def agent_rng(self): # Swap to the agent RNG state if necessary, then return random if not self.rng_for_agent: self.internal_state = random.getstate() random.setstate(self.agent_state) self.rng_for_agent = True return random
def test_choice_distribution(self):
from faker.utils.distribution import choice_distribution
a = ('a', 'b', 'c', 'd')
p = (0.5, 0.2, 0.2, 0.1)
sample = choice_distribution(a, p)
self.assertTrue(sample in a)
with open(os.path.join(TEST_DIR, 'random_state.json'), 'r') as fh:
random_state = json.load(fh)
random_state[1] = tuple(random_state[1])
random.setstate(random_state)
samples = [choice_distribution(a, p) for i in range(100)]
a_pop = len([i for i in samples if i == 'a'])
b_pop = len([i for i in samples if i == 'b'])
c_pop = len([i for i in samples if i == 'c'])
d_pop = len([i for i in samples if i == 'd'])
boundaries = []
tolerance = 5
for probability in p:
boundaries.append([100 * probability + tolerance, 100 * probability - tolerance])
self.assertTrue(boundaries[0][0] > a_pop > boundaries[0][1])
self.assertTrue(boundaries[1][0] > b_pop > boundaries[1][1])
self.assertTrue(boundaries[2][0] > c_pop > boundaries[2][1])
self.assertTrue(boundaries[3][0] > d_pop > boundaries[3][1])
def gradcheck(f, x):
"""
Gradient check for a function f
- f should be a function that takes a single argument and outputs the cost
- x is the point (numpy array) to check the gradient at
"""
rndstate = random.getstate()
random.setstate(rndstate)
fx = f(x) # Evaluate function value at original point
h = 1e-4
eps = 1e-5
numgrad = np.zeros_like(x)
# iterate over all indexes in x
it = np.nditer(x, flags=['multi_index'], op_flags=['readwrite'])
while not it.finished:
# evaluate function at x+h
ix = it.multi_index
oldval = x[ix]
x[ix] = oldval + h # increment by h
fxph = f(x) # evalute f(x + h)
x[ix] = oldval - h
fxmh = f(x) # evaluate f(x - h)
x[ix] = oldval # restore
numgrad[ix] = (fxph - fxmh) / (2 * h) # the slope
it.iternext() # step to next dimension
return numgrad
def internal_rng(self): # Swap to the internal RNG state if necessary, then return random if self.rng_for_agent: self.agent_state = random.getstate() random.setstate(self.internal_state) self.rng_for_agent = False return random
def lorem(randseed=None, count=1, method=None):
u"""
Creates Lorem Ipsum text.
Usage format:
{% lorem [randseed] [count] [method] %}
``randseed`` is any hashable object used to initialize the random numbers generator.
If ``randseed`` is not given the common "Lorem ipsum dolor sit..." text is used.
``count`` is a number of paragraphs or sentences to generate (default is 1).
``method`` is either ``p`` for HTML paragraphs enclosed in ``<p>`` tags, or ``b`` for
plain-text paragraph blocks (default is ``b``).
Notice: This filter is rewrited ``lorem`` filter from ``webdesign`` modul from default Django
package ``django.contrib.webdesign``. The original ``lorem`` filter does not give stable random
text, thus its generated paragraphs change on every page refresh. We stabilize the generated
text by setting a fixed randseed before generating the paragraph.
"""
state = random.getstate()
random.seed(randseed)
res = paragraphs(count, common=(randseed is None))
random.setstate(state)
if method == u'p':
res = [u'<p>{}</p>'.format(p) for p in res]
return u'\n'.join(res)
def start_again(self):
"""Restart the game, preserving state. Save a new virtual keyboard,
and add the most recent set of commands to the army of ghost ships.
"""
# Save a new virtual keyboard.
self.keyboard.history += [(self.keyboard.time, sf.Key.SPACE, 'u')]
self.virtual_keyboards += [VirtualKeyboard(self.keyboard)]
# Switch over the last bot to running on the virtual keyboard.
self.players[-1].reset(self.virtual_keyboards[-1],
self.view_size / 2.0)
self.players += [Ship(self.keyboard, self.view_size / 2.0)]
# Reset all of the keyboards, real and virtual.
for vkb in self.virtual_keyboards:
vkb.reset()
self.keyboard.reset()
# Reset player locations
for player in self.players:
player.reset(None, self.view_size / 2.0)
self.players[-1].is_clone = False
self.bullets = []
random.setstate(self.rand_state)
self.asteroids = [Asteroid(self, clear_zone = self.view_size / 2.0,
seed = random.random())
for i in range(self.num_asteroids)]
def __makeResistanceToon(self):
if self.resistanceToon:
return
npc = Toon.Toon()
npc.setName(TTLocalizer.ResistanceToonName)
npc.setPickable(0)
npc.setPlayerType(NametagGroup.CCNonPlayer)
dna = ToonDNA.ToonDNA()
dna.newToonRandom(11237, 'f', 1)
dna.head = 'pls'
npc.setDNAString(dna.makeNetString())
npc.animFSM.request('neutral')
self.resistanceToon = npc
self.resistanceToon.setPosHpr(*ToontownGlobals.CashbotRTBattleOneStartPosHpr)
state = random.getstate()
random.seed(self.doId)
self.resistanceToon.suitType = SuitDNA.getRandomSuitByDept('m')
random.setstate(state)
self.fakeGoons = []
for i in range(self.numFakeGoons):
goon = DistributedCashbotBossGoon.DistributedCashbotBossGoon(base.cr)
goon.doId = -1 - i
goon.setBossCogId(self.doId)
goon.generate()
goon.announceGenerate()
self.fakeGoons.append(goon)
self.__hideFakeGoons()
def test_seeds_off_random():
s = settings(max_shrinks=0, database=None)
r = random.getstate()
x = find(st.integers(), lambda x: True, settings=s)
random.setstate(r)
y = find(st.integers(), lambda x: True, settings=s)
assert x == y
def sgd(f, x0, step, iterations, useSaved = False, PRINT_EVERY=10):
# possibly more arguments for postprocessing, save trained variables,
# print status lines
# Anneal learning rate every several iterations
ANNEAL_EVERY = 20000
if useSaved:
start_iter, oldx, state = load_saved_params()
if start_iter > 0:
x0 = oldx
step *= 0.5 ** (start_iter / ANNEAL_EVERY)
if state:
random.setstate(state)
else:
start_iter = 0
x = x0
for iter in xrange(1, iterations + 1):
cost, grad = f(x)
x -= step * grad
if iter % PRINT_EVERY == 0:
print "iter %d: %f" % (iter, cost)
if iter % SAVE_PARAMS_EVERY == 0 and useSaved:
save_params(iter, x)
if iter % ANNEAL_EVERY == 0:
step *= 0.5
return x
def shuffle(self, i, seed=None):
if seed is not None:
rand_state = random.getstate()
random.seed(seed)
move_list = []
last_face = None
for _ in range(i):
face = last_face
while face == last_face:
face = Face.random()
turn_type = TurnType.random()
move_list.append((face, turn_type))
last_face = face
shuffle_algorithm = Algorithm(reversed(move_list))
self.apply_algorithm(shuffle_algorithm)
if seed is not None:
random.setstate(rand_state)
return shuffle_algorithm
def simplified_data(num_train, num_dev, num_test):
rndstate = random.getstate()
random.seed(0)
trees = loadTrees('train') + loadTrees('dev') + loadTrees('test')
#filter extreme trees
pos_trees = [t for t in trees if t.root.label==4]
neg_trees = [t for t in trees if t.root.label==0]
#binarize labels
binarize_labels(pos_trees)
binarize_labels(neg_trees)
#split into train, dev, test
print len(pos_trees), len(neg_trees)
pos_trees = sorted(pos_trees, key=lambda t: len(t.get_words()))
neg_trees = sorted(neg_trees, key=lambda t: len(t.get_words()))
num_train/=2
num_dev/=2
num_test/=2
train = pos_trees[:num_train] + neg_trees[:num_train]
dev = pos_trees[num_train : num_train+num_dev] + neg_trees[num_train : num_train+num_dev]
test = pos_trees[num_train+num_dev : num_train+num_dev+num_test] + neg_trees[num_train+num_dev : num_train+num_dev+num_test]
random.shuffle(train)
random.shuffle(dev)
random.shuffle(test)
random.setstate(rndstate)
return train, dev, test
def determine_action(state, dict, random_state, score, max_score): #print state random.setstate(random_state[1]) state_click = dict.get((state, True), 0) state_nothing = dict.get((state, False), 0) print state print state_click print state_nothing value = random.randint(1,10) random_state[1] = random.getstate() if value < 3 and score >= max_score: value = random.randint(0,1) random_state[1] = random.getstate() if value == 1: print "RNG VALUE OF 1" return True else: print "RNG VALUE OF 0" return False elif state_click > state_nothing: print "state_click greater than state_nothing" return True print "state_nothing greater than state_click" return False
def load(self, filename):
data = Data.load(filename)
random.setstate(data['random'])
self.tiles = data['tiles']
self.initindexes()
self.popcache = {}
def gradcheck_naive(f, x):
"""
Gradient check for a function f
- f should be a function that takes a single argument and outputs the cost and its gradients
- x is the point (numpy array) to check the gradient at
"""
rndstate = random.getstate()
random.setstate(rndstate)
fx, grad = f(x) # Evaluate function value at original point
h = 1e-4
# Iterate over all indexes in x
it = np.nditer(x, flags=['multi_index'], op_flags=['readwrite'])
while not it.finished:
ix = it.multi_index
### try modifying x[ix] with h defined above to compute numerical gradients
### make sure you call random.setstate(rndstate) before calling f(x) each time, this will make it
### possible to test cost functions with built in randomness later
### YOUR CODE HERE:
raise NotImplementedError
### END YOUR CODE
# Compare gradients
reldiff = abs(numgrad - grad[ix]) / max(1, abs(numgrad), abs(grad[ix]))
if reldiff > 1e-5:
print "Gradient check failed."
