.cuda.HostFromGpu) assert sum([isinstance(node.op, theano.sandbox.cuda.HostFromGpu)
...@@ -553,8 +552,8 @@ class T_Scan(unittest.TestCase): ...@@ -553,8 +552,8 @@ class T_Scan(unittest.TestCase):
v_out2[step] = numpy.int64(v_u[step] + v_out1[step - 1]) v_out2[step] = numpy.int64(v_u[step] + v_out1[step - 1])
theano_out1, theano_out2 = f2(v_u, v_x0, W_in, W) theano_out1, theano_out2 = f2(v_u, v_x0, W_in, W)
assert numpy.allclose(theano_out1, v_out1) utt.assert_allclose(theano_out1, v_out1)
assert numpy.allclose(theano_out2, v_out2) utt.assert_allclose(theano_out2, v_out2)
topo = f2.maker.fgraph.toposort() topo = f2.maker.fgraph.toposort()
scan_node = [node for node in topo scan_node = [node for node in topo
...@@ -651,8 +650,8 @@ class T_Scan(unittest.TestCase): ...@@ -651,8 +650,8 @@ class T_Scan(unittest.TestCase):
v_y[i] = numpy.dot(v_x[i - 1], vWout) v_y[i] = numpy.dot(v_x[i - 1], vWout)
(theano_x, theano_y) = f4(v_u1, v_u2, v_x0, v_y0, vW_in1) (theano_x, theano_y) = f4(v_u1, v_u2, v_x0, v_y0, vW_in1)
assert numpy.allclose(theano_x, v_x), (theano_x, v_x, theano_x - v_x) utt.assert_allclose(theano_x, v_x)
assert numpy.allclose(theano_y, v_y), (theano_y, v_y, theano_y - v_y) utt.assert_allclose(theano_y, v_y)
def test_multiple_outs_taps(self): def test_multiple_outs_taps(self):
l = 5 l = 5
...@@ -797,7 +796,7 @@ class T_Scan(unittest.TestCase): ...@@ -797,7 +796,7 @@ class T_Scan(unittest.TestCase):
numpy_out = numpy.zeros((2,)) numpy_out = numpy.zeros((2,))
numpy_out[0] = vu[0] * vW_in + vx0[1] * vW + vx0[0] numpy_out[0] = vu[0] * vW_in + vx0[1] * vW + vx0[0]
numpy_out[1] = vu[1] * vW_in + numpy_out[0] * vW + vx0[1] numpy_out[1] = vu[1] * vW_in + numpy_out[0] * vW + vx0[1]
assert numpy.allclose(numpy_out, theano_out) utt.assert_allclose(numpy_out, theano_out)
# simple rnn, one input, one state, weights for each; input/state are # simple rnn, one input, one state, weights for each; input/state are
# vectors, weights are scalars; using shared variables and past # vectors, weights are scalars; using shared variables and past
...@@ -836,7 +835,7 @@ class T_Scan(unittest.TestCase): ...@@ -836,7 +835,7 @@ class T_Scan(unittest.TestCase):
# and vx0[0] as vx0[-2], vx0[1] as vx0[-1] # and vx0[0] as vx0[-2], vx0[1] as vx0[-1]
numpy_out[0] = (vu[0] + vu[4]) * vW_in + vx0[1] * vW + vx0[0] numpy_out[0] = (vu[0] + vu[4]) * vW_in + vx0[1] * vW + vx0[0]
numpy_out[1] = (vu[1] + vu[5]) * vW_in + numpy_out[0] * vW + vx0[1] numpy_out[1] = (vu[1] + vu[5]) * vW_in + numpy_out[0] * vW + vx0[1]
assert numpy.allclose(numpy_out, theano_out) utt.assert_allclose(numpy_out, theano_out)
# simple rnn ; compute inplace version 1 # simple rnn ; compute inplace version 1
def test_inplace1(self): def test_inplace1(self):
...@@ -899,18 +898,16 @@ class T_Scan(unittest.TestCase): ...@@ -899,18 +898,16 @@ class T_Scan(unittest.TestCase):
# equivalent is done # equivalent is done
(theano_x0, theano_x1) = f9(vu0, vu1, vu2, vx0, vx1) (theano_x0, theano_x1) = f9(vu0, vu1, vu2, vx0, vx1)
# assert that theano does what it should # assert that theano does what it should
assert numpy.allclose(theano_x0, numpy_x0), (theano_x0, numpy_x0, utt.assert_allclose(theano_x0, numpy_x0)
theano_x0 - numpy_x0) utt.assert_allclose(theano_x1, numpy_x1)
assert numpy.allclose(theano_x1, numpy_x1), (theano_x1, numpy_x1,
theano_x1 - numpy_x1)
# assert that it was done in place # assert that it was done in place
# !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
# Old way of doing inplace operations is deprecated .. tests don't # Old way of doing inplace operations is deprecated .. tests don't
# make sense anymore. # make sense anymore.
##assert numpy.allclose( theano_x0 , vu2) ##utt.assert_allclose( theano_x0 , vu2)
## assert numpy.allclose( theano_x1 , vu1) ## utt.assert_allclose( theano_x1 , vu1)
# simple rnn ; compute inplace version 2 # simple rnn ; compute inplace version 2
def test_inplace2(self): def test_inplace2(self):
...@@ -981,8 +978,8 @@ class T_Scan(unittest.TestCase): ...@@ -981,8 +978,8 @@ class T_Scan(unittest.TestCase):
# equivalent is done # equivalent is done
(theano_x0, theano_x1) = f9(vu0, vu1, vu2, vx0, vx1) (theano_x0, theano_x1) = f9(vu0, vu1, vu2, vx0, vx1)
# assert that theano does what it should # assert that theano does what it should
assert numpy.allclose(theano_x0, numpy_x0), (theano_x0, numpy_x0) utt.assert_allclose(theano_x0, numpy_x0)
assert numpy.allclose(theano_x1, numpy_x1), (theano_x1, numpy_x1) utt.assert_allclose(theano_x1, numpy_x1)
# assert that it was done in place # assert that it was done in place
# not that x0 should not be inplace of vu2 because you are using # not that x0 should not be inplace of vu2 because you are using
# past values of u2, and therefore you are not allowed to work # past values of u2, and therefore you are not allowed to work
...@@ -992,7 +989,7 @@ class T_Scan(unittest.TestCase): ...@@ -992,7 +989,7 @@ class T_Scan(unittest.TestCase):
# Old way of doing inplace operations is deprecated .. tests don't # Old way of doing inplace operations is deprecated .. tests don't
