def train(self, save_path, log, form, dep, lang, examples, dim, batch,
epoch):
# TODO train for a parameter M
exam = []
for a, v1, v2 in examples:
if a not in self.vocabulary or v1 not in self.vocabulary or v2 not in self.vocabulary:
continue
vec_a = self.embeddings[a]
vec_v1 = self.embeddings[v1]
vec_v2 = self.embeddings[v2]
exam.append((tensor(vec_a), tensor(vec_v1), tensor(vec_v2)))
self.para = M.train(save_path, log, form, dep, lang, exam, dim, batch,
epoch)
return self.para
#DaydayUp1
import MyDay as M
'''
def dayUP(df):
dayup=1
for i in range(365):
if i % 7 in[6,0]:
dayup=(1-0.01)*dayup
else:
dayup=(1+df)*dayup
return dayup
'''
dayfactory = 0.001
while M(dayfactory) < 37.78:
dayfactory += 0.001
print('{:.3f}'.format(dayfactory))
#!/usr/bin/env python
if __name__ == '__main__':
import M
import csv
import numpy as np
rows = list(csv.reader(open('../csv/train.csv')))
x, e = M.getColumns(rows, (0, 1, 3, 4), (int, float, lambda x: 1 if x=='male' else 0, float))
x = np.matrix(x)
y = x[0,:] #class
x = x[1:,:] #attributes
pca = M.PCA(x)
pca.plot(x, c = ['r' if i == 0 else 'b' for i in list(np.array(y)[0])])
def main():
m=M()
t1=T1(m)
t2=T2(m)
t1.join()
t2.join()
import M
print (M.hello())
print (M.hello("Manu"))
c = M.C1 (2)
print (c.method ("foo"))
print (c.id)
print (c.name)
c.prop = 5 # Calling setter
print c.prop # Calling getter
try:
print c.wo_prop # Error
except AttributeError:
print "Expected exception when getting wo_prop"
import M
print(M.hello())
print(M.hello("Manu"))
print("%1.2f" % M.print_float(1.2))
c = M.C1 (2)
print(c.method ("foo"))
print(c.id)
print(c.name)
c.prop = 5 # Calling setter
print(c.prop) # Calling getter
try:
print(c.wo_prop) # Error
except AttributeError:
print("Expected exception when getting wo_prop")
l = M.MyList((1, 2))
print(l)
print(l.dump())
def count_finite_det(design, one_purpose):
return np.linalg.det(M(purpose=one_purpose, design=design,
kernel=kernel, p=p, h=h))
#!/usr/bin/env python
if __name__ == '__main__':
import M
import csv
import numpy as np
rows = list(csv.reader(open('../csv/train.csv')))
'''
x, e = M.getColumns(rows, (0, 3, (int, lambda x: 1 if x=='male' else 0)))
print(M.count(x))
'''
x, e = M.getColumns(rows, (0, 1, 3, 4), (int, float, lambda x: 1 if x=='male' else 0, float))
x = np.matrix(x)
y = x[0,:] #class
x = x[1:,:] #attributes
x0, x1 = M.split(x, y, lambda _y: _y==0)
'''
success, failure = M.crossValidation(x0, x1, M.Fisher)
print(success, failure)
'''
d = M.Gauss(x0, x1)
rows = list(csv.reader(open('../csv/test.csv')))[1:]
for row in rows:
try:
x = []
x.append(float(row[0]))
x.append(1 if row[2] == 'male' else 0)
x.append(float(row[3]))
print(d.do(np.matrix(x).T))
except ValueError:
def func():
...do something else...
# O.py
from M import func
from N import func # This overwites the one we got from M
func() # Calls N.func only
# O.py
import M, N # Get the whole modules, not their names
M.func() # We can call both names now
N.func() # The module names make them unique
### file: module2.py
print('starting to load...')
import sys
name = 42
def func(): pass
class klass: pass
print('done loading.')
def self_play(storage, player1, player2=None, explore=True, num_games=1, joseki=False):
for n in range(num_games):
if not parallell:
print("Self-play game: %s" % n)
#Handle the fact that ordinary self-play uses a single tree structure
#whereas evaluation uses two different ones
if player2 != None: evaluation = True
else: evaluation = False
#Initialize game structure
game = santorini.Game()
#Initialize players with networks and tree structures. Make the structures
#globally available to facilitiate inspection or debugging
p1 = M.MCTS(game, player1, sess, explore)
global P1
P1 = p1
if player2 != None:
evaluation = True
p2 = M.MCTS(game, player2, sess, explore)
global P2
P2 = p2
players = [p1, p2]
else:
evaluation = False
#Store state history, but don't add it to global history yet as we need
#to know the outcome first
temp_history = []
done = False
while done == False:
if evaluation:
player = game.turn_count%2
tree = players[player]
other_tree = players[(player+1)%2]
else:
tree = p1
#Execute tree search and make move
t0 = time.time()
done = tree.consider_resigning(v_resign, observe_games)
a, pi_s, P, v = tree.run_simulation(search_depth)
temp_history.extend([[game.stack_s(), pi_s, game.legal_moves(binaryV=True)]])
if evaluation: #This is not very neat, and I should fix it up at some point...
other_tree.prepare_adversarial_move(a)
game.move(a)
done = game.done
if evaluation:
other_tree.finish_adversarial_move(a)
tree.prepare_next_move()
if observe_games:
for i in range(10):
print("\n")
print("P (predicted tree search probs):\n%s\n\n" % np.reshape(P, [5,5]),
"pi (actual tree search probs):\n%s\n\n" % np.reshape(pi_s, [5,5]),
"v: %s\n" % v,
"Chosen move: %s\n" % a,
"Overall game state:\n%s\n\n" % game.render())
print("time: ", time.time()-t0)
z = game.outcome
#store data
t = len(temp_history)
for entry in temp_history:
storage.add(entry[0], entry[1], discount_rs(z, t), entry[2])
t -= 1
return z
# import가 필요한 경우
# M.py
def func():
...무엇인가를 수행함...
# N.py
def func():
...다른 무엇인가를 수행함...
# O.py
from M import func
from N import func # M으로부터 가져온 func를 덮어씀
func() # N.func만 호출함
# O.py
import M, N # 모듈의 이름들이 아니라, 전체 모듈을 가져옴
M.func() # 이제 두 이름 모두 호출할 수 있음
N.func() # 모듈 이름이 두 이름을 유일하게 만들어줌
# O.py
from M import func as mfunc # "as"를 이용하여 이름을 유일하게 재명명
from N import func as nfunc
mfunc(), nfunc(); # 하나 또는 나머지의 하나를 호출함
def calc_gpu_fraction(fraction_string):
idx, num = fraction_string.split('/')
idx, num = float(idx), float(num)
fraction = 1 / (num - idx + 1)
print "[*] GPU : %.4f" % fraction
return fraction
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = calc_gpu_fraction('1/3')
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
# Instantiate simulators for synthetic gradient and synthetic input
gradient_simulator = M(output_dimension, num_parameters, n_layer, sess)
input_simulator = I(IMAGE_PIXELS * IMAGE_PIXELS, input_dimension, n_layer,
sess)
iteration = 0
while iteration < 20000:
iteration += 1
batch_xs, batch_ys = mnist.train.next_batch(FLAGS.batch_size)
if n_layer != 1:
syn_input_val = input_simulator.get_syn_input(batch_xs, iteration)
else:
syn_input_val = batch_xs.astype(np.float32)
syn_input_val = append_ones(syn_input_val)