print "First gradient error found at index %s" % str(ix)
print "Your gradient: %f \t Numerical gradient: %f" % (grad[ix], numgrad)
return
it.iternext() # Step to next dimension
print "Gradient check passed!"
def movement_phase(process):
repeat = process.current_actions
Action.set_process(process)
rand_state = random.getstate()
random.shuffle(repeat)
random.setstate(rand_state)
repeat.sort()
sorted_moves = []
repeat = [Movement(a, find_shortest_path(process.map, a.pos, a.dest)) for a in repeat]
occupied, que = set(), []
while len(repeat) != len(que):
que, repeat = repeat, []
for turn in que:
if turn.path and turn.path[0] not in occupied:
occupied.add(turn.path[0])
sorted_moves.append(turn.action)
else:
repeat.append(turn)
for turn in repeat:
for node in turn.path:
if node not in occupied:
occupied.add(node)
sorted_moves.append(Action(turn.action.pos, node.pos, turn.action.attack))
break
else:
sorted_moves.append(Action(turn.action.pos, turn.action.pos, turn.action.attack))
return sorted_moves
def main(id, checkpoint_name=None):
# random.seed(64)
if checkpoint_name:
# A file name has been given, then load the data from the file
cp = pickle.load(open(checkpoint_name, "rb"))
pop = cp["population"]
start_gen = cp["generation"] + 1
hof = cp["halloffame"]
logbook = cp["logbook"]
random.setstate(cp["rndstate"])
else:
pop = toolbox.population(n=Config.MU)
start_gen = 0
hof = tools.HallOfFame(1)
logbook = tools.Logbook()
stats = tools.Statistics(lambda ind: ind.fitness.values)
stats.register("avg", np.mean)
stats.register("std", np.std)
stats.register("min", np.min)
stats.register("max", np.max)
pop, log = alg.myEAMuCommaLambda(pop, start_gen, toolbox, Config.MU, Config.LAMBDA,
cxpb=0.6, mutpb=0.2, ngen=Config.ngen,
stats=stats, halloffame=hof, logbook=logbook, verbose=True,
id=id)
return pop, log, hof
def optimize(p, k, distances, mode='clusters', seed=12345, granularity=1.):
if k == 1:
return [p]
random_state = random.getstate()
try:
# we want the same output on every call on the same data, so we use
# a fixed random seed at this point.
random.seed(seed)
clusterer = _Clusterer(
len(p), tuple(range(len(p))), None, [], p, distances.distance)
if isinstance(k, tuple) and len(k) == 2:
criterion = k[0](**k[1])
assert isinstance(criterion, SplitCriterion)
clusters = clusterer.without_k(criterion)
elif isinstance(k, int):
clusters = [c.members for c in clusterer.with_k(k)]
else:
raise ValueError('illegal parameter k "%s"' % str(k))
# sort clusters by order of their first element in the original list.
clusters = sorted(clusters, key=lambda c: c[0])
if mode == 'clusters':
return list(map(lambda c: map(lambda i: p[i], c), clusters))
elif mode == 'components':
return _components(clusters, len(p))
else:
raise ValueError('illegal mode %s' % mode)
finally:
random.setstate(random_state)
def test_unique_filename_exact_match(self):
with execution(None) as ex:
st = random.getstate()
f = touch(ex)
random.setstate(st)
g = touch(ex)
self.assertNotEqual(f, g)
def test_state_determines_filename(self):
with execution(None) as ex:
st = random.getstate()
f = unique_filename_in()
random.setstate(st)
g = unique_filename_in()
self.assertEqual(f, g)
def reify(self, langs):
timing = self._timing.routine('reifying language {}'.format(langs.index(self)))
state = random.getstate()
dictionary = reify(langs, langs.index(self), len(self._changes), set(self._concepts), timing, '')
random.setstate(state)
timing.done()
return dictionary
def sgd(f, x0, step, iterations, postprocessing = None, useSaved = False, PRINT_EVERY=10):
""" Stochastic Gradient Descent """
# Inputs:
# - f: the function to optimize, it should take a single
# argument and yield two outputs, a cost and the gradient
# with respect to the arguments
# - x0: the initial point to start SGD from
# - step: the step size for SGD
# - iterations: total iterations to run SGD for
# - postprocessing: postprocessing function for the parameters
# if necessary.
# - PRINT_EVERY: specifies every how many iterations to output
# Output:
# - x: the parameter value after SGD finishes
# Anneal learning rate every several iterations
ANNEAL_EVERY = 20000
if useSaved:
start_iter, oldx, state = load_saved_params()
if start_iter > 0:
x0 = oldx;
step *= 0.5 ** (start_iter / ANNEAL_EVERY)
if state:
random.setstate(state)
else:
start_iter = 0
x = x0
if not postprocessing:
postprocessing = lambda x: x
expcost = None
for iter in xrange(start_iter + 1, iterations + 1):
cost = None
cost, gradx = f(x)
x += - step*gradx
x = postprocessing(x)
if iter % PRINT_EVERY == 0:
if not expcost:
expcost = cost
else:
expcost = .95 * expcost + .05 * cost
print "iter %d: %f" % (iter, expcost)
if iter % SAVE_PARAMS_EVERY == 0 and useSaved:
save_params(iter, x)
if iter % ANNEAL_EVERY == 0:
step *= 0.5
return x
def __call__(self, raw_bytes, avoid, pcreg=None):
icache_flush = 1
# If randomization is disabled, ensure that the seed
# is always the same for the builder.
state = random.getstate()
if not context.randomize:
random.seed(1)
try:
b = builder.builder()
enc_data = b.enc_data_builder(raw_bytes)
dec_loop = b.DecoderLoopBuilder(icache_flush)
enc_dec_loop = b.encDecoderLoopBuilder(dec_loop)
dec = b.DecoderBuilder(dec_loop, icache_flush)
output, dec = b.buildInit(dec)
output += dec
output += enc_dec_loop
output += enc_data
finally:
random.setstate(state)
return output
def use_internal_state(self):
"""Use a specific RNG state."""
old_state = random.getstate()
random.setstate(self._random_state)
yield
self._random_state = random.getstate()
random.setstate(old_state)
def random_seed(seed):
"""Context manager to set random.seed() temporarily
"""
state = random.getstate()
random.seed(seed)
yield
random.setstate(state)
async def penis(self, ctx, *users: discord.Member):
"""Detects user's penis length
This is 100% accurate.
Enter multiple users for an accurate comparison!"""
if not users:
await self.bot.send_cmd_help(ctx)
return
dongs = {}
msg = ""
state = random.getstate()
for user in users:
random.seed(user.id)
dongs[user] = "8{}D".format("=" * random.randint(0, 30))
random.setstate(state)
dongs = sorted(dongs.items(), key=lambda x: x[1])
for user, dong in dongs:
msg += "**{}'s size:**\n{}\n".format(user.display_name, dong)
for page in pagify(msg):
await self.bot.say(page)
def _numpy_do_teardown():
global _old_python_random_state
global _old_numpy_random_state
random.setstate(_old_python_random_state)
numpy.random.set_state(_old_numpy_random_state)
_old_python_random_state = None
_old_numpy_random_state = None
def testAgainstRandom(p):
state = random.getstate()
score = 0
#random.seed(0)
for iter in range(100):
population = []
nIndividuals = 3
for n in range(nIndividuals):
newIndividual = {}
for i in ['PE', 'LB', 'PK', 'OE', 'RD', 'YW', 'GN', 'DB', 'RR', 'UY',
'PE_up', 'LB_up', 'PK_up', 'OE_up', 'RD_up', 'YW_up', 'GN_up', 'DB_up', 'RR_up', 'UY_up']:
#newIndividual[i] = random.random()
#newIndividual[i] = 1.0
if random.random() > .5:
newIndividual[i] = random.random()
else:
newIndividual[i] = random.random()
population.append(newIndividual)
points = playGame(p, population[0], population[1], population[2])
score += points[0]
random.setstate(state)
return score
def randomOpening(size, seed):
oldstate = random.getstate()
random.seed(seed)
r = random.randint(0, (size*size - 1))
random.setstate(oldstate)
move = str(chr(ord('a') + (r / size))) + str((r % size) + 1)
return move
def spatial_graph_variable_spatial_scale(cell_positions,
spatial_scale=1.,
connection_probability=connection_probability_vervaeke_2010,
synaptic_weight=synaptic_weight_vervaeke_2010):
state = random.getstate()
g_2010 = spatial_graph_2010(cell_positions)
weights_2010 = [e[2]['weight'] for e in g_2010.edges(data=True)]
total_weight_2010 = sum(weights_2010)
# reset RNG to make sure we will rescale strengths fairly
random.setstate(state)
# generate spatial network with 2010 rules but scaling all distances
n_cells = len(cell_positions)
edges = []
for i, p in enumerate(cell_positions):
for j, q in enumerate(cell_positions[i+1:]):
d = distance(p, q) / spatial_scale
if random.random() < connection_probability(d):
edges.append((i, i+1+j, {'weight': synaptic_weight(d)}))
# rescale weights to keep the same total value across the network
weights = [e[2]['weight'] for e in edges]
total_weight = sum(weights)
for e in edges:
e[2]['weight'] *= total_weight_2010 / total_weight
# create graph object
g = nx.Graph()
g.add_nodes_from(range(n_cells))
for node in g.nodes():
g.node[node]['x'] = cell_positions[node][0]
g.node[node]['y'] = cell_positions[node][1]
g.node[node]['z'] = cell_positions[node][2]
g.add_edges_from(edges)
return g
def get_random_condition(conn):
# hits can be in one of three states:
# jobs that are finished
# jobs that are started but not finished
# jobs that are never going to finish (user decided not to do it)
# our count should be based on the first two so, lets stay anything finished or anything not finished that was started in the last 45 minutes should be counted
starttime = datetime.datetime.now() + datetime.timedelta(minutes=-30)
s = select([participantsdb.c.cond], and_(participantsdb.c.codeversion==CODE_VERSION, or_(participantsdb.c.endhit!=null, participantsdb.c.beginhit>starttime)), from_obj=[participantsdb])
result = conn.execute(s)
counts = [0]*NUMCONDS
# Excluding less interesting conditions:
counts[2] = 5000
counts[4] = 5000
for row in result:
counts[row[0]]+=1
# choose randomly from the ones that have the least in them (so will tend to fill in evenly)
indicies = [i for i, x in enumerate(counts) if x == min(counts)]
rstate = getstate()
seed()
if TESTINGPROBLEMSIX:
#subj_cond = choice([5,11])
subj_cond = 5
else:
subj_cond = choice(indicies)
setstate(rstate)
return subj_cond
def solve_again(self, new_sentences, **options):
randstate = random.getstate()
self.engine.start_again(new_sentences, **options)
self.result = self.engine.get_result()
random.setstate(randstate)
return self.result
def sgd(f,
x0,
step,
iterations,
postprocessing=None,
use_saved=False,
print_every=10):
""" Stochastic Gradient Descent
Implement the stochastic gradient descent method in this function.