# make sense anymore. # make sense anymore.
#assert not numpy.allclose( theano_x0 , vu2[1:4]) #assert not numpy.allclose( theano_x0 , vu2[1:4])
#assert numpy.allclose( theano_x1 , vu1[0:3]) #utt.assert_allclose( theano_x1 , vu1[0:3])
def test_inplace3(self): def test_inplace3(self):
rng = numpy.random.RandomState(utt.fetch_seed()) rng = numpy.random.RandomState(utt.fetch_seed())
...@@ -1119,11 +1116,11 @@ class T_Scan(unittest.TestCase): ...@@ -1119,11 +1116,11 @@ class T_Scan(unittest.TestCase):
numpy_W1 = numpy_W1 + .1 numpy_W1 = numpy_W1 + .1
numpy_W2 = numpy_W2 + .05 numpy_W2 = numpy_W2 + .05
assert numpy.allclose(theano_y0, numpy_y0[3:]) utt.assert_allclose(theano_y0, numpy_y0[3:])
assert numpy.allclose(theano_y1, numpy_y1[1:]) utt.assert_allclose(theano_y1, numpy_y1[1:])
assert numpy.allclose(theano_y2, numpy_y2) utt.assert_allclose(theano_y2, numpy_y2)
assert numpy.allclose(W1.get_value(), numpy_W1) utt.assert_allclose(W1.get_value(), numpy_W1)
assert numpy.allclose(W2.get_value(), numpy_W2) utt.assert_allclose(W2.get_value(), numpy_W2)
def test_grad_dtype_change(self): def test_grad_dtype_change(self):
x = tensor.fscalar('x') x = tensor.fscalar('x')
...@@ -1191,9 +1188,9 @@ class T_Scan(unittest.TestCase): ...@@ -1191,9 +1188,9 @@ class T_Scan(unittest.TestCase):
numpy_v[i] = rng.uniform(-1, 1, size=(2,)) numpy_v[i] = rng.uniform(-1, 1, size=(2,))
theano_v = my_f() theano_v = my_f()
assert numpy.allclose(theano_v, numpy_v[:5, :]) utt.assert_allclose(theano_v, numpy_v[:5, :])
theano_v = my_f() theano_v = my_f()
assert numpy.allclose(theano_v, numpy_v[5:, :]) utt.assert_allclose(theano_v, numpy_v[5:, :])
def test_cuda_gibbs_chain(self): def test_cuda_gibbs_chain(self):
from theano.sandbox import cuda from theano.sandbox import cuda
...@@ -1293,7 +1290,7 @@ class T_Scan(unittest.TestCase): ...@@ -1293,7 +1290,7 @@ class T_Scan(unittest.TestCase):
t_result = my_f(v_vsample) t_result = my_f(v_vsample)
n_result = numpy_implementation(v_vsample) n_result = numpy_implementation(v_vsample)
assert numpy.allclose(t_result, n_result) utt.assert_allclose(t_result, n_result)
def test_only_shared_no_input_no_output(self): def test_only_shared_no_input_no_output(self):
rng = numpy.random.RandomState(utt.fetch_seed()) rng = numpy.random.RandomState(utt.fetch_seed())
...@@ -1317,7 +1314,7 @@ class T_Scan(unittest.TestCase): ...@@ -1317,7 +1314,7 @@ class T_Scan(unittest.TestCase):
n_steps = 3 n_steps = 3
this_f(n_steps) this_f(n_steps)
numpy_state = v_state * (2 ** (n_steps)) numpy_state = v_state * (2 ** (n_steps))
assert numpy.allclose(state.get_value(), numpy_state) utt.assert_allclose(state.get_value(), numpy_state)
def test_map_functionality(self): def test_map_functionality(self):
def f_rnn(u_t): def f_rnn(u_t):
...@@ -1342,7 +1339,7 @@ class T_Scan(unittest.TestCase): ...@@ -1342,7 +1339,7 @@ class T_Scan(unittest.TestCase):
v_u = rng.uniform(size=(5,), low=-5., high=5.) v_u = rng.uniform(size=(5,), low=-5., high=5.)
numpy_result = v_u + 3 numpy_result = v_u + 3
theano_result = f2(v_u) theano_result = f2(v_u)
assert numpy.allclose(theano_result, numpy_result) utt.assert_allclose(theano_result, numpy_result)
def test_map(self): def test_map(self):
v = theano.tensor.vector('v') v = theano.tensor.vector('v')
...@@ -1362,7 +1359,7 @@ class T_Scan(unittest.TestCase): ...@@ -1362,7 +1359,7 @@ class T_Scan(unittest.TestCase):
vals = rng.uniform(size=(10,), low=-5., high=5.) vals = rng.uniform(size=(10,), low=-5., high=5.)
abs_vals = abs(vals) abs_vals = abs(vals)
theano_vals = f(vals) theano_vals = f(vals)
assert numpy.allclose(abs_vals, theano_vals) utt.assert_allclose(abs_vals, theano_vals)
def test_backwards(self): def test_backwards(self):
def f_rnn(u_t, x_tm1, W_in, W): def f_rnn(u_t, x_tm1, W_in, W):
...@@ -1399,7 +1396,7 @@ class T_Scan(unittest.TestCase): ...@@ -1399,7 +1396,7 @@ class T_Scan(unittest.TestCase):
v_out[step] = v_u[3 - step] * W_in + v_out[step - 1] * W v_out[step] = v_u[3 - step] * W_in + v_out[step - 1] * W
theano_values = f2(v_u, v_x0, W_in, W) theano_values = f2(v_u, v_x0, W_in, W)
assert numpy.allclose(theano_values, v_out) utt.assert_allclose(theano_values, v_out)
def test_reduce(self): def test_reduce(self):
v = theano.tensor.vector('v') v = theano.tensor.vector('v')
...@@ -1780,7 +1777,7 @@ class T_Scan(unittest.TestCase): ...@@ -1780,7 +1777,7 @@ class T_Scan(unittest.TestCase):
num_grad = multiple_outputs_numeric_grad( num_grad = multiple_outputs_numeric_grad(
reset_rng_cost_fn, [v_u, v_x0, vW_in]) reset_rng_cost_fn, [v_u, v_x0, vW_in])
analytic_grad = reset_rng_grad_fn(v_u, v_x0, vW_in) analytic_grad = reset_rng_grad_fn(v_u, v_x0, vW_in)
assert numpy.allclose(analytic_grad[0][:2], numpy.zeros((2, 2))) utt.assert_allclose(analytic_grad[0][:2], numpy.zeros((2, 2)))
def test_draw_as_input_to_scan(self): def test_draw_as_input_to_scan(self):
trng = theano.tensor.shared_randomstreams.RandomStreams(123) trng = theano.tensor.shared_randomstreams.RandomStreams(123)
