def check_on_previous():
""" Test a hypothesis on all previous examples """
t = gl.current_hypothesis
new = t['new']
specialized = t['specialized']
retained = ['retained']
x = utils.merge_all_lists([new, specialized, retained])
cwd = os.getcwd()
testpath = os.path.join(os.path.dirname(cwd), 'runtime', 'test_file.lp')
testfile = open(testpath, 'w')
for y in x:
testfile.write(y.as_string_with_var_types)
testfile.write('\n\n')
testfile.write('#hide.')
testfile.close()
import asp
allok = False
for i in gl.seen_examples:
utils.get_example(i)
if not asp.test_hypothesis(debug=True):
print('unsatisfiable at example %s') % (str(i))
allok = True
if allok:
print('Hypothesis consistent with all past examples')
os.remove(testfile)
def check_on_previous():
""" Test a hypothesis on all previous examples """
t = gl.current_hypothesis
new = t['new']
specialized = t['specialized']
retained = ['retained']
x = utils.merge_all_lists([new,specialized,retained])
cwd = os.getcwd()
testpath = os.path.join(os.path.dirname(cwd),'runtime','test_file.lp')
testfile = open(testpath, 'w')
for y in x:
testfile.write(y.as_string_with_var_types)
testfile.write('\n\n')
testfile.write('#hide.')
testfile.close()
import asp
allok = False
for i in gl.seen_examples:
utils.get_example(i)
if not asp.test_hypothesis(debug=True):
print('unsatisfiable at example %s')%(str(i))
allok = True
if allok:
print('Hypothesis consistent with all past examples')
os.remove(testfile)
def test_default(**kwargs):
import utils
result = None
testfile = test_theory
options = [clingo,bk,ex,testfile,exmpl_constr,'0 --asp09']
command = ' '.join(options)
out = os.popen(command).read().split('. ')
if out[0].strip() == 'UNSATISFIABLE':
result = False
else:
out = set([x.strip() for x in out if not x.strip() == ''])
out = filter(lambda x: 'OPTIMUM FOUND' not in x,out)
if list(out) == [] :
result = True
elif all('posCovered' in x or 'negsCovered' in x for x in out):
(_,_,score) = utils.get_score(out)
if score == gl.current_example_object.positive_count:
return True
else:
return False
else:
raise excps.HypothesisTestingException('ASP reasoner returned %s'%(' '.join(out)),gl.logger)
if 'last_seen' in kwargs:
if result:
print(gl.current_example,'ok')
else:
print(gl.current_example,'Not ok!')
i = kwargs['last_seen']
utils.get_example(i)
return result
def test_default(**kwargs):
import utils
result = None
testfile = test_theory
options = [clingo,bk,ex,testfile,exmpl_constr,'0 --asp09']
command = ' '.join(options)
out = os.popen(command).read().split('. ')
if out[0].strip() == 'UNSATISFIABLE':
result = False
else:
out = set([x.strip() for x in out if not x.strip() == ''])
out = filter(lambda x: 'OPTIMUM FOUND' not in x,out)
if list(out) == [] :
result = True
elif all('posCovered' in x or 'negsCovered' in x for x in out):
(_,_,score) = utils.get_score(out)
if score == gl.current_example_object.positive_count:
return True
else:
return False
else:
raise excps.HypothesisTestingException('ASP reasoner returned %s'%(' '.join(out)),gl.logger)
if 'last_seen' in kwargs:
if result:
print(gl.current_example,'ok')
else:
print(gl.current_example,'Not ok!')
