def path_experiment(p_mat, sources=None, sinks=None, verbose=False):
counts = {}
for i in range(p_mat.shape[0]):
counts[i + 1] = 0
n_paths = 10**4
paths = generate_random_routes(p_mat,
n_paths,
1,
12,
sources=sources,
sinks=sinks)
for path in paths:
for vtx in path:
counts[vtx + 1] += 1
total = sum(list(map(lambda x: counts[x], counts)))
pcts = []
if total == 0:
debug()
for i in range(p_mat.shape[0]):
pcts.append((i + 1, round(100 * counts[i + 1] / total, 2)))
if verbose:
print('Vertex\tPercent of Total Visits')
for row in pcts:
print(row)
else:
return np.array(pcts)
def xferToTarget(cell, target, amount, resource):
target_dir = 'X'
for dir, child in cell.adjacent.iteritems():
if child == target:
target_dir = dir
break
if target_dir == 'X':
# Tried to transfer to a non-adjacent target
debug()
else:
xferResource(cell, target_dir, amount, resource)
def choose_move(self):
# empty board
sleep(0.4)
if self.game.state.moves_on_board==0:
rand1 = randrange(self.game.board_dimension)
rand2 = randrange(self.game.board_dimension)
self.game.move(self, (rand1, rand2))
return
self.timer = time()
next_states = list(self.game.state.next_states())
num_states = len(next_states)
# scores = [0]*len(next_states)
max_score = (-float('inf'), None) # (score, coords)
all_max = []
stop_interval = .03*(self.time_limit/num_states)
# debug()
for index, state in enumerate(next_states):
new_state = GomokuState(self.game.state, (self.symbol, state))
self.node_count+=1
# if time()<(self.timer+self.time_limit):
# s = 'not beyond stop time in choose #{}/{}'.format(index+1, len(next_states))
# print s
# self.print_debug(s)
s = 'branch #{}/{}'.format(index+1, len(next_states))
# self.print_debug(s)
# scores[index]=self.max_value(new_state, stop_time)
# if new_state.is_win_state:
# self.game.move(self, state)
# return
my_max = self.min_value(new_state, time()+stop_interval, -float('inf'), float('inf'), 1)
print 'score: {}'.format(my_max)
all_max.append(my_max)
if my_max>max_score[0]:
max_score = (my_max, state)
all_hugely_negative = True
for thing in all_max:
if thing>-10000000000:
all_hugely_negative = False
if all_hugely_negative:
debug()
# self.game.move(self, (-1,-1))
# return
self.game.move(self,max_score[1])
def add_daughter(self, cell, direction, init_sugar, init_water, newMemory={}):
x, y = self.coordinates[cell]
coordinate = adjacent_coords(x, y, direction)
program = cell.program
if coordinate not in self.coordinates:
newCell = Cell(self, program, newMemory, 'GENERIC', init_sugar, init_water)
self.add_cell(newCell, coordinate)
for adjcell in self.get_adjacent(newCell).itervalues():
adjcell.update_world_state()
return newCell
else:
# Attempted to divide into a spot where a cell already existed
debug()
def initialize(cell):
if cell == 'EMPTY':
print "ERROR: Failed division"
debug()
if 'growth_sugar' not in cell.memory:
cell.memory['growth_sugar'] = cell.sugar_consumption * 20
if 'growth_water' not in cell.memory:
cell.memory['growth_water'] = cell.water_consumption * 20
if 'sugar_children' not in cell.memory:
cell.memory['sugar_children'] = []
if 'water_children' not in cell.memory:
cell.memory['water_children'] = []
cell.memory['initialized'] = 1
if 'demand' not in cell.memory:
cell.memory['demand'] = ((0, 0), (0, 0))
def getRegularlySampledPrfByOffset(self, singlePrfObj, colOffset,
rowOffset):
"""Private function of `getSingleRegularlySampledPrf()`
The 13x13 pixel PRFs on at each grid location are sampled at a 9x9 intra-pixel grid, to
describe how the PRF changes as the star moves by a fraction of a pixel in row or column.
