def test_min_informative_str():
# evaluates a reference output to make sure the
# min_informative_str function works as intended
A = tensor.matrix(name="A")
B = tensor.matrix(name="B")
C = A + B
C.name = "C"
D = tensor.matrix(name="D")
E = tensor.matrix(name="E")
F = D + E
G = C + F
mis = min_informative_str(G).replace("\t", " ")
reference = """A. Elemwise{add,no_inplace}
B. C
C. Elemwise{add,no_inplace}
D. D
E. E"""
if mis != reference:
print("--" + mis + "--")
print("--" + reference + "--")
assert mis == reference
def test_min_informative_str():
""" evaluates a reference output to make sure the
min_informative_str function works as intended """
A = tensor.matrix(name='A')
B = tensor.matrix(name='B')
C = A + B
C.name = 'C'
D = tensor.matrix(name='D')
E = tensor.matrix(name='E')
F = D + E
G = C + F
mis = min_informative_str(G).replace("\t", " ")
reference = """A. Elemwise{add,no_inplace}
B. C
C. Elemwise{add,no_inplace}
D. D
E. E"""
if mis != reference:
print('--' + mis + '--')
print('--' + reference + '--')
assert mis == reference
def test_min_informative_str():
""" evaluates a reference output to make sure the
min_informative_str function works as intended """
A = tensor.matrix(name='A')
B = tensor.matrix(name='B')
C = A + B
C.name = 'C'
D = tensor.matrix(name='D')
E = tensor.matrix(name='E')
F = D + E
G = C + F
mis = min_informative_str(G).replace("\t", " ")
reference = """A. Elemwise{add,no_inplace}
B. C
C. Elemwise{add,no_inplace}
D. D
E. E"""
if mis != reference:
print('--' + mis + '--')
print('--' + reference + '--')
assert mis == reference
def test_min_informative_str():
""" evaluates a reference output to make sure the
min_informative_str function works as intended """
A = tensor.matrix(name="A")
B = tensor.matrix(name="B")
C = A + B
C.name = "C"
D = tensor.matrix(name="D")
E = tensor.matrix(name="E")
F = D + E
G = C + F
mis = min_informative_str(G).replace("\t", " ")
reference = """A. Elemwise{add,no_inplace}
B. C
C. Elemwise{add,no_inplace}
D. D
E. E"""
if mis != reference:
print "--" + mis + "--"
print "--" + reference + "--"
assert mis == reference
def handle_line(fgraph, line, i, node, fn):
"""
Records new node computation.
Parameters
----------
line : string
Line to record. For example, the function name or node name.
i : integer
Node number in the toposort order.
node : Apply,
The Apply node which created the entry.
fn : Function,
Function related to Apply node.
"""
try:
self.record.handle_line(line)
except MismatchError as e:
print("Got this MismatchError:")
print(e)
print(f"while processing node i={i}:")
print(f"str(node):{node}")
print("Symbolic inputs: ")
for elem in node.inputs:
print(min_informative_str(elem))
print("str(output) of outputs: ")
for elem in fn.outputs:
assert isinstance(elem, list)
(elem, ) = elem
print(str(elem))
print(f"function name: {fgraph.name}")
raise MismatchError("Non-determinism detected by WrapLinker")
orig_obj = orig_obj[field[1:-1]]
elif field.startswith('.'):
obj_name += '.' + field
orig_obj = getattr(orig_obj,field[1:])
else:
obj_name + '[' + field + ']'
orig_obj = orig_obj[eval(field)]
if id(orig_obj) in cycle_check:
print "You're going in circles, "+obj_name+" is the same as "+cycle_check[id(orig_obj)]
quit()
cycle_check[id(orig_obj)] = obj_name
print 'type of object: '+str(type(orig_obj))
print 'object: '+str(orig_obj)
print 'object, longer description:\n'+min_informative_str(orig_obj, indent_level = 1)
t1 = time.time()
s = cPickle.dumps(orig_obj)
t2 = time.time()
prev_ts = t2 - t1
prev_bytes = len(s)
print 'orig_obj bytes: \t\t\t\t'+str(prev_bytes)
t1 = time.time()
x = cPickle.loads(s)
t2 = time.time()
prev_t = t2 - t1
print 'orig load time: '+str(prev_t)
print 'orig save time: '+str(prev_ts)
nodes = func.maker.fgraph.toposort()
count = 0
for node in nodes:
if str(type(node.op)).lower().find('hostfrom') != -1:
count += 1
found = 0
for ipt in node.inputs:
if ipt.owner is not None and str(type(ipt.owner.op)).lower().find('hostfrom') != -1:
found += 1
try:
print ipt.ndim,'dimensions'
except:
print 'no ndm'
print min_informative_str(ipt)
if found > 0:
print type(node.op), found
try:
print '\t',type(node.op.scalar_op)
except:
pass
print count
"""
i = 58
for key in mf1mod.hidden_layers[0].transformer.get_params():
func = function([Xb,yb,alpha], updates[key], on_unused_input = 'ignore')
nodes = func.maker.fgraph.toposort()
count = 0
for node in nodes:
if str(type(node.op)).lower().find('hostfrom') != -1:
count += 1
found = 0
for ipt in node.inputs:
if ipt.owner is not None and str(type(
ipt.owner.op)).lower().find('hostfrom') != -1:
found += 1
try:
print ipt.ndim, 'dimensions'
except:
print 'no ndm'
print min_informative_str(ipt)
if found > 0:
print type(node.op), found
try:
print '\t', type(node.op.scalar_op)
except:
pass
print count
test = CIFAR10(which_set='test', one_hot=True, gcn=55.)