...@@ -1799,8 +1796,8 @@ class T_Scan(unittest.TestCase): ...@@ -1799,8 +1796,8 @@ class T_Scan(unittest.TestCase):
ny1, nz1 = f(nx) ny1, nz1 = f(nx)
ny2, nz2 = f(nx) ny2, nz2 = f(nx)
assert numpy.allclose([ny1, ny1], nz1) utt.assert_allclose([ny1, ny1], nz1)
assert numpy.allclose([ny2, ny2], nz2) utt.assert_allclose([ny2, ny2], nz2)
assert not numpy.allclose(ny1, ny2) assert not numpy.allclose(ny1, ny2)
def test_grad_of_shared(self): def test_grad_of_shared(self):
...@@ -1813,7 +1810,7 @@ class T_Scan(unittest.TestCase): ...@@ -1813,7 +1810,7 @@ class T_Scan(unittest.TestCase):
m = theano.tensor.grad(y.sum(), x1) m = theano.tensor.grad(y.sum(), x1)
f = theano.function([x2], m, allow_input_downcast=True) f = theano.function([x2], m, allow_input_downcast=True)
assert numpy.allclose(f([2, 3]), 5) utt.assert_allclose(f([2, 3]), 5)
def test_computing_gradient(self): def test_computing_gradient(self):
x1 = theano.tensor.scalar('x1') x1 = theano.tensor.scalar('x1')
...@@ -1918,7 +1915,7 @@ class T_Scan(unittest.TestCase): ...@@ -1918,7 +1915,7 @@ class T_Scan(unittest.TestCase):
vR = numpy.array([[3.6, 1.8], [1.8, 0.9]], dtype=theano.config.floatX) vR = numpy.array([[3.6, 1.8], [1.8, 0.9]], dtype=theano.config.floatX)
out = f(vx, vA) out = f(vx, vA)
assert numpy.allclose(out, vR) utt.assert_allclose(out, vR)
def test_cloning_no_replace_strict_copy_inputs(self): def test_cloning_no_replace_strict_copy_inputs(self):
# This has nothing to do with scan, but it refers to the clone # This has nothing to do with scan, but it refers to the clone
...@@ -2097,8 +2094,8 @@ class T_Scan(unittest.TestCase): ...@@ -2097,8 +2094,8 @@ class T_Scan(unittest.TestCase):
v_y0, v_y0,
vW_in1) vW_in1)
assert numpy.allclose(theano_x, v_x) utt.assert_allclose(theano_x, v_x)
assert numpy.allclose(theano_y, v_y) utt.assert_allclose(theano_y, v_y)
def test_scan_as_tensor_on_gradients(self): def test_scan_as_tensor_on_gradients(self):
""" """
...@@ -2182,8 +2179,8 @@ class T_Scan(unittest.TestCase): ...@@ -2182,8 +2179,8 @@ class T_Scan(unittest.TestCase):
(theano_dump, theano_x, theano_y) = f4(v_u1, v_u2, v_x0, v_y0, vW_in1) (theano_dump, theano_x, theano_y) = f4(v_u1, v_u2, v_x0, v_y0, vW_in1)
assert numpy.allclose(theano_x, v_x[-1:]) utt.assert_allclose(theano_x, v_x[-1:])
assert numpy.allclose(theano_y, v_y[-1:]) utt.assert_allclose(theano_y, v_y[-1:])
def caching_nsteps_by_scan_op(self): def caching_nsteps_by_scan_op(self):
W = tensor.matrix('weights') W = tensor.matrix('weights')
...@@ -2220,7 +2217,7 @@ class T_Scan(unittest.TestCase): ...@@ -2220,7 +2217,7 @@ class T_Scan(unittest.TestCase):
rval = numpy.asarray([[5187989] * 5] * 5, dtype=theano.config.floatX) rval = numpy.asarray([[5187989] * 5] * 5, dtype=theano.config.floatX)
arg1 = numpy.ones((5, 5), dtype=theano.config.floatX) arg1 = numpy.ones((5, 5), dtype=theano.config.floatX)
arg2 = numpy.ones((10, 5), dtype=theano.config.floatX) arg2 = numpy.ones((10, 5), dtype=theano.config.floatX)
assert numpy.allclose(f(arg1, arg2), rval) utt.assert_allclose(f(arg1, arg2), rval)
def test_save_mem_reduced_number_of_steps(self): def test_save_mem_reduced_number_of_steps(self):
def f_rnn(u_t): def f_rnn(u_t):
...@@ -2259,13 +2256,13 @@ class T_Scan(unittest.TestCase): ...@@ -2259,13 +2256,13 @@ class T_Scan(unittest.TestCase):
# compute the output in numpy # compute the output in numpy
tx1, tx2, tx3, tx4, tx5, tx6, tx7 = f2(v_u, 3, 15) tx1, tx2, tx3, tx4, tx5, tx6, tx7 = f2(v_u, 3, 15)
assert numpy.allclose(tx1, v_u[:2] + 1.) utt.assert_allclose(tx1, v_u[:2] + 1.)
assert numpy.allclose(tx2, v_u[4] + 2.) utt.assert_allclose(tx2, v_u[4] + 2.)
assert numpy.allclose(tx3, v_u[3] + 3.) utt.assert_allclose(tx3, v_u[3] + 3.)
assert numpy.allclose(tx4, v_u[:3] + 4.) utt.assert_allclose(tx4, v_u[:3] + 4.)
assert numpy.allclose(tx5, v_u[-10] + 5.) utt.assert_allclose(tx5, v_u[-10] + 5.)
assert numpy.allclose(tx6, v_u[-15] + 6.) utt.assert_allclose(tx6, v_u[-15] + 6.)
assert numpy.allclose(tx7, v_u[:-15] + 7.) utt.assert_allclose(tx7, v_u[:-15] + 7.)
scan_node = f2.maker.fgraph.outputs[0].owner.inputs[0] scan_node = f2.maker.fgraph.outputs[0].owner.inputs[0]
# Maybe ugly, way to check if the optimization had # Maybe ugly, way to check if the optimization had
...@@ -2315,11 +2312,11 @@ class T_Scan(unittest.TestCase): ...@@ -2315,11 +2312,11 @@ class T_Scan(unittest.TestCase):
# compute the output in numpy # compute the output in numpy
tx1, tx2, tx3, tx4, tx5 = f2(v_u, [0, 0], 0, [0, 0], 0) tx1, tx2, tx3, tx4, tx5 = f2(v_u, [0, 0], 0, [0, 0], 0)
assert numpy.allclose(tx1, v_u[-7] + 1.) utt.assert_allclose(tx1, v_u[-7] + 1.)
assert numpy.allclose(tx2, v_u[-3:-1] + 2.) utt.assert_allclose(tx2, v_u[-3:-1] + 2.)
assert numpy.allclose(tx3, v_u[-6:] + 3.) utt.assert_allclose(tx3, v_u[-6:] + 3.)
assert numpy.allclose(tx4, v_u[-1] + 4.) utt.assert_allclose(tx4, v_u[-1] + 4.)
assert numpy.allclose(tx5, v_u[-1] + 5.) utt.assert_allclose(tx5, v_u[-1] + 5.)