Arguments:
f -- the function to optimize, it should take a single
argument and yield two outputs, a loss and the gradient
with respect to the arguments
x0 -- the initial point to start SGD from
step -- the step size for SGD
iterations -- total iterations to run SGD for
postprocessing -- postprocessing function for the parameters
if necessary. In the case of word2vec we will need to
normalize the word vectors to have unit length.
print_every -- specifies how many iterations to output loss
Return:
x -- the parameter value after SGD finishes
"""
# Anneal learning rate every several iterations
anneal_every = 20000
if use_saved:
start_iter, oldx, state = load_saved_params()
if start_iter > 0:
x0 = oldx
step *= 0.5**(start_iter / anneal_every)
if state:
random.setstate(state)
else:
start_iter = 0
x = x0
if not postprocessing:
postprocessing = lambda x: x
exploss = None
for iter_ in range(start_iter + 1, iterations + 1):
# You might want to print the progress every few iterations.
loss = None
### YOUR CODE HERE
loss, grad = f(x)
x = x - step * grad
### END YOUR CODE
x = postprocessing(x)
if iter_ % print_every == 0:
if not exploss:
exploss = loss
else:
exploss = .95 * exploss + .05 * loss
print("iter %d: %f" % (iter_, exploss))
if iter_ % SAVE_PARAMS_EVERY == 0 and use_saved:
save_params(iter_, x)
if iter_ % anneal_every == 0:
step *= 0.5
return x
def sgd(f,
x0,
step,
iterations,
postprocessing=None,
useSaved=False,
PRINT_EVERY=10):
""" Stochastic Gradient Descent """
# Implement the stochastic gradient descent method in this
# function.
# Inputs:
# - f: the function to optimize, it should take a single
# argument and yield two outputs, a cost and the gradient
# with respect to the arguments
# - x0: the initial point to start SGD from
# - step: the step size for SGD
# - iterations: total iterations to run SGD for
# - postprocessing: postprocessing function for the parameters
# if necessary. In the case of word2vec we will need to
# normalize the word vectors to have unit length.
# - PRINT_EVERY: specifies every how many iterations to output
# Output:
# - x: the parameter value after SGD finishes
# Anneal learning rate every several iterations
ANNEAL_EVERY = 20000
if useSaved:
start_iter, oldx, state = load_saved_params()
if start_iter > 0:
x0 = oldx
step *= 0.5**(start_iter / ANNEAL_EVERY)
if state:
random.setstate(state)
else:
start_iter = 0
x = x0
if not postprocessing:
postprocessing = lambda x: x
expcost = None
for iter in xrange(start_iter + 1, iterations + 1):
### Don't forget to apply the postprocessing after every iteration!
### You might want to print the progress every few iterations.
cost = None
### YOUR CODE HERE
cost, grad = f(x)
x -= step * grad
x = postprocessing(x)
#raise NotImplementedError
### END YOUR CODE
if iter % PRINT_EVERY == 0:
if not expcost:
expcost = cost
else:
expcost = .95 * expcost + .05 * cost
print "iter %d: %f" % (iter, expcost)
if iter % SAVE_PARAMS_EVERY == 0 and useSaved:
save_params(iter, x)
if iter % ANNEAL_EVERY == 0:
step *= 0.5
return x
def rand_list_of(n):
# 32 extend to 64-bit integers (to avoid overflow in summation
from random import randrange, setstate
init_state = (
3, (2147483648L, 3430835514L, 2928424416L, 3147699060L, 2823572732L,
2905216632L, 1887281517L, 14272356L, 1356039141L, 2741361235L,
1824725388L, 2228169284L, 2679861265L, 3150239284L, 657657570L,
1407124159L, 517316568L, 653526369L, 139268705L, 3784719953L,
2212355490L, 3452491289L, 1232629882L, 1791207424L, 2898278956L,
1147783320L, 1824413680L, 1993303973L, 2568444883L, 4228847642L,
4163974668L, 385627078L, 3663560714L, 320542554L, 1565882322L,
3416481154L, 4219229298L, 315071254L, 778331393L, 3961037651L,
2951403614L, 3355970261L, 102946340L, 2509883952L, 215897963L,
3361072826L, 689991350L, 3348092598L, 1763608447L, 2140226443L,
3813151178L, 2619956936L, 51244592L, 2130725065L, 3867113849L,
1980820881L, 2600246771L, 3207535572L, 257556968L, 2223367443L,
3706150033L, 1711074250L, 4252385224L, 3197142331L, 4139558716L,
748471849L, 2281163369L, 2596250092L, 2804492653L, 484240110L,
3726117536L, 2483815933L, 2173995598L, 3765136999L, 3178931194L,
1237068319L, 3427263384L, 3958412830L, 2268556676L, 360704423L,
4113430429L, 3758882140L, 3743971788L, 1685454939L, 488386L,
3511218911L, 3020688912L, 2168345327L, 3149651862L, 1472484695L,
2011779229L, 1112533726L, 1873931730L, 2196153055L, 3806225492L,
1515074892L, 251489714L, 1958141723L, 2081062631L, 3703490262L,
3211541213L, 1436109217L, 2664448365L, 2350764370L, 1285829042L,
3496997759L, 2306637687L, 1571644344L, 1020052455L, 3114491401L,
2994766034L, 1518527036L, 994512437L, 1732585804L, 2089330296L,
2592371643L, 2377347339L, 2617648350L, 1478066246L, 389918052L,
1126787130L, 2728695369L, 2921719205L, 3193658789L, 2101782606L,
4284039483L, 2704867468L, 3843423543L, 119359906L, 1882384901L,
832276556L, 1862974878L, 1943541262L, 1823624942L, 2146680272L,
333006125L, 929197835L, 639017219L, 1640196300L, 1424826762L,
2119569013L, 4259272802L, 2089277168L, 2030198981L, 2950559216L,
621654826L, 3452546704L, 4085446289L, 3038316311L, 527272378L,
1679817853L, 450787204L, 3525043861L, 3838351358L, 1558592021L,
3649888848L, 3328370698L, 3247166155L, 3855970537L, 1183088418L,
2778702834L, 2820277014L, 1530905121L, 1434023607L, 3942716950L,
41643359L, 310637634L, 1537174663L, 4265200088L, 3126624846L,
2837665903L, 446994733L, 85970060L, 643115053L, 1751804182L,
1480207958L, 2977093071L, 544778713L, 738954842L, 3370733859L,
3242319053L, 2707786138L, 4041098196L, 1671493839L, 3420415077L,
2473516599L, 3949211965L, 3686186772L, 753757988L, 220738063L,
772481263L, 974568026L, 3190407677L, 480257177L, 3620733162L,
2616878358L, 665763320L, 2808607644L, 3851308236L, 3633157256L,
4240746864L, 1261222691L, 268963935L, 1449514350L, 4229662564L,
1342533852L, 1913674460L, 1761163533L, 1974260074L, 739184472L,
3811507072L, 2880992381L, 3998389163L, 2673626426L, 2212222504L,
231447607L, 2608719702L, 3509764733L, 2403318909L, 635983093L,
4233939991L, 2894463467L, 177171270L, 2962364044L, 1191007101L,
882222586L, 1004217833L, 717897978L, 2125381922L, 626199402L,
3694698943L, 1373935523L, 762314613L, 2291077454L, 2111081024L,
3758576304L, 2812129656L, 4067461097L, 3700761868L, 2281420733L,
197217625L, 460620692L, 506837624L, 1532931238L, 3872395078L,
3629107738L, 2273221134L, 2086345980L, 1240615886L, 958420495L,
4059583254L, 3119201875L, 3742950862L, 891360845L, 2974235885L,
87814219L, 4067521161L, 615939803L, 1881195074L, 2225917026L,
2775128741L, 2996201447L, 1590546624L, 3960431955L, 1417477945L,
913935155L, 1610033170L, 3212701447L, 2545374014L, 2887105562L,
2991635417L, 3194532260L, 1565555757L, 2142474733L, 621483430L,
2268177481L, 919992760L, 2022043644L, 2756890220L, 881105937L,
2621060794L, 4262292201L, 480112895L, 2557060162L, 2367031748L,
2172434102L, 296539623L, 3043643256L, 59166373L, 2947638193L,
1312917612L, 1798724013L, 75864164L, 339661149L, 289536004L,
422147716L, 1134944052L, 1095534216L, 1231984277L, 239787072L,
923053211L, 1015393503L, 2558889580L, 4194512643L, 448088150L,
707905706L, 2649061310L, 3081089715L, 3432955562L, 2217740069L,
1965789353L, 3320360228L, 3625802364L, 2420747908L, 3116949010L,
442654625L, 2157578112L, 3603825090L, 3111995525L, 1124579902L,
101836896L, 3297125816L, 136981134L, 4253748197L, 3809600572L,
1668193778L, 4146759785L, 3712590372L, 2998653463L, 3032597504L,
1046471011L, 2843821193L, 802959497L, 3307715534L, 3226042258L,
1014478160L, 3105844949L, 3209150965L, 610876993L, 2563947590L,
2482526324L, 3913970138L, 2812702315L, 4281779167L, 1026357391L,
2579486306L, 402208L, 3457975059L, 1714004950L, 2543595755L,
2421499458L, 478932497L, 3117588180L, 1565800974L, 1757724858L,