To extract out a single PRF, you need to address the 117x117 array in a slightly funny way
(117 = 13x9),
.. code-block:: python
img = array[ [colOffset, colOffset+9, colOffset+18, ...],
[rowOffset, rowOffset+9, ...] ]
"""
#TODO Reference documentation about samplesPerPixel
gridSize = self.gridSize
#TODO: Bounds checks like this don't belong in an inner loop. Remove them
#once you're sure they never get triggered.
if colOffset >= gridSize:
raise ValueError("Requested column offset (%i) too large" %
(colOffset))
if rowOffset >= gridSize:
raise ValueError("Requested row offset (%i) too large" %
(colOffset))
assert colOffset < gridSize
assert rowOffset < gridSize
fullImage = singlePrfObj.values / float(singlePrfObj.samplesPerPixel)
#Number of pixels in regularly sampled PRF. Typically 13x13
nColOut, nRowOut = fullImage.shape
nColOut /= float(gridSize)
nRowOut /= float(gridSize)
iCol = colOffset + (np.arange(nColOut) * gridSize).astype(np.int)
iRow = rowOffset + (np.arange(nRowOut) * gridSize).astype(np.int)
if np.max(iCol) == 117:
debug()
#Don't understand why this must be a twoliner
tmp = fullImage[iRow, :]
return tmp[:, iCol]
def get_kubernetes_version_info(self):
"""
fetching the kubernetes version installed on the host
:return: None
"""
try:
res = requests.get(urljoin(base_url, 'version'),
headers=def_header)
debug()
if res.status_code == 200:
print(f" Kubernetes version installed is {res.text}")
else:
print("Kubernetes Version Not available")
except ApiException as e:
print(f"Error : {e}")
except Exception as e:
print(f"Error: {e}")
def optimize_model(memory, BATCH_SIZE, policy_net, target_net, optimizer, GAMMA, device = 'cpu'):
if len(memory) < BATCH_SIZE:
return
transitions = memory.sample(BATCH_SIZE)
# Transpose the batch (see https://stackoverflow.com/a/19343/3343043 for
# detailed explanation). This converts batch-array of Transitions
# to Transition of batch-arrays.
batch = Transition(*zip(*transitions))
# Compute a mask of non-final states and concatenate the batch elements
# (a final state would've been the one after which simulation ended)
non_final_mask = torch.tensor(tuple(map(lambda s: s is not None,
batch.next_state)), device=device, dtype=torch.bool)
non_final_next_states = torch.cat([s for s in batch.next_state
if s is not None])
state_batch = torch.cat(batch.state)
action_batch = torch.cat(batch.action)
reward_batch = torch.cat(batch.reward)
# Compute Q(s_t, a) - the model computes Q(s_t), then we select the
# columns of actions taken. These are the actions which would've been taken
# for each batch state according to policy_net
state_action_values = policy_net(state_batch).gather(1, action_batch)
# Compute V(s_{t+1}) for all next states.
# Expected values of actions for non_final_next_states are computed based
# on the "older" target_net; selecting their best reward with max(1)[0].
# This is merged based on the mask, such that we'll have either the expected
# state value or 0 in case the state was final.
next_state_values = torch.zeros(BATCH_SIZE, device=device)
next_state_values[non_final_mask] = target_net(non_final_next_states).max(1)[0].detach()
# Compute the expected Q values
expected_state_action_values = (next_state_values * GAMMA) + reward_batch
# Compute Huber loss
loss = F.smooth_l1_loss(state_action_values, expected_state_action_values.unsqueeze(1))
debug()
# Optimize the model
optimizer.zero_grad()
loss.backward()
for param in policy_net.parameters():
param.grad.data.clamp_(-1, 1)
optimizer.step()
def divide(self, direction, sugar_transfer, water_transfer, newMemory={}):
"""Spawn a daughter cell in the given direction.
New cell starts with specified sugar and water, and memory initialized to newMemory. Note division has additional costs (specified in cell_types.py as COSTS['GENERIC']).