# The following test will fail in DebugMode if there are # The following test will fail in DebugMode if there are
# some problems in Scan.infer_shape # some problems in Scan.infer_shape
...@@ -2456,7 +2453,7 @@ class T_Scan(unittest.TestCase): ...@@ -2456,7 +2453,7 @@ class T_Scan(unittest.TestCase):
f_vals = f(x_val) f_vals = f(x_val)
memory.set_value(mem_val.copy()) memory.set_value(mem_val.copy())
f2_vals = f2(x_val) f2_vals = f2(x_val)
assert numpy.allclose(f_vals, f2_vals) utt.assert_allclose(f_vals, f2_vals)
def test_reduce_memory_consumption(self): def test_reduce_memory_consumption(self):
...@@ -2479,7 +2476,7 @@ class T_Scan(unittest.TestCase): ...@@ -2479,7 +2476,7 @@ class T_Scan(unittest.TestCase):
assert f1().shape[0] == 1 assert f1().shape[0] == 1
gx = theano.tensor.grad(o, x) gx = theano.tensor.grad(o, x)
f2 = theano.function([], gx) f2 = theano.function([], gx)
assert numpy.allclose(f2(), numpy.ones((10,))) utt.assert_allclose(f2(), numpy.ones((10,)))
def test_foldl_memory_consumption(self): def test_foldl_memory_consumption(self):
x = theano.shared(numpy.asarray( x = theano.shared(numpy.asarray(
...@@ -2502,7 +2499,7 @@ class T_Scan(unittest.TestCase): ...@@ -2502,7 +2499,7 @@ class T_Scan(unittest.TestCase):
assert f1().shape[0] == 1 assert f1().shape[0] == 1
gx = theano.tensor.grad(o, x) gx = theano.tensor.grad(o, x)
f2 = theano.function([], gx) f2 = theano.function([], gx)
assert numpy.allclose(f2(), numpy.ones((10,))) utt.assert_allclose(f2(), numpy.ones((10,)))
def test_foldr_memory_consumption(self): def test_foldr_memory_consumption(self):
...@@ -2526,7 +2523,7 @@ class T_Scan(unittest.TestCase): ...@@ -2526,7 +2523,7 @@ class T_Scan(unittest.TestCase):
assert f1().shape[0] == 1 assert f1().shape[0] == 1
gx = theano.tensor.grad(o, x) gx = theano.tensor.grad(o, x)
f2 = theano.function([], gx) f2 = theano.function([], gx)
assert numpy.allclose(f2(), numpy.ones((10,))) utt.assert_allclose(f2(), numpy.ones((10,)))
def test_rop2(self): def test_rop2(self):
seed = utt.fetch_seed() seed = utt.fetch_seed()
...@@ -2600,9 +2597,9 @@ class T_Scan(unittest.TestCase): ...@@ -2600,9 +2597,9 @@ class T_Scan(unittest.TestCase):
vnu, vnh0, vnW, vno = fn_rop(v_u, v_h0, v_W, v_eu, v_eh0, v_eW) vnu, vnh0, vnW, vno = fn_rop(v_u, v_h0, v_W, v_eu, v_eh0, v_eW)
tnu, tnh0, tnW, tno = fn_test(v_u, v_h0, v_W, v_eu, v_eh0, v_eW) tnu, tnh0, tnW, tno = fn_test(v_u, v_h0, v_W, v_eu, v_eh0, v_eW)
assert numpy.allclose(vnu, tnu, atol=1e-6) utt.assert_allclose(vnu, tnu, atol=1e-6)
assert numpy.allclose(vnh0, tnh0, atol=1e-6) utt.assert_allclose(vnh0, tnh0, atol=1e-6)
assert numpy.allclose(vnW, tnW, atol=1e-6) utt.assert_allclose(vnW, tnW, atol=1e-6)
def test_rop(self): def test_rop(self):
seed = utt.fetch_seed() seed = utt.fetch_seed()
...@@ -2673,9 +2670,9 @@ class T_Scan(unittest.TestCase): ...@@ -2673,9 +2670,9 @@ class T_Scan(unittest.TestCase):
vnu, vnh0, vnW = fn_rop(v_u, v_h0, v_W, v_eu, v_eh0, v_eW) vnu, vnh0, vnW = fn_rop(v_u, v_h0, v_W, v_eu, v_eh0, v_eW)
tnu, tnh0, tnW = fn_test(v_u, v_h0, v_W, v_eu, v_eh0, v_eW) tnu, tnh0, tnW = fn_test(v_u, v_h0, v_W, v_eu, v_eh0, v_eW)
assert numpy.allclose(vnu, tnu, atol=1e-6) utt.assert_allclose(vnu, tnu, atol=1e-6)
assert numpy.allclose(vnh0, tnh0, atol=1e-6) utt.assert_allclose(vnh0, tnh0, atol=1e-6)
assert numpy.allclose(vnW, tnW, atol=1e-6) utt.assert_allclose(vnW, tnW, atol=1e-6)
def test_pushout_all(self): def test_pushout_all(self):
W1 = tensor.matrix('W1') W1 = tensor.matrix('W1')
...@@ -2709,7 +2706,7 @@ class T_Scan(unittest.TestCase): ...@@ -2709,7 +2706,7 @@ class T_Scan(unittest.TestCase):
# theano. Note that what we ask theano to do is to repeat the 2 # theano. Note that what we ask theano to do is to repeat the 2
# elements vector v_out 5 times # elements vector v_out 5 times
sol[:, :] = v_out sol[:, :] = v_out
assert numpy.allclose(sol, f(v_h, v_W1, v_W2)) utt.assert_allclose(sol, f(v_h, v_W1, v_W2))
def test_pushout(self): def test_pushout(self):
W1 = tensor.matrix('W1') W1 = tensor.matrix('W1')
...@@ -3085,8 +3082,8 @@ class T_Scan(unittest.TestCase): ...@@ -3085,8 +3082,8 @@ class T_Scan(unittest.TestCase):
(theano_dump, theano_x, theano_y) = f4(v_u1, v_u2, v_x0, v_y0, vW_in1) (theano_dump, theano_x, theano_y) = f4(v_u1, v_u2, v_x0, v_y0, vW_in1)
assert numpy.allclose(theano_x, v_x[-2:]) utt.assert_allclose(theano_x, v_x[-2:])
assert numpy.allclose(theano_y, v_y[-4:]) utt.assert_allclose(theano_y, v_y[-4:])
def test_opt_order(self): def test_opt_order(self):
""" """