1483685124L, 2262270397L, 3794544469L, 3986696110L, 2914756339L,
1952061826L, 2672480198L, 3793151752L, 309930721L, 1861137379L,
94571340L, 1162935802L, 3681554226L, 4027302061L, 21079572L,
446709644L, 1587253187L, 1845056582L, 3080553052L, 3575272255L,
2526224735L, 3569822959L, 2685900491L, 918305237L, 1399881227L,
1554912161L, 703181091L, 738501299L, 269937670L, 1078548118L,
2313670525L, 3495159622L, 2659487842L, 11394628L, 1222454456L,
3392065094L, 3426833642L, 1153231613L, 1234517654L, 3144547626L,
2148039080L, 3790136587L, 684648337L, 3956093475L, 1384378197L,
2042781475L, 759764431L, 222267088L, 3187778457L, 3795259108L,
2817237549L, 3494781277L, 3762880618L, 892345749L, 2153484401L,
721588894L, 779278769L, 3306398772L, 4221452913L, 1981375723L,
379087895L, 1604791625L, 1426046977L, 4231163093L, 1344994557L,
1341041093L, 1072537134L, 1829925137L, 3791772627L, 3176876700L,
2553745117L, 664821113L, 473469583L, 1076256869L, 2406012795L,
3141453822L, 4123012649L, 3058620143L, 1785080140L, 1181483189L,
3587874749L, 1453504375L, 707249496L, 2022787257L, 2436320047L,
602521701L, 483826957L, 821599664L, 3333871672L, 3024431570L,
3814441382L, 416508285L, 1217138244L, 3975201118L, 3077724941L,
180118569L, 3754556886L, 4121534265L, 3495283397L, 700504668L,
3113972067L, 719371171L, 910731026L, 619936911L, 2937105529L,
2039892965L, 3853404454L, 3783801801L, 783321997L, 1135195902L,
326690505L, 1774036419L, 3476057413L, 1518029608L, 1248626026L,
427510490L, 3443223611L, 4087014505L, 2858955517L, 1918675812L,
3921514056L, 3929126528L, 4048889933L, 1583842117L, 3742539544L,
602292017L, 3393759050L, 3929818519L, 3119818281L, 3472644693L,
1993924627L, 4163228821L, 2943877721L, 3143487730L, 4087113198L,
1149082355L, 1713272081L, 1243627655L, 3511633996L, 3358757220L,
3812981394L, 650044449L, 2143650644L, 3869591312L, 3719322297L,
386030648L, 2633538573L, 672966554L, 3498396042L, 3907556L,
2308686209L, 2878779858L, 1475925955L, 2701537395L, 1448018484L,
2962578755L, 1383479284L, 3731453464L, 3659512663L, 1521189121L,
843749206L, 2243090279L, 572717972L, 3400421356L, 3440777300L,
1393518699L, 1681924551L, 466257295L, 568413244L, 3288530316L,
2951425105L, 2624424893L, 2410788864L, 2243174464L, 1385949609L,
2454100663L, 1113953725L, 2127471443L, 1775715557L, 3874125135L,
1901707926L, 3152599339L, 2277843623L, 1941785089L, 3171888228L,
802596998L, 3397391306L, 1743834429L, 395463904L, 2099329462L,
3761809163L, 262702111L, 1868879810L, 2887406426L, 1160032302L,
4164116477L, 2287740849L, 3312176050L, 747117003L, 4048006270L,
3955419375L, 2724452926L, 3141695820L, 791246424L, 524525849L,
1794277132L, 295485241L, 4125127474L, 825108028L, 1582794137L,
1259992755L, 2938829230L, 912029932L, 1534496985L, 3075283272L,
4052041116L, 1125808104L, 2032938837L, 4008676545L, 1638361535L,
1649316497L, 1302633381L, 4221627277L, 1206130263L, 3114681993L,
3409690900L, 3373263243L, 2922903613L, 349048087L, 4049532385L,
3458779287L, 1737687814L, 287275672L, 645786941L, 1492233180L,
3925845678L, 3344829077L, 1669219217L, 665224162L, 2679234088L,
1986576411L, 50610077L, 1080114376L, 1881648396L, 3818465156L,
1486861008L, 3824208930L, 1782008170L, 4115911912L, 656413265L,
771498619L, 2709443211L, 1919820065L, 451888753L, 1449812173L,
2001941180L, 2997921765L, 753032713L, 3011517640L, 2386888602L,
3181040472L, 1280522185L, 1036471598L, 1243809973L, 2985144032L,
2238294821L, 557934351L, 347132246L, 1797956016L, 624L), None)
setstate(init_state)
return [
rffi.r_longlong(randrange(-(1 << 31), (1 << 31) - 1)) for _ in range(n)
]
def not_so_random():
old_state = random.getstate()
random.seed(42)
yield
random.setstate(old_state)
def __exit__(self, *args):
np.random.set_state(self.np_rng_state)
random.setstate(self.rand_rng_state)
def gradcheck_naive(f, x):
""" Gradient check for a function f.
Arguments:
f -- a function that takes a single argument and outputs the
cost and its gradients(f是一个计算损失函数值和梯度的函数)
x -- the point (numpy array) to check the gradient at（数据，gradcheck方法以该数据值为基础，
通过梯度的定义方法对梯度的合理取值范围进行计算）
"""
rndstate = random.getstate()
random.setstate(rndstate)
fx, grad = f(x) # Evaluate function value at original point
h = 1e-4 # Do not change this!
# Iterate over all indexes ix in x to check the gradient.
it = np.nditer(x, flags=['multi_index'], op_flags=['readwrite'])
# np.nditer是numpy array自带的迭代器,flags=['multi_index']表示对a进行多重索引
# op_flags=['readwrite']表示不仅可以对a进行read（读取），还可以write（写入）
while not it.finished:
ix = it.multi_index
# it.multi_index表示输出元素的索引，形式类似(1,1),(1,2),...,(x,y)...
# Try modifying x[ix] with h defined above to compute numerical
# gradients (numgrad).
# Use the centered difference of the gradient.
# It has smaller asymptotic error than forward / backward difference
# methods. If you are curious, check out here:
# https://math.stackexchange.com/questions/2326181/when-to-use-forward-or-central-difference-approximations
# Make sure you call random.setstate(rndstate)
# before calling f(x) each time. This will make it possible
# to test cost functions with built in randomness later.
### YOUR CODE HERE:
x[ix] += h
random.setstate(rndstate)
new_f1 = f(x)[0]
x[ix] -= 2 * h
random.setstate(rndstate)
new_f2 = f(x)[0]
x[ix] += h
numgrad = (new_f1 - new_f2) / (2 * h)
# lim h -->0，这里用1e-4近似
### END YOUR CODE
# Compare gradients
reldiff = abs(numgrad - grad[ix]) / max(1, abs(numgrad), abs(grad[ix]))
if reldiff > 1e-5:
print("Gradient check failed.")
print("First gradient error found at index %s" % str(ix))
print("Your gradient: %f \t Numerical gradient: %f" %
(grad[ix], numgrad))
return
it.iternext() # Step to next dimension
print("Gradient check passed!")
def __exit__(self, type, value, traceback):
setstate(self.state)
def sgd(f, x0, step, iterations, postprocessing=None, useSaved=False,
PRINT_EVERY=10):
""" Stochastic Gradient Descent
Implement the stochastic gradient descent method in this function.
Arguments:
f -- the function to optimize, it should take a single
argument and yield two outputs, a cost and the gradient
with respect to the arguments
x0 -- the initial point to start SGD from
step -- the step size for SGD
iterations -- total iterations to run SGD for
postprocessing -- postprocessing function for the parameters
if necessary. In the case of word2vec we will need to
normalize the word vectors to have unit length.
PRINT_EVERY -- specifies how many iterations to output loss
Return:
x -- the parameter value after SGD finishes
"""
# Anneal learning rate every several iterations
ANNEAL_EVERY = 20000
if useSaved:
start_iter, oldx, state = load_saved_params()
if start_iter > 0:
x0 = oldx
step *= 0.5 ** (start_iter / ANNEAL_EVERY)
if state:
random.setstate(state)
else:
start_iter = 0
x = x0
if not postprocessing:
postprocessing = lambda x: x
expcost = None
for iter in xrange(start_iter + 1, iterations + 1):
# Don't forget to apply the postprocessing after every iteration!
# You might want to print the progress every few iterations.
cost = None
### YOUR CODE HERE
cost, nabla_J = f(x)
del_x = -step * nabla_J
x += del_x
if postprocessing:
x = postprocessing(x)
### END YOUR CODE
if iter % PRINT_EVERY == 0:
if not expcost:
expcost = cost
else:
expcost = .95 * expcost + .05 * cost
print "iter %d: %f" % (iter, expcost)
if iter % SAVE_PARAMS_EVERY == 0 and useSaved:
save_params(iter, x)
if iter % ANNEAL_EVERY == 0:
step *= 0.5
return x
def restore(self):
setstate(self.__state)
def set_rng_state(state):
torch.set_rng_state(state['torch'])
random.setstate(state['random'])
if 'numpy' in state:
import numpy as np
np.random.set_state(state['numpy'])
def main(checkpoint=None, FREQ=2, verbose=True):
"""
This algorithm reproduce the simplest evolutionary algorithm as
presented in chapter 7 of [Back2000]_.