If the cell doesn't have enough water or sugar for the division, then division will fail but resources will not be lost.
"""
sugar_transfer = max(0, sugar_transfer)
water_transfer = max(0, water_transfer)
if direction not in self.adjacent:
sugar_cost, water_cost = COSTS['GENERIC']
if self.sugar >= sugar_transfer + sugar_cost and \
self.water >= water_transfer + water_cost:
self.sugar -= sugar_transfer + sugar_cost
self.water -= water_transfer + water_cost
return self.world.add_daughter(self, direction, sugar_transfer, water_transfer, newMemory)
else:
print "Warning: Bad division!!!"
debug()
def dna_grass(cell):
"""Entry point for the program. Automates cell behavior."""
if getMem(cell, 'initialized') == 0:
initialize(cell)
role = getMem(cell, 'role')
if role == 0:
print "Error: Cell has no role!"
debug()
elif role == 'stem':
automate_stem(cell)
elif role == 'root':
automate_root(cell)
elif role == 'leaf':
automate_leaf(cell)
elif role == 'store':
automate_store(cell)
elif role == 'origin':
automate_origin(cell)
elif role == 'bud':
automate_bud(cell)
manage_resource_flow(cell)
def add_station(p, vtx_idx, alpha=1.25):
for i in range(p.shape[0]):
edge_prob = p[i][vtx_idx]
if edge_prob < 10e-5:
continue
beta = (1 - edge_prob * alpha) / (1 - edge_prob)
p[i] = p[i] * beta
p[i, vtx_idx] = p[i, vtx_idx] / beta * alpha
if np.max(p) > 1:
print('Error: chosen alpha produces invalid matrix')
raise ValueError
for i in range(p.shape[0]):
try:
assert (np.sum(p[i]) - 1.0) < 10e-4
except:
print('Invalid row', i)
print(alpha, beta)
debug()
return p
def generate_random_routes(p,
num_routes,
route_min_len,
route_max_len,
sources=None,
sinks=None):
num_steps = np.random.uniform(route_min_len, route_max_len, num_routes)
start_vtxs = []
for i in range(num_routes):
start_vtxs.append(sources[np.floor(np.random.uniform(
0, len(sources))).astype(int)])
paths = []
for i in tqdm(range(num_routes), desc='computing paths', ncols=80):
# debug()
if start_vtxs[i] not in sources:
continue
paths.append(random_route(start_vtxs[i], int(num_steps[i]), p))
paths1 = list(
filter(lambda path: path[0] in sources and path[-1] in sinks, paths))
if not paths1:
debug()
return paths
def distribute(cell, amount, resource, pattern):
"""Automates distribution of resource (given amount) to adjacent cells according to pattern
cell: A cell
amount: A float
resource: 'sugar' or 'water'
pattern: 'high' distributes according to high demand profile for given resource, 'low' according to the low demand profile, 'even' gives equally to each adjacent cell
"""
report(cell, 'Distributing {0} {1} in pattern {2}'.format(amount, resource, pattern))
index1 = int(resource == 'water')
if pattern in ('high', 'low'):
index2 = int(pattern == 'low')
demand = get_children_demand(cell)[index1][index2]
if demand == 0:
distribution_factor = 1
else:
distribution_factor = float(amount) / demand
# The proportion of amount demanded to give to each adjacent cell
if resource == 'sugar' and getMem(cell, 'sugar_children') != 0:
for (sugar_child, dir) in getMem(cell, 'sugar_children'):
amt_to_send = getMem(sugar_child, 'demand')[index1][index2] * distribution_factor
xferResource(cell, dir, amt_to_send, resource)
elif resource == 'water' and getMem(cell, 'water_children') != 0:
for (water_child, dir) in getMem(cell, 'water_children'):
amt_to_send = getMem(water_child, 'demand')[index1][index2] * distribution_factor
xferResource(cell, dir, amt_to_send, resource)
else:
if pattern != 'even':
debug()