i = kwargs['last_seen']
utils.get_example(i)
return result
def test_all(last_example):
import utils
all_,unsat = True,[]
for i in gl.seen_examples:
utils.get_example(i)
test = test_hypothesis
if not test:
all_ = False
unsat.append(i)
utils.get_example(last_example) # clean up
if all_:
print('Hypothesis ok with all seen examples')
else:
print('Not ok with %s')%(','.join([str(x) for x in unsat]))
return (all_,unsat)
def test_all(last_example):
import utils
all_,unsat = True,[]
for i in gl.seen_examples:
utils.get_example(i)
test = test_hypothesis
if not test:
all_ = False
unsat.append(i)
utils.get_example(last_example) # clean up
if all_:
print('Hypothesis ok with all seen examples')
else:
print('Not ok with %s')%(','.join([str(x) for x in unsat]))
return (all_,unsat)
def get_questions(self):
questions = []
for d in self.data_dicts:
relation = d['relation']
coll = d['collection']
prop = d['property']
scale = self.prop_info[prop][0]['scale']
coll_rel = (coll, relation)
if prop.startswith('made_of'):
coll = 'parts_material'
prop = prop.lstrip('made_of_')
scale = self.prop_info[prop][0]['scale']
if scale == 'T':
coll = coll+'_scale'
coll_rel = (coll, relation)
if coll_rel in self.question_templates:
q_d = dict()
q_d.update(d)
qu_temp = self.question_templates[coll_rel]
# print(prop)
# print(self.prop_info[prop])
cat = self.prop_info[prop][0]['category']
q_d['quid'] = d['quid']
q_d['relation'] = relation
q_d['question'] = utils.create_question(prop, d['concept'], qu_temp, cat)
examples = self.relation_examples_dict[relation][coll]
ex_dict = utils.get_example(examples, qu_temp, self.prop_info, relation)
if 'collection' in ex_dict:
ex_dict.pop('collection')
q_d.update(ex_dict)
# print(q_d['collection'])
# print('---')
questions.append(q_d)
return questions
def display_example(word):
example = get_example(word)
window['output'].print("Word: " + word)
if example:
window['output'].print("Example: ", example)
else:
display_error("Meaning Not Found")
def test_hypothesis(**kwargs):
"""
Test a hypothesis on examples. How to call:
- If called with no kwargs it simply checks a hypothesis on gl.current_example (normal mode)
- If called with kwargs = {example:i}, where i is an integer then its gets example
i from hte database, performs check and cleans up, i.e it restores gl.current_example (debugging mode)
- If called with kwargs = {example:'all'} it checks a hypothesis on all seen examples
and cleans up afterwards. (debugging mode)
"""
import utils
current_example = gl.current_example # remember to clean up later
if kwargs == {}: # default
return test_default()
else:
if 'example' in kwargs: # perform a test with a specific example (for debugging)
i = kwargs['example']
if utils.isint(i):
utils.get_example(i)
return test_default(last_seen = current_example)
else: # then we want to check all seen examples for correctness (argument = 'all')
return test_all(current_example)
def test_hypothesis(**kwargs):
"""
Test a hypothesis on examples. How to call:
- If called with no kwargs it simply checks a hypothesis on gl.current_example (normal mode)
- If called with kwargs = {example:i}, where i is an integer then its gets example
i from hte database, performs check and cleans up, i.e it restores gl.current_example (debugging mode)
- If called with kwargs = {example:'all'} it checks a hypothesis on all seen examples
and cleans up afterwards. (debugging mode)
"""
import utils
current_example = gl.current_example # remember to clean up later
if kwargs == {}: # default
return test_default()
else:
if 'example' in kwargs: # perform a test with a specific example (for debugging)
i = kwargs['example']
if utils.isint(i):
utils.get_example(i)
return test_default(last_seen = current_example)
else: # then we want to check all seen examples for correctness (argument = 'all')
return test_all(current_example)
def main():
args = utils.get_args()
print("Prepare dataset...")
mnist = input_data.read_data_sets("mnist/", one_hot = True)
with tf.Graph().as_default(), tf.Session() as session:
autoencoder = Autoencoder(
784, args.hid_shape, args.lat_shape,
optimizer = tf.train.AdagradOptimizer(args.lr),
batch_size = args.batch_size,
dropout = args.dropout)
session.run(tf.initialize_all_variables())
if args.save_model or args.load_model:
saver = tf.train.Saver()
if args.load_model:
try:
saver.restore(session, utils.SAVER_FILE)
except ValueError:
print("Cant find model file")
sys.exit(1)
if args.make_imgs:
index = 0
print("Prepare images directory...")
utils.prepare_image_folder()
example = utils.get_example(args.digit, mnist.test)
print("Start training...")