:param checkpoint: string, checkpoint file path
:param FREQ: int, save result to checkpoint every n*FREQ generation
:param verbose: boolean, whether or not to log the statistics
:return: A list of varied individuals that are independent of their
parents.
"""
random.seed(318)
# probability of crossover
cxpb = 0.5
# probability for mutation
mutpb = 0.2
# number of generations
ngen = 2
if checkpoint:
with open(checkpoint, "r") as cp_file:
cp = pickle.load(cp_file)
population = cp["population"]
start_gen = cp["generation"]
halloffame = cp["halloffame"]
logbook = cp["logbook"]
random.setstate(cp["rndstate"])
else:
population = toolbox.population(n=70)
start_gen = 0
halloffame = tools.HallOfFame(maxsize=1)
logbook = tools.Logbook()
# set initial configuration
stats_fit = tools.Statistics(lambda ind: ind.fitness.values)
stats_size = tools.Statistics(len)
mstats = tools.MultiStatistics(fitness=stats_fit, size=stats_size)
mstats.register("avg", np.mean)
mstats.register("std", np.std)
mstats.register("min", np.min)
mstats.register("max", np.max)
logbook.header = ['gen', 'nevals'] + (mstats.fields if mstats else [])
# Evaluate the individuals with an invalid fitness
invalid_ind = [ind for ind in population if not ind.fitness.valid]
fitnesses = toolbox.map(toolbox.evaluate, invalid_ind)
for ind, fit in zip(invalid_ind, fitnesses):
ind.fitness.values = fit
if halloffame is not None:
halloffame.update(population)
record = mstats.compile(population) if mstats else {}
logbook.record(gen=0, nevals=len(invalid_ind), **record)
if verbose:
print logbook.stream
# Begin the generational process
for gen in range(1, ngen+ 1):
# Select the next generation individuals
offspring = toolbox.select(population, len(population))
# Vary the pool of individuals
offspring = algorithms.varAnd(offspring, toolbox, cxpb, mutpb)
# Evaluate the individuals with an invalid fitness
invalid_ind = [ind for ind in offspring if not ind.fitness.valid]
fitnesses = toolbox.map(toolbox.evaluate, invalid_ind)
for ind, fit in zip(invalid_ind, fitnesses):
ind.fitness.values = fit
# Update the hall of fame with the generated individuals
if halloffame is not None:
halloffame.update(offspring)
# Replace the current population by the offspring
population[:] = offspring
# Append the current generation statistics to the logbook
record = mstats.compile(population) if mstats else {}
logbook.record(gen=gen, nevals=len(invalid_ind), **record)
if verbose:
print logbook.stream
if gen % FREQ == 0:
cp = dict(population=population, generation=gen, halloffame=halloffame,
logbook=logbook, rndstate=random.getstate())
with open("checkpoint_name.pkl", "wb") as cp_file:
pickle.dump(cp, cp_file)
return population, logbook
pickle_idx = 0
while not successful_restart:
try:
pickled_pops = glob("pickledPops{}/*".format(SEED))
last_gen = natural_sort(pickled_pops, reverse=True)[pickle_idx]
with open(last_gen, 'rb') as handle:
[optimizer, random_state,
numpy_random_state] = cPickle.load(handle)
successful_restart = True
my_pop = optimizer.pop
my_optimization = optimizer
my_optimization.continued_from_checkpoint = True
my_optimization.start_time = time()
random.setstate(random_state)
np.random.set_state(numpy_random_state)
print "Starting from pickled checkpoint: generation {}".format(
my_pop.gen)
except EOFError:
# something went wrong writing the checkpoint : use previous checkpoint and redo last generation
sub.call("touch IO_ERROR_$(date +%F_%R)", shell=True)
pickle_idx += 1
pass
my_optimization.run(max_hours_runtime=MAX_TIME,
max_gens=GENS,
checkpoint_every=CHECKPOINT_EVERY,
directory=DIRECTORY)
def prepare(self, src: str, dst: str, force_download: bool = False) -> None:
"""See ``rainbow.preparation.preparer.Preparer``."""
# Create the directory for saving the source files.
tf.io.gfile.makedirs(os.path.join(src, self.JOCI["name"]))
# Create the directory for saving the prepared files.
tf.io.gfile.makedirs(os.path.join(dst, self.JOCI["name"]))
src_path = os.path.join(src, self.JOCI["name"], self.JOCI["file_name"])
# Copy the dataset to src_path from the URL.
if not tf.io.gfile.exists(src_path) or force_download:
logger.info(
f"Downloading {self.JOCI['name']} from {self.JOCI['url']}"
f" to {src_path}."
)
preparer_utils.copy_url_to_gfile(self.JOCI["url"], src_path)
with tf.io.gfile.GFile(src_path, "rb") as src_file:
# Verify the dataset file against its checksum.
sha256 = hashlib.sha256()
chunk = None
while chunk != b"":
# Read in 64KB at a time.
chunk = src_file.read(64 * 1024)
sha256.update(chunk)
checksum = sha256.hexdigest()
if checksum != self.JOCI["checksum"]:
raise IOError(
f"The file for {self.JOCI['name']} did not have the"
f" expected checksum. Try running with force_download=True"
f" to redownload all files, or consider updating the"
f" datasets' checksums."
)
# Return to the beginning of the file.
src_file.seek(0)
# Read the data from the JOCI file.
with zipfile.ZipFile(src_file, "r") as src_zip:
with src_zip.open(self.JOCI["csv_path"], "r") as joci_csv:
joci_csv = codecs.getreader("utf-8")(joci_csv)
reader = csv.DictReader(joci_csv)
data = [x for x in reader]
# Prepare and write the splits to dst.
# Shuffle and split the JOCI data.
random_state = random.getstate()
random.seed(rainbow_utils.string_to_seed(self.JOCI["name"]))
random.shuffle(data)
random.setstate(random_state)
for split in self.JOCI["splits"].values():
dst_path = os.path.join(
dst,
self.JOCI["name"],
settings.PREPROCESSED_SPLIT_FILE_NAME_TEMPLATE.format(
split=split["name"], dataset=self.JOCI["name"]
),
)
with tf.io.gfile.GFile(dst_path, "w") as dst_file:
rows_written = 0
writer = csv.DictWriter(
dst_file,
fieldnames=["index", "inputs", "targets"],
dialect="unix",
)
writer.writeheader()
split_data, data = data[: split["size"]], data[split["size"] :]
for i, row_in in enumerate(split_data):
row_out = {
"index": rows_written,
"inputs": (
f"[{self.JOCI['name']}]:\n"
f"<context>{row_in['CONTEXT']}</context>\n"
f"<hypothesis>{row_in['HYPOTHESIS']}</hypothesis>"
),
"targets": row_in["LABEL"],
}
if i == 0:
logger.info(
f"\n\n"
f"Example {row_out['index']} from"
f" {self.JOCI['name']}'s {split['name']} split:\n"
f"inputs:\n"
f"{row_out['inputs']}\n"
f"targets:\n"
f"{row_out['targets']}\n"
f"\n"
)
# Write to the CSV.
writer.writerow(row_out)
rows_written += 1
if rows_written != split["size"]:
logger.error(
f"Expected to write {split.size} rows for the"
f" {split['name']} split of {self.JOCI['name']}, instead"
f" {rows_written} were written."
)
logger.info(f"Finished processing JOCI.")
def patch_and_output(settings, window, spoiler, rom, start):
logger = logging.getLogger('')
logger.info('Patching ROM.')
worlds = spoiler.worlds
cosmetics_log = None
settings_string_hash = hashlib.sha1(
settings.settings_string.encode('utf-8')).hexdigest().upper()[:5]
if settings.output_file:
outfilebase = settings.output_file
elif settings.world_count > 1:
outfilebase = 'OoT_%s_%s_W%d' % (settings_string_hash, settings.seed,
settings.world_count)
else:
outfilebase = 'OoT_%s_%s' % (settings_string_hash, settings.seed)
output_dir = default_output_path(settings.output_dir)
if settings.compress_rom == 'Patch':
rng_state = random.getstate()
file_list = []
window.update_progress(65)
for world in worlds:
if settings.world_count > 1:
window.update_status('Patching ROM: Player %d' %
(world.id + 1))
patchfilename = '%sP%d.zpf' % (outfilebase, world.id + 1)
else:
window.update_status('Patching ROM')
patchfilename = '%s.zpf' % outfilebase
random.setstate(rng_state)
patch_rom(spoiler, world, rom)
cosmetics_log = patch_cosmetics(settings, rom)
rom.update_header()
window.update_progress(65 + 20 *
(world.id + 1) / settings.world_count)
window.update_status('Creating Patch File')
output_path = os.path.join(output_dir, patchfilename)
file_list.append(patchfilename)
create_patch_file(rom, output_path)
rom.restore()
window.update_progress(65 + 30 *
(world.id + 1) / settings.world_count)
if settings.create_cosmetics_log and cosmetics_log:
window.update_status('Creating Cosmetics Log')
if settings.world_count > 1:
cosmetics_log_filename = "%sP%d_Cosmetics.txt" % (
outfilebase, world.id + 1)
else:
cosmetics_log_filename = '%s_Cosmetics.txt' % outfilebase
cosmetics_log.to_file(
os.path.join(output_dir, cosmetics_log_filename))
file_list.append(cosmetics_log_filename)
cosmetics_log = None
if settings.world_count > 1:
window.update_status('Creating Patch Archive')
output_path = os.path.join(output_dir, '%s.zpfz' % outfilebase)
with zipfile.ZipFile(output_path, mode="w") as patch_archive:
for file in file_list:
file_path = os.path.join(output_dir, file)
patch_archive.write(file_path,
file.replace(outfilebase, ''),
compress_type=zipfile.ZIP_DEFLATED)
for file in file_list:
os.remove(os.path.join(output_dir, file))
logger.info("Created patchfile at: %s" % output_path)
window.update_progress(95)
elif settings.compress_rom != 'None':
window.update_status('Patching ROM')
patch_rom(spoiler, worlds[settings.player_num - 1], rom)
cosmetics_log = patch_cosmetics(settings, rom)
window.update_progress(65)
window.update_status('Saving Uncompressed ROM')
if settings.world_count > 1:
filename = "%sP%d.z64" % (outfilebase, settings.player_num)
else:
filename = '%s.z64' % outfilebase
output_path = os.path.join(output_dir, filename)
rom.write_to_file(output_path)
if settings.compress_rom == 'True':
window.update_status('Compressing ROM')
logger.info('Compressing ROM.')