num_adjacent_cells = 9 - cell.free_spaces
# Used 9 as starting number so that the cell will always keep a portion for itself!
amt_to_send = float(amount) / num_adjacent_cells
for dir, adjcell in cell.adjacent.iteritems():
if adjcell != 'EMPTY':
xferResource(cell, dir, amt_to_send, resource)
def forward(self, x):
crop_h, crop_w = int(x.size()[-2]), int(x.size()[-1])
x = self.stages[0](x)
side = []
side_out = []
for i in range(1, len(self.stages)):
x = self.stages[i](x)
side_temp = self.side_prep[i - 1](x)
side.append(
center_crop(self.upscale[i - 1](side_temp), crop_h, crop_w))
side_out.append(
center_crop(
self.upscale_[i - 1](self.score_dsn[i - 1](side_temp)),
crop_h, crop_w))
out = torch.cat(side[:], dim=1)
out = self.fuse(out)
side_out.append(out)
debug()
print(side_out)
return side_out
def automate_bud(cell):
if getMem(cell.adjacent['S'], 'role') == 'stem':
# Bud to the NW, NE, and revert to stem with budcounter
if isEmpty(cell, 'NW'): dir = 'NW'
elif isEmpty(cell, 'NE'): dir = 'NE'
elif isEmpty(cell, 'N' ): dir = 'N'
else:
dir = 'X'
debug()
if dir in ('NW', 'NE'):
if cell.sugar > 40 and cell.water > 40:
new_mem = {'role': 'bud', 'growth_dir': dir, 'bud_growth_limit': 5}
new_cell = cell.divide(dir, 10, 10, new_mem)
initialize(new_cell)
add_sugar_child(new_cell, cell)
add_water_child(cell, new_cell)
else:
cell.memory['growth_sugar'] = 40
cell.memory['growth_water'] = 40
if dir == 'N':
cell.memory['role'] = 'stem'
cell.memory['growth_to_bud'] = 3
else:
growth_limit = getMem(cell, 'bud_growth_limit')
growth_dir = getMem(cell, 'growth_dir')
leafdir = 'X'
for dir in outsideDirs(growth_dir):
if isEmpty(cell, dir):
leafdir = dir
if leafdir != 'X':
grow_leaf(cell, leafdir)
elif isEmpty(cell, growth_dir) and growth_limit > 0 and can_grow(cell, 40, 40):
new_mem = {'role': 'bud', 'growth_dir': growth_dir, 'bud_growth_limit': growth_limit - 1, 'sugar_children': [cell]}
new_bud = cell.divide(growth_dir, 10, 10, new_mem)
initialize(new_bud)
add_water_child(cell, new_bud)
import cPickle as pk
from plot import plotLines
import numpy as np
import theano
from pylearn2.space import Conv2DSpace, VectorSpace
from pdb import set_trace as debug
mon = np.array(pk.load(open('monitor.pkle')))
axis = [x for x in xrange(len(mon))]
plotLines([
[axis, mon[:, 1]],
[axis, mon[:, 0]],
], title='Training Graph')
debug()
net = pk.load(open('best.pkle'))
debug()
vec_space = VectorSpace(28**2)
# test_X = vec_space.np_format_as(test_X, net.get_input_space())
# train_X = vec_space.np_format_as(train_X, net.get_input_space())
i = 0
X = net.get_input_space().make_theano_batch()
Y = net.fprop(X)
predict = theano.function([X], Y)
debug()
def get_sets(self):
"""
Get a training and a test set for the ESN learning.