for epoch in range(args.epoches):
for i, batch in enumerate(utils.gen_data(args.batch_size, mnist.train.images)):
autoencoder.fit_on_batch(session, batch)
if (i+1) % args.log_after == 0:
test_cost = autoencoder.evaluate(session, mnist.test.images)
print("Test error = {0:.4f} on {1} batch in {2} epoch".format(test_cost, i+1, epoch+1))
if args.make_imgs:
path = os.path.join(utils.IMG_FOLDER, "{0:03}.png".format(index))
autoencoded = autoencoder.encode_decode(session, example.reshape(1, 784))
utils.save_image(autoencoded.reshape((28, 28)), path)
index += 1
if args.save_model:
saver.save(session, utils.SAVER_FILE)
print("Model saved")
def get_questions(self):
questions = []
for d in self.data_dicts:
certainty = d['certainty']
label = d['label']
levels = utils.get_levels(label, certainty, self.restrict)
for l in levels:
relations = self.level_dict[l]
for rel in relations:
coll = d['collection']
prop = d['property']
if prop.startswith('made_of'):
coll = 'parts_material'
prop = prop.lstrip('made_of_')
scale = self.prop_info[prop][0]['scale']
if scale == 'T':
coll = coll+'_scale'
coll_rel = (coll, rel)
if coll_rel in self.question_templates:
q_d = dict()
q_d.update(d)
qu_temp = self.question_templates[coll_rel]
# print(prop)
# print(self.prop_info[prop])
cat = self.prop_info[prop][0]['category']
q_d['quid'] = uuid.uuid4()
q_d['relation'] = rel
q_d['question'] = utils.create_question(prop, d['concept'], qu_temp, cat)
examples = self.relation_examples_dict[rel][coll]
ex_dict = utils.get_example(examples, qu_temp, self.prop_info, rel)
if 'collection' in ex_dict:
ex_dict.pop('collection')
q_d.update(ex_dict)
# print(q_d['collection'])
# print('---')
questions.append(q_d)
# else:
# print('not found:,', coll_rel)
return questions
def learn(**kwargs):
u.clear_prior()
found_new_clauses = False
hs,scs,cls,incsolve = False,False,False,False
(step,up_to_exmpl) = utils.get_time_step(gl)
time_interval = (i for i in range(0,up_to_exmpl*step+1,step) )
(newclauses,retained,specialized) = ([],[],[])
if 'heuristic_search' in kwargs and kwargs['heuristic_search'] : hs = True
if 'set_cover_search' in kwargs and kwargs['set_cover_search'] : scs = True
if 'clause_level_search' in kwargs and kwargs['clause_level_search'] : cls = True
if 'incremental_solve' in kwargs and kwargs['incremental_solve'] : incsolve = True
u.post_exmpl_constraints(heuristic_search=hs,set_cover_search=scs,clause_level_search=cls)
"""
for i in range(1,600):
if utils.get_example(i):
gl.current_example = i
#if not asp.test_hypothesis():
print(i)
asp.show_negs()
print('here')
"""
#for i in time_interval:
#for i in range(1,100000):
for i in range(1,1000000):
#for i in (110,120):
#i = 250
#print(i)
if i == 1672:
stop = 'stop'
if utils.get_example(i):
gl.current_example = i
if not asp.test_hypothesis():
print(i)
print('revising')
(newclauses,retained,specialized) = \
u.revise(is_new_example=True,
debug=False,
newcl=newclauses,
refcl=specialized,
retcl=retained,heuristic_search=hs,
set_cover_search=scs,
clause_level_search=cls,
incremental_solve=incsolve)
found_new_clauses = newclauses != []
else:
print(i)
print('correct')
if gl.current_example_object.has_positives:
u.update_support([newclauses,
retained,
specialized],
simple_update=True)
if found_new_clauses:
for j in gl.seen_examples:
utils.get_example(j)
if not asp.test_hypothesis():
(newclauses,retained,specialized) = \
u.revise(is_new_example=False,
debug=False,
newcl=newclauses,
refcl=specialized,
retcl=retained,heuristic_search=hs,
set_cover_search=scs,
clause_level_search=cls,
incremental_solve=incsolve)
else: pass
u.tideup(newclauses,retained,specialized,
kernel='clear',delta='clear',
usedict='clear',updatexmpl=i)
found_new_clauses = False