...@@ -3113,7 +3110,7 @@ class T_Scan(unittest.TestCase): ...@@ -3113,7 +3110,7 @@ class T_Scan(unittest.TestCase):
vA = numpy.array([[1., 1.], [1., 0.]], dtype=theano.config.floatX) vA = numpy.array([[1., 1.], [1., 0.]], dtype=theano.config.floatX)
vR = numpy.array([[[2, 1], [4, 2]], [[2, 1], [4, 2]]], vR = numpy.array([[[2, 1], [4, 2]], [[2, 1], [4, 2]]],
dtype=theano.config.floatX) dtype=theano.config.floatX)
assert numpy.allclose(f(vx, vA), vR) utt.assert_allclose(f(vx, vA), vR)
def test_savemem_opt(self): def test_savemem_opt(self):
y0 = theano.shared(numpy.ones((2, 10))) y0 = theano.shared(numpy.ones((2, 10)))
...@@ -3334,7 +3331,7 @@ class T_Scan(unittest.TestCase): ...@@ -3334,7 +3331,7 @@ class T_Scan(unittest.TestCase):
rng = numpy.random.RandomState(utt.fetch_seed()) rng = numpy.random.RandomState(utt.fetch_seed())
vA = rng.uniform(size=(5, 5)).astype(theano.config.floatX) vA = rng.uniform(size=(5, 5)).astype(theano.config.floatX)
vB = rng.uniform(size=(5, 5)).astype(theano.config.floatX) vB = rng.uniform(size=(5, 5)).astype(theano.config.floatX)
assert numpy.allclose(f(vA, vB), numpy.dot(vA.T, vB)) utt.assert_allclose(f(vA, vB), numpy.dot(vA.T, vB))
def test_pregreedy_optimizer(self): def test_pregreedy_optimizer(self):
...@@ -3390,10 +3387,10 @@ class T_Scan(unittest.TestCase): ...@@ -3390,10 +3387,10 @@ class T_Scan(unittest.TestCase):
rng = numpy.random.RandomState(utt.fetch_seed()) rng = numpy.random.RandomState(utt.fetch_seed())
v_u = asarrayX(rng.uniform(size=(5,))) v_u = asarrayX(rng.uniform(size=(5,)))
outs = f(v_u, [0, 0, 0], 0) outs = f(v_u, [0, 0, 0], 0)
assert numpy.allclose(outs[0], v_u + 1) utt.assert_allclose(outs[0], v_u + 1)
assert numpy.allclose(outs[1], v_u + 2) utt.assert_allclose(outs[1], v_u + 2)
assert numpy.allclose(outs[2], v_u + 3) utt.assert_allclose(outs[2], v_u + 3)
assert numpy.allclose(sh.get_value(), v_u[-1] + 4) utt.assert_allclose(sh.get_value(), v_u[-1] + 4)
def test_eliminate_nonseqs(self): def test_eliminate_nonseqs(self):
W = tensor.scalar('W') W = tensor.scalar('W')
...@@ -3423,10 +3420,10 @@ class T_Scan(unittest.TestCase): ...@@ -3423,10 +3420,10 @@ class T_Scan(unittest.TestCase):
rng = numpy.random.RandomState(utt.fetch_seed()) rng = numpy.random.RandomState(utt.fetch_seed())
v_w = asarrayX(rng.uniform()) v_w = asarrayX(rng.uniform())
outs = f(v_w, [0, 0, 0], 0) outs = f(v_w, [0, 0, 0], 0)
assert numpy.allclose(outs[0], v_w + 1) utt.assert_allclose(outs[0], v_w + 1)
assert numpy.allclose(outs[1], v_w + 2) utt.assert_allclose(outs[1], v_w + 2)
assert numpy.allclose(outs[2], v_w + 3) utt.assert_allclose(outs[2], v_w + 3)
assert numpy.allclose(sh.get_value(), v_w + 4) utt.assert_allclose(sh.get_value(), v_w + 4)
def test_grad_bug_disconnected_input(self): def test_grad_bug_disconnected_input(self):
W = theano.shared(numpy.zeros((3, 3)), name='W') W = theano.shared(numpy.zeros((3, 3)), name='W')
...@@ -3435,7 +3432,7 @@ class T_Scan(unittest.TestCase): ...@@ -3435,7 +3432,7 @@ class T_Scan(unittest.TestCase):
#This used to raise an exception #This used to raise an exception
f = theano.function([v], theano.tensor.grad(y.sum(), W)) f = theano.function([v], theano.tensor.grad(y.sum(), W))
assert numpy.allclose(f([1,2]), [[0,0,0],[1,1,1],[1,1,1]]) utt.assert_allclose(f([1,2]), [[0,0,0],[1,1,1],[1,1,1]])
def test_clone(self): def test_clone(self):
def test(x, y, mention_y): def test(x, y, mention_y):
...@@ -3448,9 +3445,9 @@ class T_Scan(unittest.TestCase): ...@@ -3448,9 +3445,9 @@ class T_Scan(unittest.TestCase):
return theano.function([], out)() return theano.function([], out)()
x = theano.shared(numpy.asarray(0., dtype=theano.config.floatX)) x = theano.shared(numpy.asarray(0., dtype=theano.config.floatX))
assert numpy.allclose(test(x, tensor.sum((x+1)**2), mention_y=False), utt.assert_allclose(test(x, tensor.sum((x+1)**2), mention_y=False),
1.21000003815) 1.21000003815)
assert numpy.allclose(test(x, tensor.sum((x+1)**2), mention_y=True), utt.assert_allclose(test(x, tensor.sum((x+1)**2), mention_y=True),
1.21000003815) 1.21000003815)
def test_grad_find_input(self): def test_grad_find_input(self):
...@@ -3526,7 +3523,7 @@ class T_Scan(unittest.TestCase): ...@@ -3526,7 +3523,7 @@ class T_Scan(unittest.TestCase):
assert len(inp) == 1 assert len(inp) == 1
assert (len(inp) == len(set(inp))) assert (len(inp) == len(set(inp)))
#import pdb;pdb.set_trace() #import pdb;pdb.set_trace()
#assert numpy.allclose(f([1, 2]), [[0, 0, 0], [1, 1, 1], [1, 1, 1]]) #utt.assert_allclose(f([1, 2]), [[0, 0, 0], [1, 1, 1], [1, 1, 1]])
def test_speed(): def test_speed():
......