These are picked from all the samples we got in 'get_samples'
"""
if not self.testing_set_speakers:
# Chose test and training set at random!
# Refresh initialisation
self.training_set = []
self.test_set = []
# We created representative samples of /a/, /e/ etc. as well as null samples in ambient_speech.py.
# Now, we'll see how many null, or /a/, or /e/ sounds we want.
n_train = self.n_samples['train'] * self.n_speakers
# (remember: self.n_samples[...] was defined as 'per speaker'.
# Same for test samples
n_test = self.n_samples['test'] * self.n_speakers
for vowel in self.vowels_and_null:
# Get representative part of training/test set.
random.shuffle(self.labeled_samples[vowel])
# If many vowels were moved out of a folder, our list of labeled samples might
# be quite short - shorter even than training_samples + test_samples!
for sample in range(n_train):
try:
self.training_set.append(self.labeled_samples[vowel][sample])
except IndexError:
raise RuntimeError('Too few '+vowel+' samples!')
#self.training_set.append(random.choice(self.labeled_samples[vowel]))
for sample in range(n_train,n_train+n_test):
try:
self.test_set.append(self.labeled_samples[vowel][sample])
except IndexError:
raise RuntimeError('Too few '+vowel+' samples!')
#self.test_set.append(random.choice(self.labeled_samples[vowel]))
elif type(self.testing_set_speakers) == list and not self.training_set and not self.test_set:
# This is the case of excluding speakers from the training set and using them for the test set.
# Refresh initialisation
self.training_set = []
self.test_set = []
for vowel in self.vowels_and_null:
for sample in range(len(self.labeled_samples[vowel])):
if self.labeled_samples[vowel+'_speaker_label'][sample] in self.testing_set_speakers:
self.test_set.append(self.labeled_samples[vowel][sample])
else:
self.training_set.append(self.labeled_samples[vowel][sample])
# Change this from a list to a string (this is a flag for the next time round.)
self.testing_set_speakers = "already set"
elif self.testing_set_speakers == "already set":
# in this case, we only need to shuffle them..
pass
else:
# Something has gone wrong.
print self.testing_set_speakers
debug()
print "Length of training set "+str(len(self.training_set))
print "Length of test set "+str(len(self.test_set))
raise RuntimeError("Illegal use of get_sets()")
# Shuffle representative with non-represantive training elements.
random.shuffle(self.training_set)
random.shuffle(self.test_set)
def check_for_rotten_eggs(x):#
"""
Check for 'rotten eggs' in set
"""
i_rotten=0
i = 0
while i in range(len(x)):
if x[i][0] == []:
i_rotten+=1
x.pop(i)
else:
i+=1
print '('+str(i_rotten)+' rotten samples found in set)'
check_for_rotten_eggs(self.training_set)
check_for_rotten_eggs(self.test_set)
def simulate_ESN(self):
""" main function simulating both leaky and non-leaky ESNs
globals:
- self.reservoir_sizes: list of reservoir sizes
- self.n_vowels: number of vowels used
- self.do_plot_hearing: boolean defining if plots are created
- self.separate: boolean defining if infant samples are used as test data
- self.n_channels: number of used channels
"""
# Compare leaky networks with non-leaky networks?
# ------------------------------------------------------------------------
for j in xrange(len(self.reservoir_sizes)): # loop over network sizes
self.current_reservoir = j
for train in xrange(self.trains_per_worker):
# Leaky (done anyway)
# ...........................................................
print('worker', self.rank+1, 'of', self.n_workers, 'simulating leaky network of size', self.reservoir_sizes[self.current_reservoir],
'('+str(train+1)+'/'+str(self.trains_per_worker)+')')
# call learn function
error_leaky, c_matrix_leaky = self.learn(output_dim=self.reservoir_sizes[self.current_reservoir], leaky=True)
if (train==0) and self.do_plot_hearing and (self.rank == 0):
self.plot_prototypes(N=self.reservoir_sizes[self.current_reservoir])
# ...........................................................
if self.do_compare_leaky:
# Non-leaky (optional)
# ...........................................................
print('worker', self.rank+1, 'of', self.n_workers, 'simulating non-leaky network of size', self.reservoir_sizes[self.current_reservoir],
'('+str(train+1)+'/'+str(self.trains_per_worker)+')')
# call learn function to execute one simulation run
error_nonleaky, c_matrix_nonleaky = self.learn(output_dim=self.reservoir_sizes[self.current_reservoir], leaky=False)
if (train==0) and self.do_plot_hearing and (self.rank == 0):
self.plot_prototypes(N=self.reservoir_sizes[self.current_reservoir])