def learn(**kwargs):
u.clear_prior()
found_new_clauses = False
hs, scs, cls, incsolve = False, False, False, False
(step, up_to_exmpl) = utils.get_time_step(gl)
time_interval = (i for i in range(0, up_to_exmpl * step + 1, step))
(newclauses, retained, specialized) = ([], [], [])
if 'heuristic_search' in kwargs and kwargs['heuristic_search']: hs = True
if 'set_cover_search' in kwargs and kwargs['set_cover_search']: scs = True
if 'clause_level_search' in kwargs and kwargs['clause_level_search']:
cls = True
if 'incremental_solve' in kwargs and kwargs['incremental_solve']:
incsolve = True
u.post_exmpl_constraints(heuristic_search=hs,
set_cover_search=scs,
clause_level_search=cls)
"""
for i in range(1,600):
if utils.get_example(i):
gl.current_example = i
#if not asp.test_hypothesis():
print(i)
asp.show_negs()
print('here')
"""
#for i in time_interval:
#for i in range(1,100000):
for i in range(1, 1000000):
#for i in (110,120):
#i = 250
#print(i)
if i == 1672:
stop = 'stop'
if utils.get_example(i):
gl.current_example = i
if not asp.test_hypothesis():
print(i)
print('revising')
(newclauses,retained,specialized) = \
u.revise(is_new_example=True,
debug=False,
newcl=newclauses,
refcl=specialized,
retcl=retained,heuristic_search=hs,
set_cover_search=scs,
clause_level_search=cls,
incremental_solve=incsolve)
found_new_clauses = newclauses != []
else:
print(i)
print('correct')
if gl.current_example_object.has_positives:
u.update_support([newclauses, retained, specialized],
simple_update=True)
if found_new_clauses:
for j in gl.seen_examples:
utils.get_example(j)
if not asp.test_hypothesis():
(newclauses,retained,specialized) = \
u.revise(is_new_example=False,
debug=False,
newcl=newclauses,
refcl=specialized,
retcl=retained,heuristic_search=hs,
set_cover_search=scs,
clause_level_search=cls,
incremental_solve=incsolve)
else:
pass
u.tideup(newclauses,
retained,
specialized,
kernel='clear',
delta='clear',
usedict='clear',
updatexmpl=i)
found_new_clauses = False
def train_net(self):
"""
trains the restoration net
"""
net = self.build_net()
print("### net created successfully ###")
# cost function
cost_average_pixels = tf.reduce_mean(tf.square(self.label - self.out))
cost_similar_filters = tf.reduce_mean(
tf.square(
tf.stack(self.gradients) -
tf.reduce_mean(tf.stack(self.gradients))))
cost = cost_average_pixels + self.alpha * cost_similar_filters
optimizer = tf.train.AdamOptimizer(
learning_rate=self.learning_rate).minimize(cost)
saver = tf.train.Saver()
loss_graph = []
loss_a = []
loss_b = []
loss_g = []
with tf.Session() as sess:
print("### initializing variables ###")
sess.run(tf.global_variables_initializer())
gen = utils.get_example(self.path, self.num_images)
print("### strating the training ###")
for iter_index in range(self.num_epochs):
for index in range(self.num_images):
(x, y) = next(gen)
fd = {}
for i in range(len(x[0])):
fd[self.convs[i]] = x[0][i]
for i in range(len(x[1])):
fd[self.pool_dims[i]] = x[1][i]
fd[self.label] = y
_, c, c_a, c_b, c_g = sess.run([
optimizer, cost, cost_average_pixels,
cost_similar_filters, self.gradients
],
feed_dict=fd)
loss_graph.append(c)
loss_a.append(c_a)
loss_b.append(c_b)
loss_g.append(c_g)
if iter_index % 50 == 0 or iter_index == self.num_epochs - 1:
print("Finished iteration " + str(iter_index + 1) +
" out of " + str(self.num_epochs) + " with loss: " +
str(c))
print("pixelwise: %d", c_a)
print("gradients: %d", c_b)
print(c_g)
print("### finished training ###")
save_path = saver.save(sess, self.model_location)
print("Model saved in file: %s" % save_path)
np.save("stats/loss_total.npy", loss_graph)
np.save("stats/loss_a.npy", loss_a)
np.save("stats/loss_b.npy", loss_b)
np.save("stats/loss_g.npy", loss_g)