theano/sparse/tests/test_basic.py
浏览文件 @ 60ad703d
...@@ -48,7 +48,6 @@ from theano.sparse import ( ...@@ -48,7 +48,6 @@ from theano.sparse import (
from theano.sparse.opt import (StructuredDotCSC, UsmmCscDense, CSMGradC) from theano.sparse.opt import (StructuredDotCSC, UsmmCscDense, CSMGradC)
from theano.tests import unittest_tools as utt from theano.tests import unittest_tools as utt
from theano.tensor.basic import _allclose
def as_sparse_format(data, format): def as_sparse_format(data, format):
...@@ -886,7 +885,7 @@ class test_structureddot(unittest.TestCase): ...@@ -886,7 +885,7 @@ class test_structureddot(unittest.TestCase):
scipy_result = spmat * mat scipy_result = spmat * mat
assert theano_result.shape == scipy_result.shape assert theano_result.shape == scipy_result.shape
assert theano_result.dtype == scipy_result.dtype assert theano_result.dtype == scipy_result.dtype
assert _allclose(theano_result, scipy_result) utt.assert_allclose(scipy_result, theano_result)
def test_opt_unpack(self): def test_opt_unpack(self):
# #
...@@ -999,7 +998,7 @@ class test_structureddot(unittest.TestCase): ...@@ -999,7 +998,7 @@ class test_structureddot(unittest.TestCase):
# fail if Theano is slower than scipy by more than a certain amount # fail if Theano is slower than scipy by more than a certain amount
overhead_tol = 0.003 # seconds overall overhead_tol = 0.003 # seconds overall
overhead_rtol = 1.2 # times as long overhead_rtol = 1.2 # times as long
self.assertTrue(numpy.allclose(theano_result, scipy_result)) utt.assert_allclose(scipy_result, theano_result)
if not theano.config.mode in ["DebugMode", "DEBUG_MODE"]: if not theano.config.mode in ["DebugMode", "DEBUG_MODE"]:
self.assertFalse(theano_time > overhead_rtol * scipy_time + self.assertFalse(theano_time > overhead_rtol * scipy_time +
overhead_tol) overhead_tol)
...@@ -1035,7 +1034,7 @@ class test_structureddot(unittest.TestCase): ...@@ -1035,7 +1034,7 @@ class test_structureddot(unittest.TestCase):
# print 'scipy took', scipy_time # print 'scipy took', scipy_time
overhead_tol = 0.002 # seconds overhead_tol = 0.002 # seconds
overhead_rtol = 1.1 # times as long overhead_rtol = 1.1 # times as long
self.assertTrue(numpy.allclose(theano_result, scipy_result)) utt.assert_allclose(scipy_result, theano_result)
if (not theano.config.mode in ["DebugMode", "DEBUG_MODE"] and if (not theano.config.mode in ["DebugMode", "DEBUG_MODE"] and
theano.config.cxx): theano.config.cxx):
self.assertFalse(theano_time > overhead_rtol * scipy_time + self.assertFalse(theano_time > overhead_rtol * scipy_time +
...@@ -1075,7 +1074,7 @@ class DotTests(utt.InferShapeTester): ...@@ -1075,7 +1074,7 @@ class DotTests(utt.InferShapeTester):
f_a = theano.function([x, y], theano.sparse.dot(x, y)) f_a = theano.function([x, y], theano.sparse.dot(x, y))
f_b = lambda x, y: x * y f_b = lambda x, y: x * y
assert _allclose(f_a(x_v, y_v), f_b(x_v, y_v)) utt.assert_allclose(f_a(x_v, y_v), f_b(x_v, y_v))
# Test infer_shape # Test infer_shape
self._compile_and_check([x, y], [theano.sparse.dot(x, y)], self._compile_and_check([x, y], [theano.sparse.dot(x, y)],
...@@ -1094,7 +1093,7 @@ class DotTests(utt.InferShapeTester): ...@@ -1094,7 +1093,7 @@ class DotTests(utt.InferShapeTester):
f_a = theano.function([x, y], theano.sparse.dot(x, y)) f_a = theano.function([x, y], theano.sparse.dot(x, y))
f_b = lambda x, y: x * y f_b = lambda x, y: x * y
assert _allclose(f_a(x_v, y_v), f_b(x_v, y_v)) utt.assert_allclose(f_a(x_v, y_v), f_b(x_v, y_v))
# Test infer_shape # Test infer_shape
self._compile_and_check([x, y], [theano.sparse.dot(x, y)], self._compile_and_check([x, y], [theano.sparse.dot(x, y)],
...@@ -1117,12 +1116,12 @@ class DotTests(utt.InferShapeTester): ...@@ -1117,12 +1116,12 @@ class DotTests(utt.InferShapeTester):
x = theano.sparse.SparseType(format=x_f, dtype=d1)('x') x = theano.sparse.SparseType(format=x_f, dtype=d1)('x')
y = theano.sparse.SparseType(format=x_f, dtype=d2)('x') y = theano.sparse.SparseType(format=x_f, dtype=d2)('x')
f_a = theano.function([x, y], theano.sparse.dot(x, y)) f_a = lambda x, y: x * y
f_b = lambda x, y: x * y f_b = theano.function([x, y], theano.sparse.dot(x, y))
vx = getattr(self, 'x_' + x_f).astype(d1) vx = getattr(self, 'x_' + x_f).astype(d1)
vy = getattr(self, 'y_' + y_f).astype(d2) vy = getattr(self, 'y_' + y_f).astype(d2)
assert _allclose(f_a(vx, vy), f_b(vx, vy).toarray()) utt.assert_allclose(f_a(vx, vy).toarray(), f_b(vx, vy))
# Test infer_shape # Test infer_shape
f_a = theano.function([x, y], theano.sparse.dot(x, y).shape) f_a = theano.function([x, y], theano.sparse.dot(x, y).shape)
...@@ -1150,7 +1149,7 @@ class DotTests(utt.InferShapeTester): ...@@ -1150,7 +1149,7 @@ class DotTests(utt.InferShapeTester):
a_val = scipy.sparse.csr_matrix(random_lil((5, 3), 'float32', 5)) a_val = scipy.sparse.csr_matrix(random_lil((5, 3), 'float32', 5))
d_theano = f(a_val) d_theano = f(a_val)
d_numpy = a_val * b.get_value() d_numpy = a_val * b.get_value()
assert numpy.allclose(d_theano, d_numpy) utt.assert_allclose(d_numpy, d_theano)
def test_int32_dtype(self): def test_int32_dtype(self):
# Reported on the theano-user mailing-list: # Reported on the theano-user mailing-list:
...@@ -1256,7 +1255,7 @@ class UsmmTests(unittest.TestCase): ...@@ -1256,7 +1255,7 @@ class UsmmTests(unittest.TestCase):
theano.tensor.basic.float64_atol = orig_atol theano.tensor.basic.float64_atol = orig_atol
theano.tensor.basic.float64_rtol = orig_rtol theano.tensor.basic.float64_rtol = orig_rtol
assert _allclose(f_a_out, f_b_out, rtol=1e-5), (f_a_out, f_b_out) utt.assert_allclose(f_a_out, f_b_out, rtol=1e-5)