# ...........................................................
# Collect current error rates and append current confusion matrix.
# ----------------------------------------------------------------
self.errors['non-leaky'][train][self.current_reservoir] = error_nonleaky
self.c_matrices['non-leaky'][self.current_reservoir] += c_matrix_nonleaky
self.errors['leaky'][train][self.current_reservoir] = error_leaky
self.c_matrices['leaky'][self.current_reservoir] += c_matrix_leaky
# ----------------------------------------------------------------
if (c_matrix_leaky>1).any():
debug()
# Print current cmatrices?
if self.verbose and self.do_compare_leaky:
print('c_matrix_leaky:', c_matrix_leaky)
print('c_matrix_nonleaky:', c_matrix_nonleaky)
elif self.verbose:
print('c_matrix_leaky:', c_matrix_leaky)
# Divide by number of trains per worker
if self.do_compare_leaky:
self.c_matrices['leaky'] /= self.trains_per_worker
self.c_matrices['non-leaky'] /= self.trains_per_worker
else:
self.c_matrices['leaky'] /= self.trains_per_worker
if (self.c_matrices['leaky']>1).any():
debug()
# Print cmatrices?
if self.do_compare_leaky and self.verbose:
print('total_c_matrices_leaky:', self.c_matrices['leaky'])
print('total_c_matrices_nonleaky:', self.c_matrices['non-leaky'])
elif self.verbose:
print('total_c_matrices_leaky:', self.c_matrices['leaky'])
def main(self):
"""
hear.main() does the auditory learning with the BRIAN hears and the echo state
network (ESN).
Initialize..
Murakami:
# n_training used to be 183
# n_samples used to be 204
"""
# Setting up correct paths and folders
self.setup_folders()
# Make reservoir states inspectable for plotting..
self.make_Oger_inspectable()
# Only partially train the ESN. analysis. See get_params.
if self.generalisation_age:
# Basically run through most of what we've done already, but for different sample data.
# Train only on speakers that are over the age of firstspeaker_male.
# --------------------------------------------------------------------------------------------------------------
# Get the first speaker to include in data samples.
# -------------------------------------------------------------------------
try:
firstspeaker_male = self.age_m.index(self.generalisation_age)
except ValueError:
print "Error! No generalisation age!"
debug()
try:
firstspeaker_female = self.age_f.index(self.generalisation_age)
except ValueError:
print "Error! No generalisation age!"
debug()
firstspeaker = min(firstspeaker_female,firstspeaker_male)
# -------------------------------------------------------------------------
# Execute this 'setup' only by master.
if self.rank == 0:
# A list of speakers that are excluded in training, but used for testing.
self.testing_set_speakers = [speaker for speaker in self.speakers if speaker < firstspeaker]
else:
# The normal case. Train using data from all speakers. Training set and test set are randomly drawn.
self.testing_set_speakers = False
# Get all the samples produced in ambient speech..
self.get_samples()
# Simulate ESN (basically the main function, which then in itself calls the interesting function 'learn')..
self.simulate_ESN()
if self.rank == 0:
self.master_collect_and_postprocess(choose=False) # SEE BELOW. Set choose True if you want to chose one of the produced ESNs for learning (essentially rename it).