if is_bundled():
compressor_path = "."
else:
compressor_path = "Compress"
if platform.system() == 'Windows':
if 8 * struct.calcsize("P") == 64:
compressor_path += "\\Compress.exe"
else:
compressor_path += "\\Compress32.exe"
elif platform.system() == 'Linux':
if platform.uname()[4] == 'aarch64' or platform.uname(
)[4] == 'arm64':
compressor_path += "/Compress_ARM64"
else:
compressor_path += "/Compress"
elif platform.system() == 'Darwin':
compressor_path += "/Compress.out"
else:
compressor_path = ""
logger.info('OS not supported for compression')
output_compress_path = output_path[:output_path.
rfind('.')] + '-comp.z64'
if compressor_path != "":
run_process(
window, logger,
[compressor_path, output_path, output_compress_path])
os.remove(output_path)
logger.info("Created compressed rom at: %s" % output_compress_path)
else:
logger.info("Created uncompressed rom at: %s" % output_path)
window.update_progress(95)
if not settings.create_spoiler or settings.output_settings:
settings.distribution.update_spoiler(spoiler, False)
window.update_status('Creating Settings Log')
settings_path = os.path.join(output_dir,
'%s_Settings.json' % outfilebase)
settings.distribution.to_file(settings_path, False)
logger.info("Created settings log at: %s" %
('%s_Settings.json' % outfilebase))
if settings.create_spoiler:
settings.distribution.update_spoiler(spoiler, True)
window.update_status('Creating Spoiler Log')
spoiler_path = os.path.join(output_dir,
'%s_Spoiler.json' % outfilebase)
settings.distribution.to_file(spoiler_path, True)
logger.info("Created spoiler log at: %s" %
('%s_Spoiler.json' % outfilebase))
if settings.create_cosmetics_log and cosmetics_log:
window.update_status('Creating Cosmetics Log')
if settings.world_count > 1 and not settings.output_file:
filename = "%sP%d_Cosmetics.txt" % (outfilebase,
settings.player_num)
else:
filename = '%s_Cosmetics.txt' % outfilebase
cosmetic_path = os.path.join(output_dir, filename)
cosmetics_log.to_file(cosmetic_path)
logger.info("Created cosmetic log at: %s" % cosmetic_path)
window.update_progress(100)
if cosmetics_log and cosmetics_log.error:
window.update_status(
'Success: Rom patched successfully. Some cosmetics could not be applied.'
)
else:
window.update_status('Success: Rom patched successfully')
logger.info('Done. Enjoy.')
logger.debug('Total Time: %s', time.process_time() - start)
return worlds[settings.player_num - 1]
def sample(self, num_rows=1, seed=None):
"""Generating samples from vine model."""
s1 = np.random.get_state()
s2 = random.getstate()
np.random.seed(seed)
random.setstate(seed)
unis = np.random.uniform(0, 1, self.n_var)
# randomly select a node to start with
first_ind = random.randint(0, self.n_var - 1)
np.random.seed(s1)
random.setstate(s2)
adj = self.trees[0].get_adjacent_matrix()
visited = []
explore = [first_ind]
sampled = np.zeros(self.n_var)
itr = 0
while explore:
current = explore.pop(0)
neighbors = np.where(adj[current, :] == 1)[0].tolist()
if itr == 0:
new_x = self.ppfs[current](unis[current])
else:
for i in range(itr - 1, -1, -1):
current_ind = -1
if i >= self.truncated:
continue
current_tree = self.trees[i].edges
# get index of edge to retrieve
for edge in current_tree:
if i == 0:
if (edge.L == current and edge.R == visited[0]) or\
(edge.R == current and edge.L == visited[0]):
current_ind = edge.index
break
else:
if edge.L == current or edge.R == current:
condition = set(edge.D)
condition.add(edge.L)
condition.add(edge.R)
visit_set = set(visited).add(current)
if condition.issubset(visit_set):
current_ind = edge.index
break
if current_ind != -1:
# the node is not indepedent contional on visited node
copula_type = current_tree[current_ind].name
copula_para = current_tree[current_ind].param
cop = Bivariate(CopulaTypes(copula_type))
derivative = cop.get_h_function()
# start with last level
if i == itr - 1:
tmp = optimize.fminbound(derivative,
EPSILON,
1.0,
args=(unis[visited[0]],
copula_para,
unis[current]))
else:
tmp = optimize.fminbound(derivative,
EPSILON,
1.0,
args=(unis[visited[0]],
copula_para, tmp))
tmp = min(max(tmp, EPSILON), 0.99)
new_x = self.ppfs[current](tmp)
sampled[current] = new_x
for s in neighbors:
if s not in visited:
explore.insert(0, s)
itr += 1
visited.insert(0, current)
return sampled
def __init__(self,
offline_dataset,
batch_size,
shuffle_batches=True,
num_buckets=None,
shuffle_buckets=True):
if not isinstance(offline_dataset, OfflineDataset):
raise TypeError('Expecting an OfflineDataset instance.')
if not dist.is_available():
raise RuntimeError('Expecting distributed training.')
self._world_size = dist.get_world_size()
self._rank = dist.get_rank()
# Randomly drop a number of traces so that the number of all minibatches in the whole dataset is an integer multiple of world size
num_batches_to_drop = math.floor(
len(offline_dataset._sorted_indices) /
batch_size) % self._world_size
num_traces_to_drop = num_batches_to_drop * batch_size
# Ensure all ranks choose the same traces to drop
st = random.getstate()
random.seed(0)
self._batches = list(
util.chunks(
util.drop_items(list(offline_dataset._sorted_indices),
num_traces_to_drop), batch_size)
) # List of all minibatches, where each minibatch is a list of trace indices
random.setstate(st)
# Discard last minibatch if it's smaller than batch_size
if len(self._batches[-1]) < batch_size:
del (self._batches[-1])
if num_buckets is None:
num_buckets = len(self._batches) / self._world_size
self._num_buckets = num_buckets
self._bucket_size = math.ceil(len(self._batches) / num_buckets)
if self._bucket_size < self._world_size:
raise RuntimeError(
'offline_dataset:{}, batch_size:{} and num_buckets:{} imply a bucket_size:{} smaller than world_size:{}'
.format(len(offline_dataset), batch_size, num_buckets,
self._bucket_size, self._world_size))
# List of buckets, where each bucket is a list of minibatches
self._buckets = list(util.chunks(self._batches, self._bucket_size))
# Unify last two buckets if the last bucket is smaller than other buckets
if len(self._buckets[-1]) < self._bucket_size:
if len(self._buckets) < 2:
raise RuntimeError(
'offline_dataset:{} too small for given batch_size:{} and num_buckets:{}'
.format(len(offline_dataset), batch_size, num_buckets))
self._buckets[-2].extend(self._buckets[-1])
del (self._buckets[-1])
self._shuffle_batches = shuffle_batches
self._shuffle_buckets = shuffle_buckets
self._epoch = 0
self._current_bucket_id = 0
print('DistributedTraceBatchSampler')
print('OfflineDataset size : {:,}'.format(len(offline_dataset)))
print('World size : {:,}'.format(self._world_size))
print('Batch size : {:,}'.format(batch_size))
print('Num. batches dropped: {:,}'.format(num_batches_to_drop))
print('Num. batches : {:,}'.format(len(self._batches)))
print('Bucket size : {:,}'.format(self._bucket_size))
print('Num. buckets : {:,}'.format(self._num_buckets))
def gradcheck_naive(f, x):
"""
Gradient check for a function f
- f should be a function that takes a single argument
and outputs the cost and its gradients
- x is the point (numpy array) to check the gradient at
"""
rndstate = random.getstate()
random.setstate(rndstate)
# Evaluate function value at original point
fx, grad = f(x)
h = 1e-4
test = True
"""
--- about nd.nditer ---
flags = multi_index causes a multi-index, or a tuple of indices
with one per iteration dimension, to be tracked.
op_flags : list of list of str, optional
this is a list of flags for each operand. At minimum, one of readonly,
readwrite, or writeonly must be specified.
readwrite indicates the operand will be read from and written to.