yl = T.argmax(yb, axis=1)
mf1acc = 1. - T.neq(yl, T.argmax(ymf1, axis=1)).mean()
#mfnacc = 1.-T.neq(yl , T.argmax(mfny,axis=1)).mean()
X = space.make_theano_batch()
X.tag.test_value = space.get_origin_batch(m).astype(X.dtype)
inputs = [X]
history = model.mf(X, return_history=True)
for elem in history:
assert isinstance(elem, (list, tuple))
assert len(elem) == len(model.hidden_layers)
outputs = [elem[-1] for elem in history]
for elem in outputs:
for value in get_debug_values(elem):
if value.shape[0] != m:
print 'culprint is',id(elem)
print min_informative_str(elem)
quit(-1)
f = function(inputs, outputs)
n_classes = model.hidden_layers[-1].n_classes
if isinstance(n_classes, float):
assert n_classes == int(n_classes)
n_classes = int(n_classes)
assert isinstance(n_classes, int)
templates = np.zeros((n_classes, space.get_total_dimension()))
for i in xrange(n_classes):
for j in xrange(-1, -dataset.X.shape[0], -1):
if dataset.y[j,i]:
templates[i, :] = dataset.X[j, :]
do_one_V_update()
do_one_optimized_H_update()
cur_V = V
cur_H = H
for i in xrange(num_rounds):
cur_V = update_V(cur_H)
cur_H = update_H(cur_V)
print 'Compiling unrolled theano'
unrolled_theano = function([], updates={V: cur_V, H: cur_H})
from theano.printing import min_informative_str
print min_informative_str(unrolled_theano.maker.env.outputs[0])
assert False
def unrolled_loop():
init()
unrolled_theano()
print 'Timing python loop'
run_timed_trial(python_loop)
print 'Timing unrolled loop'
run_timed_trial(unrolled_loop)
print 'Timing optimized python loop'
elif field.startswith('.'):
obj_name += '.' + field
orig_obj = getattr(orig_obj, field[1:])
else:
obj_name + '[' + field + ']'
orig_obj = orig_obj[eval(field)]
if id(orig_obj) in cycle_check:
print("You're going in circles, " + obj_name + " is the same as " +
cycle_check[id(orig_obj)])
quit()
cycle_check[id(orig_obj)] = obj_name
print('type of object: ' + str(type(orig_obj)))
print('object: ' + str(orig_obj))
print('object, longer description:\n' +
min_informative_str(orig_obj, indent_level=1))
t1 = time.time()
s = cPickle.dumps(orig_obj, hp)
t2 = time.time()
prev_ts = t2 - t1
prev_bytes = len(s)
print('orig_obj bytes: \t\t\t\t' + str(prev_bytes))
t1 = time.time()
x = cPickle.loads(s)
t2 = time.time()
prev_t = t2 - t1
print('orig load time: ' + str(prev_t))
print('orig save time: ' + str(prev_ts))
for i in xrange(num_rounds):
do_one_V_update()
do_one_optimized_H_update()
cur_V = V
cur_H = H
for i in xrange(num_rounds):
cur_V = update_V(cur_H)
cur_H = update_H(cur_V)
print 'Compiling unrolled theano'
unrolled_theano = function([], updates = { V : cur_V, H : cur_H } )
from theano.printing import min_informative_str
print min_informative_str(unrolled_theano.maker.env.outputs[0])
assert False
def unrolled_loop():
init()
unrolled_theano()
print 'Timing python loop'
run_timed_trial(python_loop)
print 'Timing unrolled loop'
run_timed_trial(unrolled_loop)
print 'Timing optimized python loop'
run_timed_trial(optimized_python_loop)
X = space.make_theano_batch()
X.tag.test_value = space.get_origin_batch(m).astype(X.dtype)
inputs = [X]
history = model.mf(X, return_history=True)
for elem in history:
assert isinstance(elem, (list, tuple))
assert len(elem) == len(model.hidden_layers)
outputs = [elem[-1] for elem in history]
for elem in outputs:
for value in get_debug_values(elem):
if value.shape[0] != m:
print 'culprint is', id(elem)
print min_informative_str(elem)
quit(-1)
f = function(inputs, outputs)
n_classes = model.hidden_layers[-1].n_classes
if isinstance(n_classes, float):
assert n_classes == int(n_classes)
n_classes = int(n_classes)
assert isinstance(n_classes, int)
templates = np.zeros((n_classes, space.get_total_dimension()))
for i in xrange(n_classes):
for j in xrange(-1, -dataset.X.shape[0], -1):
if dataset.y[j, i]:
templates[i, :] = dataset.X[j, :]