# Save this class instance
# ---------------------------------------------------
f = open('data/classes/hear_instance.pickle', 'w+')
f.truncate()
pickle.dump(self,f)
f.close()
def parse_download_url(self, response):
hxs = HtmlXPathSelector(response)
debug()
yum = hxs.select("//img")
def _get_kaggle_passwd(username):
debug()
world.draw_messageboard(screen, MESSAGE_RECT)
world.updateDisplay()
pygame.display.flip()
autoRunning = False
tick_amount = 500
while running:
for event in pygame.event.get():
if event.type == pygame.QUIT:
running = False
elif event.type == pygame.KEYDOWN:
if event.key == pygame.K_SPACE:
world.update_cells()
world.draw_messageboard(screen, MESSAGE_RECT)
pygame.display.flip()
if event.key == pygame.K_x:
debug()
if event.key == pygame.K_q:
running = False
if event.key == pygame.K_g:
if autoRunning:
tick_amount = 500
else:
tick_amount = 6
autoRunning = not(autoRunning)
elif ( event.type == pygame.MOUSEBUTTONDOWN and
pygame.mouse.get_pressed()[0]):
world.change_selected_cell(pygame.mouse.get_pos())
world.draw_messageboard(screen, MESSAGE_RECT)
pygame.display.flip()
if autoRunning:
def evaluate_score(self):
if self.score != None:
return self.score
else:
self.score = 0
if self.new_move[0]=='X':
eval_dict = self.x_dict
other_dict = self.o_dict
inst_dict = self.o_instant
elif self.new_move[0]=='O':
eval_dict = self.o_dict
other_dict = self.x_dict
inst_dict = self.x_instant
# check all horizontal strings
for column in range(len(self.board)-self.win_length):
for row in range(len(self.board)):
s = ''
for i in range(self.win_length+1):
s+=self.board[row][column+i]
self.score+=0.1*eval_dict[s]
self.score-=other_dict[s]
if inst_dict[s]:
self.instant_lose.append(s)
self.score += -10000000000
# return
# check all vertical strings
for row in range(len(self.board)-self.win_length):
for column in range(len(self.board)):
s = ''
for i in range(self.win_length+1):
s+=self.board[row+i][column]
self.score+=0.1*eval_dict[s]
self.score-=other_dict[s]
if inst_dict[s]:
self.instant_lose.append(s)
self.score += -10000000000
# return
# check all diagonal strings len(8-5)
for column in range(self.board_dimension-self.win_length):
for up_row in range(self.board_dimension-self.win_length):
low_row = up_row+5
south_east = ''
north_east = ''
for i in range(self.win_length+1):
south_east+=self.board[up_row+i][column+i]
try:
north_east+=self.board[low_row-i][column+i]
except:
debug()
# if 'XXXX' in north_east:
# debug()
self.score+=0.1*eval_dict[south_east]
self.score-=other_dict[south_east]
self.score+=0.1*eval_dict[north_east]
self.score-=other_dict[north_east]
if inst_dict[south_east]:
self.instant_lose.append(south_east)
self.score += -10000000000
# return
if inst_dict[north_east]:
self.instant_lose.append(north_east)
self.score += -10000000000
# check four remaining
# top right corner
r = 0
c = self.board_dimension-self.win_length
s='.'
for i in range(self.win_length):
s+=self.board[r+i][c+i]
if not self.new_move[0] in s:
self.score+=0.1*eval_dict[s]
self.score-=other_dict[s]
if inst_dict[s]:
self.instant_lose.append(s)
self.score += -10000000000
# return
# top left corner
r = self.win_length-1
c = 0
s='.'
for i in range(self.win_length):
s+=self.board[r-i][c+i]
if not self.new_move[0] in s:
self.score+=0.1*eval_dict[s]
self.score-=other_dict[s]
if inst_dict[s]:
self.instant_lose.append(s)
self.score += -10000000000
# return
# bottom left corner
r = self.board_dimension-self.win_length
c = 0
s='.'
for i in range(self.win_length):
s+=self.board[r+i][c+i]
if not self.new_move[0] in s:
self.score+=0.1*eval_dict[s]
self.score-=other_dict[s]
if inst_dict[s]:
self.instant_lose.append(s)
self.score += -10000000000
# return
# bottom right corner
r = self.board_dimension-1
c = self.board_dimension-self.win_length
s='.'
for i in range(self.win_length):
s+=self.board[r-i][c+i]
if not self.new_move[0] in s:
self.score+=0.1*eval_dict[s]
self.score-=other_dict[s]
if inst_dict[s]:
self.instant_lose.append(s)