topo = f_a.maker.fgraph.toposort() topo = f_a.maker.fgraph.toposort()
up = theano.scalar.upcast(dtype1, dtype2, dtype3, dtype4) up = theano.scalar.upcast(dtype1, dtype2, dtype3, dtype4)
...@@ -1492,7 +1491,7 @@ class ColScaleCSCTester(utt.InferShapeTester): ...@@ -1492,7 +1491,7 @@ class ColScaleCSCTester(utt.InferShapeTester):
expected = x * s expected = x * s
assert tested.format == format assert tested.format == format
assert numpy.allclose(tested.toarray(), expected) utt.assert_allclose(expected, tested.toarray())
def test_infer_shape(self): def test_infer_shape(self):
for format, cls in [('csc', sparse.ColScaleCSC), for format, cls in [('csc', sparse.ColScaleCSC),
...@@ -1533,7 +1532,7 @@ class RowScaleCSCTester(utt.InferShapeTester): ...@@ -1533,7 +1532,7 @@ class RowScaleCSCTester(utt.InferShapeTester):
expected = x * s expected = x * s
assert tested.format == format assert tested.format == format
assert numpy.allclose(tested.toarray(), expected) utt.assert_allclose(expected, tested.toarray())
def test_infer_shape(self): def test_infer_shape(self):
for format, cls in [('csc', sparse.RowScaleCSC), for format, cls in [('csc', sparse.RowScaleCSC),
...@@ -1579,7 +1578,7 @@ class SpSumTester(utt.InferShapeTester): ...@@ -1579,7 +1578,7 @@ class SpSumTester(utt.InferShapeTester):
f = theano.function(variable, self.op(variable[0], axis=axis)) f = theano.function(variable, self.op(variable[0], axis=axis))
tested = f(*data) tested = f(*data)
expected = data[0].todense().sum(axis).ravel() expected = data[0].todense().sum(axis).ravel()
assert numpy.allclose(tested, expected) utt.assert_allclose(expected, tested)
def test_infer_shape(self): def test_infer_shape(self):
for format in sparse.sparse_formats: for format in sparse.sparse_formats:
...@@ -1621,7 +1620,7 @@ class DiagTester(utt.InferShapeTester): ...@@ -1621,7 +1620,7 @@ class DiagTester(utt.InferShapeTester):
tested = f(*data) tested = f(*data)
expected = data[0].toarray().diagonal() expected = data[0].toarray().diagonal()
assert numpy.allclose(tested, expected) utt.assert_allclose(expected, tested)
def test_infer_shape(self): def test_infer_shape(self):
for format in sparse.sparse_formats: for format in sparse.sparse_formats:
...@@ -1659,7 +1658,7 @@ class SquareDiagonalTester(utt.InferShapeTester): ...@@ -1659,7 +1658,7 @@ class SquareDiagonalTester(utt.InferShapeTester):
tested = f(*data).toarray() tested = f(*data).toarray()
expected = numpy.diag(*data) expected = numpy.diag(*data)
assert numpy.allclose(tested, expected) utt.assert_allclose(expected, tested)
assert tested.dtype == expected.dtype assert tested.dtype == expected.dtype
assert tested.shape == expected.shape assert tested.shape == expected.shape
...@@ -1701,7 +1700,7 @@ class EnsureSortedIndicesTester(utt.InferShapeTester): ...@@ -1701,7 +1700,7 @@ class EnsureSortedIndicesTester(utt.InferShapeTester):
tested = f(*data).toarray() tested = f(*data).toarray()
expected = data[0].sorted_indices().toarray() expected = data[0].sorted_indices().toarray()
assert numpy.allclose(tested, expected) utt.assert_allclose(expected, tested)
def test_infer_shape(self): def test_infer_shape(self):
for format in sparse.sparse_formats: for format in sparse.sparse_formats:
...@@ -1744,7 +1743,7 @@ class CleanTester(utt.InferShapeTester): ...@@ -1744,7 +1743,7 @@ class CleanTester(utt.InferShapeTester):
tested = tested.toarray() tested = tested.toarray()
expected = expected.toarray() expected = expected.toarray()
assert numpy.allclose(tested, expected) utt.assert_allclose(expected, tested)
def test_grad(self): def test_grad(self):
for format in sparse.sparse_formats: for format in sparse.sparse_formats:
...@@ -2048,9 +2047,9 @@ class CastTester(utt.InferShapeTester): ...@@ -2048,9 +2047,9 @@ class CastTester(utt.InferShapeTester):
t_cls = t_cls.toarray() t_cls = t_cls.toarray()
t_prop = t_prop.toarray() t_prop = t_prop.toarray()
assert numpy.allclose(t_func, expected) utt.assert_allclose(expected, t_func)
assert numpy.allclose(t_cls, expected) utt.assert_allclose(expected, t_cls)
assert numpy.allclose(t_prop, expected) utt.assert_allclose(expected, t_prop)
def test_infer_shape(self): def test_infer_shape(self):
for format in sparse.sparse_formats: for format in sparse.sparse_formats:
...@@ -2120,7 +2119,7 @@ class _HVStackTester(utt.InferShapeTester): ...@@ -2120,7 +2119,7 @@ class _HVStackTester(utt.InferShapeTester):
format=out_f, format=out_f,
dtype=dtype) dtype=dtype)
assert numpy.allclose(tested.toarray(), expected.toarray()) utt.assert_allclose(expected.toarray(), tested.toarray())
assert tested.format == expected.format assert tested.format == expected.format
assert tested.dtype == expected.dtype assert tested.dtype == expected.dtype
...@@ -2191,7 +2190,7 @@ class AddSSDataTester(utt.InferShapeTester): ...@@ -2191,7 +2190,7 @@ class AddSSDataTester(utt.InferShapeTester):
tested = f(*self.a[format]) tested = f(*self.a[format])
expected = 2 * self.a[format][0] expected = 2 * self.a[format][0]
assert numpy.allclose(tested.toarray(), expected.toarray()) utt.assert_allclose(expected.toarray(), tested.toarray())
assert tested.format == expected.format assert tested.format == expected.format
assert tested.dtype == expected.dtype assert tested.dtype == expected.dtype
...@@ -2286,7 +2285,7 @@ def elemwise_checker(op, expected_f, gap=None, test_dtypes=None, ...@@ -2286,7 +2285,7 @@ def elemwise_checker(op, expected_f, gap=None, test_dtypes=None,
tested = tested.toarray() tested = tested.toarray()
try: try:
assert numpy.allclose(tested, expected) utt.assert_allclose(expected, tested)
except AssertionError: except AssertionError:
raise AssertionError(self.__name__) raise AssertionError(self.__name__)
...@@ -2348,7 +2347,7 @@ def elemwise_checker(op, expected_f, gap=None, test_dtypes=None, ...@@ -2348,7 +2347,7 @@ def elemwise_checker(op, expected_f, gap=None, test_dtypes=None,
tested = tested.toarray() tested = tested.toarray()
try: try:
assert numpy.allclose(tested, expected, rtol=1e-2) utt.assert_allclose(tested, expected, rtol=1e-2)
except AssertionError: except AssertionError:
raise AssertionError(self.__name__) raise AssertionError(self.__name__)
...@@ -2578,7 +2577,7 @@ class MulSVTester(unittest.TestCase): ...@@ -2578,7 +2577,7 @@ class MulSVTester(unittest.TestCase):
out = f(spmat, mat) out = f(spmat, mat)
assert numpy.allclose(out.toarray(), spmat.toarray() * mat) utt.assert_allclose(spmat.toarray() * mat, out.toarray())
class StructuredAddSVTester(unittest.TestCase): class StructuredAddSVTester(unittest.TestCase):
...@@ -2615,8 +2614,8 @@ class StructuredAddSVTester(unittest.TestCase): ...@@ -2615,8 +2614,8 @@ class StructuredAddSVTester(unittest.TestCase):
out = f(spmat, mat) out = f(spmat, mat)
assert numpy.allclose(out.toarray(), utt.assert_allclose(spones.multiply(spmat + mat),
spones.multiply(spmat + mat)) out.toarray())
class SamplingDotTester(utt.InferShapeTester): class SamplingDotTester(utt.InferShapeTester):
...@@ -2645,7 +2644,7 @@ class SamplingDotTester(utt.InferShapeTester): ...@@ -2645,7 +2644,7 @@ class SamplingDotTester(utt.InferShapeTester):
x, y, p = self.a x, y, p = self.a
expected = p.multiply(numpy.dot(x, y.T)) expected = p.multiply(numpy.dot(x, y.T))
assert numpy.allclose(tested.toarray(), expected) utt.assert_allclose(expected, tested.toarray())
assert tested.format == 'csr' assert tested.format == 'csr'
assert tested.dtype == expected.dtype assert tested.dtype == expected.dtype
......
theano/tensor/basic.py
浏览文件 @ 60ad703d
...@@ -474,6 +474,8 @@ else: ...@@ -474,6 +474,8 @@ else:
def _allclose(a, b, rtol=None, atol=None): def _allclose(a, b, rtol=None, atol=None):
a = numpy.asarray(a)
b = numpy.asarray(b)
narrow = 'float32', 'complex64' narrow = 'float32', 'complex64'
if (str(a.dtype) in narrow) or (str(b.dtype) in narrow): if (str(a.dtype) in narrow) or (str(b.dtype) in narrow):
atol_ = float32_atol atol_ = float32_atol
......
theano/tests/unittest_tools.py
浏览文件 @ 60ad703d
from copy import copy, deepcopy from copy import copy, deepcopy
import logging import logging
from StringIO import StringIO
import sys import sys
import unittest import unittest
...@@ -239,3 +240,94 @@ class InferShapeTester(unittest.TestCase): ...@@ -239,3 +240,94 @@ class InferShapeTester(unittest.TestCase):
numeric_shapes = shapes_function(*numeric_inputs) numeric_shapes = shapes_function(*numeric_inputs)
for out, shape in zip(numeric_outputs, numeric_shapes): for out, shape in zip(numeric_outputs, numeric_shapes):
assert numpy.all(out.shape == shape) assert numpy.all(out.shape == shape)
def str_diagnostic(expected, value, rtol, atol):
"""Return a pretty multiline string representating the cause
of the exception"""
sio = StringIO()
try:
ssio = StringIO()
print >> ssio, " : shape, dtype, strides, min, max, n_inf, n_nan:"
print >> ssio, " Expected :",
print >> ssio, expected.shape,
print >> ssio, expected.dtype,
print >> ssio, expected.strides,
print >> ssio, expected.min(),
print >> ssio, expected.max(),
print >> ssio, numpy.isinf(expected).sum(),
print >> ssio, numpy.isnan(expected).sum(),
# only if all succeeds to we add anything to sio
print >> sio, ssio.getvalue()
except Exception:
pass
try:
ssio = StringIO()
print >> ssio, " Value :",
print >> ssio, value.shape,
print >> ssio, value.dtype,
print >> ssio, value.strides,
print >> ssio, value.min(),
print >> ssio, value.max(),
print >> ssio, numpy.isinf(value).sum(),
print >> ssio, numpy.isnan(value).sum(),
# only if all succeeds to we add anything to sio
print >> sio, ssio.getvalue()
except Exception:
pass
print >> sio, " expected :", expected
print >> sio, " value :", value
try:
ov = numpy.asarray(expected)
nv = numpy.asarray(value)
ssio = StringIO()
absdiff = numpy.absolute(nv - ov)
print >> ssio, " Max Abs Diff: ", numpy.max(absdiff)
print >> ssio, " Mean Abs Diff: ", numpy.mean(absdiff)
print >> ssio, " Median Abs Diff: ", numpy.median(absdiff)
print >> ssio, " Std Abs Diff: ", numpy.std(absdiff)
reldiff = numpy.absolute(nv - ov) / (numpy.absolute(nv) +
numpy.absolute(ov))
print >> ssio, " Max Rel Diff: ", numpy.max(reldiff)
print >> ssio, " Mean Rel Diff: ", numpy.mean(reldiff)
print >> ssio, " Median Rel Diff: ", numpy.median(reldiff)
print >> ssio, " Std Rel Diff: ", numpy.std(reldiff)
# only if all succeeds to we add anything to sio
print >> sio, ssio.getvalue()
except Exception:
pass
#Use the same formula as in _allclose to find the tolerance used
narrow = 'float32', 'complex64'
if ((str(expected.dtype) in narrow) or
(str(value.dtype) in narrow)):
atol_ = T.basic.float32_atol
rtol_ = T.basic.float32_rtol
else:
atol_ = T.basic.float64_atol
rtol_ = T.basic.float64_rtol
if rtol is not None:
rtol_ = rtol
if atol is not None:
atol_ = atol
print >> sio, " rtol, atol:", rtol_, atol_
return sio.getvalue()
class WrongValue(Exception):
def __init__(self, expected_val, val, rtol, atol):
self.val1 = expected_val
self.val2 = val
self.rtol = rtol
self.atol = atol
def __str__(self):
s = "WrongValue\n"
return s + str_diagnostic(self.val1, self.val2, self.rtol, self.atol)
def assert_allclose(val1, val2, rtol=None, atol=None):
if not T.basic._allclose(val1, val2, rtol, atol):
raise WrongValue(val1, val2, rtol, atol)
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