"""
# Iterate over all indexes in x
all_dif = np.array(x, copy=True)
it = np.nditer(x, flags=['multi_index'], op_flags=['readwrite'])
while not it.finished:
ix = it.multi_index
""" try modifying x[ix] with h defined above to compute numerical gradients
make sure you call random.setstate(rndstate) before calling f(x)
each time, this will make it possible to test cost functions with
built in randomness later"""
# YOUR CODE HERE:
x_plus_h = np.array(x, copy=True)
x_plus_h[ix] = x_plus_h[ix] + h
random.setstate(rndstate)
fxh_plus, _ = f(x_plus_h)
x_minus_h = np.array(x, copy=True)
x_minus_h[ix] = x_minus_h[ix] - h
random.setstate(rndstate)
fxh_minus, _ = f(x_minus_h)
numgrad = (fxh_plus - fxh_minus) / (2 * h)
# END YOUR CODE
# Compare gradients
reldiff = abs(numgrad - grad[ix]) / max(1, abs(numgrad), abs(grad[ix]))
all_dif[ix] = reldiff
if reldiff > 1e-5:
test = False
string = """
Your gradient = {0}
Numerical gradient = {1}""".format(grad[ix], numgrad)
print(" " + str(ix) + ": " + string)
print(" fx ={}".format(fx))
print(" fxh_plus ={}".format(fxh_plus))
print(" fxh_minus ={}\n".format(fxh_minus))
# For debugging with a bunch of params is
# useful to add the following:
# else:
# print(str(ix) + ": OK")
# Step to next dimension
it.iternext()
if test:
print("Gradient check passed!")
# add this return to test
return all_dif
def _load(self) -> None:
if not self.load_path:
return
# Backwards compat with older checkpoint formats. List is newest to
# oldest known state_dict locations.
potential_paths = [
["state_dict.pth"],
["determined", "state_dict.pth"],
["pedl", "state_dict.pth"],
["checkpoint.pt"],
]
for ckpt_path in potential_paths:
maybe_ckpt = self.load_path.joinpath(*ckpt_path)
if maybe_ckpt.exists():
checkpoint = torch.load(str(maybe_ckpt),
map_location="cpu") # type: ignore
break
if "model_state_dict" in checkpoint:
# Backward compatible with older checkpoint format.
check.not_in("models_state_dict", checkpoint)
check.eq(len(self.context.models), 1)
self.context.models[0].load_state_dict(
checkpoint["model_state_dict"])
else:
for idx, model in enumerate(self.context.models):
model.load_state_dict(checkpoint["models_state_dict"][idx])
if "optimizer_state_dict" in checkpoint:
# Backward compatible with older checkpoint format.
check.not_in("optimizers_state_dict", checkpoint)
check.eq(len(self.context.optimizers), 1)
self.context.optimizers[0].load_state_dict(
checkpoint["optimizer_state_dict"])
else:
for idx, optimizer in enumerate(self.context.optimizers):
optimizer.load_state_dict(
checkpoint["optimizers_state_dict"][idx])
if "lr_scheduler" in checkpoint:
# Backward compatible with older checkpoint format.
check.not_in("lr_schedulers_state_dict", checkpoint)
check.eq(len(self.context.lr_schedulers), 1)
self.context.lr_schedulers[0].load_state_dict(
checkpoint["lr_scheduler"])
else:
for idx, lr_scheduler in enumerate(self.context.lr_schedulers):
lr_scheduler.load_state_dict(
checkpoint["lr_schedulers_state_dict"][idx])
if "amp_state" in checkpoint:
if self.context._use_amp:
apex.amp.load_state_dict(checkpoint["amp_state"])
else:
logging.warning(
"There exists amp_state in checkpoint but the experiment is not using AMP."
)
else:
if self.context._use_amp:
logging.warning(
"The experiment is using AMP but amp_state does not exist in the checkpoint."
)
if "rng_state" in checkpoint:
rng_state = checkpoint["rng_state"]
np.random.set_state(rng_state["np_rng_state"])
random.setstate(rng_state["random_rng_state"])
torch.random.set_rng_state(rng_state["cpu_rng_state"])
if torch.cuda.device_count():
if "gpu_rng_state" in rng_state:
torch.cuda.set_rng_state(
rng_state["gpu_rng_state"],
device=self.context.distributed.get_local_rank())
else:
logging.warning(
"The system has a gpu but no gpu_rng_state exists in the checkpoint."
)
else:
if "gpu_rng_state" in rng_state:
logging.warning(
"There exists gpu_rng_state in checkpoint but the system has no gpu."
)
else:
logging.warning("The checkpoint has no random state to restore.")
callback_state = checkpoint.get("callbacks", {})
for name in self.callbacks:
if name in callback_state:
self.callbacks[name].load_state_dict(callback_state[name])
elif util.is_overridden(self.callbacks[name].load_state_dict,
pytorch.PyTorchCallback):
logging.warning(
"Callback '{}' implements load_state_dict(), but no callback state "
"was found for that name when restoring from checkpoint. This "
"callback will be initialized from scratch")
test='sars_cov_2_Compound_Viral_interactions_for_Supervised_Learning.csv',
format='csv',
skip_header=
True, # if your csv header has a header, make sure to pass this to ensure it doesn't get proceesed as data!
fields=datafields)
#See how the data looks like
print("Uniprot: ", data.examples[0].uniprot_accession)
print("Protein Sequence: ", data.examples[0].Sequence)
print("Canonical Smiles: ", data.examples[0].canonical_smiles)
print("Inchi Key:", data.examples[0].standard_inchi_key)
print("Target value: ", data.examples[0].pchembl_value)
#Split the data randomly into train, valid and test
train_data, valid_data = data.split(split_ratio=0.9,
random_state=random.setstate(st))
print('Number of training examples: ', len(train_data.examples))
print('Number of validation examples: ', len(valid_data.examples))
print('Number of test examples: ', len(test_data.examples))
#del data
torch.cuda.empty_cache()
# +
#Build the sequence and smiles
TEXT1.build_vocab(train_data, min_freq=2)
TEXT2.build_vocab(train_data, min_freq=1)
LABEL.build_vocab(train_data, min_freq=1)
INDEX1.build_vocab(full_data, min_freq=1)
INDEX2.build_vocab(full_data, min_freq=1)
print('Unique tokens in Sequence vocabulary: ', len(TEXT1.vocab))
def close(self) -> Dict[str, Any]:
'''Return a dict which contains
* **fw-bleu**: fw bleu value.
* **bw-bleu**: bw bleu value.
* **fw-bw-bleu**: harmony mean of fw/bw bleu value.
* **fw-bw-bleu hashvalue**: hash value for fwbwbleu metric, same hash value stands
for same evaluation settings.
'''
res = super().close()
if not self.hyps:
raise RuntimeError(
"The metric has not been forwarded data correctly.")
if not self.reference_test_list:
raise RuntimeError("Reference cannot be empty")
rng_state = random.getstate()
random.seed(self.seed)
sample_hyps = self.hyps.copy()
sample_refs = self.reference_test_list.copy()
random.shuffle(sample_hyps)
random.shuffle(sample_refs)
random.setstate(rng_state)
n_sample_max = max(self.n_sample_hyp, self.n_sample_ref)
sample_hyps = sample_hyps[:n_sample_max]
sample_refs = sample_refs[:n_sample_max]
refs: List[Any]
hyps: List[Any]
if self.tokenizer:
tokenizer: Tokenizer
if isinstance(self.tokenizer, str):
tokenizer = SimpleTokenizer(self.tokenizer)
else:
tokenizer = self.tokenizer
if isinstance(sample_refs[0], List):
ref_sents = [
self.dataloader.convert_ids_to_sentence(
ids, remove_special=True, trim=True)
for ids in sample_refs
]
else:
ref_sents = sample_refs
refs = tokenizer.tokenize_sentences(ref_sents)
hyp_sents = [
self.dataloader.convert_ids_to_sentence(ids,
remove_special=True,
trim=True)
for ids in sample_hyps
]
hyps = tokenizer.tokenize_sentences(hyp_sents)
else:
refs = [
self.dataloader.convert_ids_to_tokens(ids,
remove_special=True,
trim=True)
for ids in sample_refs
]
hyps = [
self.dataloader.convert_ids_to_tokens(ids,
remove_special=True,
trim=True)
for ids in sample_hyps
]
if "unk" in self.dataloader.get_special_tokens_mapping():
refs = replace_unk(
refs,
self.dataloader.get_special_tokens_mapping()["unk"])
bleu_irl_fw, bleu_irl_bw = [], []
weights = np.ones(self.ngram) / self.ngram
n_fw_sample = min(len(hyps), self.n_sample_hyp)
n_fw_reference = min(len(refs), self.n_sample_ref)
tasks = ((refs[:n_fw_reference], hyps[i], weights)
for i in range(n_fw_sample))
pool: Optional[Any] = None
values: Iterable[Any]
if n_fw_sample >= FwBwBleuCorpusMetric.MINIMAL_PARALLEL_N and self.cpu_count > 1:
if not pool:
pool = Pool(self.cpu_count)
values = pool.imap_unordered(_sentence_bleu, tasks, chunksize=20)
else:
values = map(_sentence_bleu, tasks)
if n_fw_sample >= FwBwBleuCorpusMetric.MINIMAL_PARALLEL_N:
values = tqdm.tqdm(values, total=n_fw_sample)
for ans in values:
bleu_irl_fw.append(ans)
n_bw_sample = min(len(refs), self.n_sample_hyp)
n_bw_reference = min(len(hyps), self.n_sample_ref)
tasks = ((hyps[:n_fw_reference], refs[i], weights)
for i in range(n_bw_sample))
if n_bw_sample >= FwBwBleuCorpusMetric.SHOW_PROGRESS and self.cpu_count > 1:
if pool is None:
pool = Pool(self.cpu_count)
values = pool.imap_unordered(_sentence_bleu, tasks, chunksize=20)
else:
values = map(_sentence_bleu, tasks)
if n_bw_sample >= FwBwBleuCorpusMetric.SHOW_PROGRESS:
values = tqdm.tqdm(values, total=n_bw_sample)
for ans in values:
bleu_irl_bw.append(ans)
if pool is not None:
pool.close()
pool.join()
fw_bleu = (1.0 * sum(bleu_irl_fw) / len(bleu_irl_fw))
bw_bleu = (1.0 * sum(bleu_irl_bw) / len(bleu_irl_bw))
if fw_bleu + bw_bleu > 0:
fw_bw_bleu = 2.0 * bw_bleu * fw_bleu / (fw_bleu + bw_bleu)
else:
fw_bw_bleu = 0
res.update({"fw-bleu" : fw_bleu, \
"bw-bleu" : bw_bleu, \
"fw-bw-bleu" : fw_bw_bleu \
})
self._hash_unordered_list(refs)
self._hash_ordered_data((self.ngram, self.seed, n_fw_sample,
n_fw_reference, n_bw_sample, n_bw_reference))
res.update({"fw-bw-bleu hashvalue": self._hashvalue()})
return res
def save_random_state(seed=None):
state = random.getstate()
if seed is not None:
random.seed(seed)
yield
random.setstate(state)
def eaAlphaMuPlusLambdaCheckpoint(
population,
toolbox,
mu,
cxpb,
mutpb,
ngen,
stats=None,
halloffame=None,
cp_frequency=1,
cp_filename=None,
continue_cp=False):
r"""This is the :math:`(~\alpha,\mu~,~\lambda)` evolutionary algorithm
Args:
population(list of deap Individuals)
toolbox(deap Toolbox)
mu(int): Total parent population size of EA
cxpb(float): Crossover probability
mutpb(float): Mutation probability
ngen(int): Total number of generation to run
stats(deap.tools.Statistics): generation of statistics
halloffame(deap.tools.HallOfFame): hall of fame
cp_frequency(int): generations between checkpoints
cp_filename(string): path to checkpoint filename
continue_cp(bool): whether to continue
"""
if continue_cp:
# A file name has been given, then load the data from the file
cp = pickle.load(open(cp_filename, "rb"))
population = cp["population"]
parents = cp["parents"]
start_gen = cp["generation"]
halloffame = cp["halloffame"]
logbook = cp["logbook"]
history = cp["history"]
random.setstate(cp["rndstate"])
else:
# Start a new evolution
start_gen = 1
parents = population[:]
logbook = deap.tools.Logbook()
logbook.header = ['gen', 'nevals'] + (stats.fields if stats else [])
history = deap.tools.History()
# TODO this first loop should be not be repeated !
invalid_count = _evaluate_invalid_fitness(toolbox, population)
_update_history_and_hof(halloffame, history, population)
_record_stats(stats, logbook, start_gen, population, invalid_count)
# Begin the generational process
for gen in range(start_gen + 1, ngen + 1):
offspring = _get_offspring(parents, toolbox, cxpb, mutpb)
population = parents + offspring
invalid_count = _evaluate_invalid_fitness(toolbox, offspring)
_update_history_and_hof(halloffame, history, population)
_record_stats(stats, logbook, gen, population, invalid_count)
# Select the next generation parents
parents = toolbox.select(population, mu)
logger.info(logbook.stream)
if(cp_filename and cp_frequency and
gen % cp_frequency == 0):
cp = dict(population=population,
generation=gen,
parents=parents,
halloffame=halloffame,
history=history,
logbook=logbook,
rndstate=random.getstate())
pickle.dump(cp, open(cp_filename, "wb"))
logger.debug('Wrote checkpoint to %s', cp_filename)
return population, halloffame, logbook, history
def shuffle_in_unision(a,b):
rng_state = random.getstate()
random.shuffle(a)
random.setstate(rng_state)
random.shuffle(b)
for k, v in chance.items():
s = sum(v.values())
if s > biggest:
biggest = s
word = k
prefix = k
for i in v.keys():
chance[k][i] = chance[k][i] / s
if chance[k][i] < epsilon:
epsilon = chance[k][i]
if (cont == 'y'):
with open(sys.argv[1] + '.ls') as data_file:
pq = ast.literal_eval(data_file.readline())
random.setstate(pq)
# zacni generovat
delta = epsilon * 10 # Pravdepodobnost pre prechod z neznameho do znameho stavu
try:
generateWords(prefix, word, delta, epsilon, sys.argv[3])
with open(sys.argv[1] + '.ls', 'w') as outfile:
outfile.write(str(random.getstate()))
print('Session saved', file=sys.stderr)
sys.exit(0)
except KeyboardInterrupt:
with open(sys.argv[1] + '.ls', 'w') as outfile:
outfile.write(str(random.getstate()))
print('Session saved', file=sys.stderr)
sys.exit(0)
def close(self) -> Dict[str, Any]:
'''Return a dict which contains
* **self-bleu**: self-bleu value.
* **self-bleu hashvalue**: hash value for self-bleu metric, same hash value stands
for same evaluation settings.
'''
res = super().close()
if not self.hyps:
raise RuntimeError(
"The metric has not been forwarded data correctly.")
if len(self.hyps) == 1:
raise RuntimeError(
"Selfbleu can't be computed because there is only 1 generated sentence."
)
rng_state = random.getstate()
random.seed(self.seed)
random.shuffle(self.hyps)
random.setstate(rng_state)
n_sample_hyp = min(self.n_sample_hyp, len(self.hyps))
n_sample_ref = min(self.n_sample_ref, len(self.hyps))
ref = self.hyps[:n_sample_ref]
if self.tokenizer:
tokenizer: Tokenizer
if isinstance(self.tokenizer, str):
tokenizer = SimpleTokenizer(self.tokenizer)
else:
tokenizer = self.tokenizer
ref = [
self.dataloader.convert_ids_to_sentence(ids,
remove_special=True,
trim=True)
for ids in ref
]
ref = tokenizer.tokenize_sentences(ref)
else:
ref = [
self.dataloader.convert_ids_to_tokens(ids,
remove_special=True,
trim=True) for ids in ref
]
if "unk" in self.dataloader.get_special_tokens_mapping():
_ref = replace_unk(
ref,
self.dataloader.get_special_tokens_mapping()["unk"])
else:
_ref = ref
bleu_irl = []
weights = np.ones(self.ngram) / self.ngram
tasks = ((ref[:i] + ref[i + 1:], _ref[i], weights)
for i in range(n_sample_hyp))
pool: Optional[Any] = None
values: Iterable[Any]
if n_sample_hyp >= SelfBleuCorpusMetric.MINIMAL_PARALLEL_N and self.cpu_count > 1:
# use multiprocessing
if pool is None:
pool = Pool(self.cpu_count)
values = pool.imap_unordered(_sentence_bleu, tasks, chunksize=20)
else:
values = map(_sentence_bleu, tasks)
if n_sample_hyp >= SelfBleuCorpusMetric.SHOW_PROGRESS:
values = tqdm.tqdm(values, total=n_sample_hyp) # use tqdm
for ans in values:
bleu_irl.append(ans)
if pool is not None:
pool.close()
pool.join()
self._hash_ordered_data((self.seed, n_sample_hyp, n_sample_ref))
res.update({"self-bleu" : 1.0 * sum(bleu_irl) / len(bleu_irl),\
"self-bleu hashvalue": self._hashvalue()})
return res
def set_rng_state(checkpoint: dict[str, Any]) -> None:
if checkpoint:
random.setstate(checkpoint["rng_state"])
np.random.set_state(checkpoint["np_rng_state"])
torch.set_rng_state(checkpoint["torch_rng_state"])
def seed_random(seed):
state = random.getstate()
random.seed(seed)
cleanup(lambda: random.setstate(state))
return RandomSeeder(seed)
if args.resume:
# Load checkpoint.
print('==> Resuming from checkpoint..')
assert os.path.isdir(
args.checkpoint_dir), 'Error: no checkpoint directory found!'
checkpoint = torch.load('./' + args.checkpoint_dir + '/ckpt.best.' +
args.sess + '_' + str(args.seed))
net1 = checkpoint['net1']
net2 = checkpoint['net2']
start_epoch = checkpoint['epoch'] + 1
torch.set_rng_state(checkpoint['rng_state'])
torch.cuda.set_rng_state(checkpoint['cuda_rng_state'])
np.random.set_state(checkpoint['np_state'])
random.setstate(checkpoint['random_state'])
if args.dataset == 'cifar10':
with open("cifar10_labelled_index.pkl", "rb") as fp:
S_idx = pickle.load(fp)
with open("cifar10_unlabelled_index.pkl", "rb") as fp:
U_idx = pickle.load(fp)
else:
with open("svhn_labelled_index.pkl", "rb") as fp:
S_idx = pickle.load(fp)
with open("svhn_unlabelled_index.pkl", "rb") as fp:
U_idx = pickle.load(fp)
else:
patience = 0
start_epoch = 0
start_time = datetime.datetime.now()
if args.resume_last or args.resume_best:
if args.resume_last:
fname = 'saved_models/%s_last.pth' % args.exp_name
else:
fname = 'saved_models/%s_best.pth' % args.exp_name
if os.path.exists(fname):
data = torch.load(fname)
torch.set_rng_state(data['torch_rng_state'])
torch.cuda.set_rng_state(data['cuda_rng_state'])
np.random.set_state(data['numpy_rng_state'])
random.setstate(data['random_rng_state'])
model.load_state_dict(data['state_dict'], strict=False)
optim.load_state_dict(data['optimizer'])
scheduler.load_state_dict(data['scheduler'])
start_epoch = data['epoch'] + 1
best_cider = data['best_cider']
patience = data['patience']
use_rl = data['use_rl']
start_time = data['start_time']
try:
middle_time = data['middle_time']
except:
middle_time = None
print(
'Resuming from epoch %d, validation loss %f, best validation cider %f, patience %d'
% (data['epoch'], data['val_loss'], data['best_cider'],