def random_pairs(
how_many_pairs,
*,
isomorph_probability=0.5,
non_isomoprh_trivial_probability=0.25,
non_isomoprh_non_trivial_probability=0.25,
**kwargs,
) -> Iterator[Tuple[CNF, CNF, int]]:
"""Iterate through pairs of cnfs generated based on the probabilities."""
# normalize probabilities
tot_probs = isomorph_probability + non_isomoprh_trivial_probability + non_isomoprh_non_trivial_probability
isomorph_probability /= tot_probs
non_isomoprh_trivial_probability /= tot_probs
non_isomoprh_non_trivial_probability /= tot_probs
funcs = [
cnf_isomorphic_generator,
cnf_generator_trivial,
non_trivial_non_isomorphic_cnf_generator,
]
funcs_probs = [
isomorph_probability,
non_isomoprh_trivial_probability,
non_isomoprh_non_trivial_probability,
]
for _ in range(how_many_pairs):
original_cnf = random_cnf(**kwargs)
func_idx = np_choice(3, p=funcs_probs)
func_generator = funcs[func_idx]
new_cnf = func_generator(original_cnf, **kwargs)
yield original_cnf, new_cnf, func_idx
def pick_move(self, pheromone, dist, visited):
pheromone = np.copy(pheromone)
pheromone[list(visited)] = 0
row = pheromone**self.alpha * ((1.0 / dist)**self.beta)
norm_row = row / row.sum()
move = np_choice(self.all_inds, 1, p=norm_row)[0]
return move
def add_node(self, trace, visibility, alpha, beta, q0, sensors):
current_node = self.path[-1]
next_nodes = self.next_nodes(current_node)
if len(next_nodes) == 0:
return False
if random() < q0:
best_node = None
best_x = 0
for next_node in next_nodes:
x = (trace[(sensors[current_node], sensors[next_node])] ** alpha) * \
(visibility[(sensors[current_node], sensors[next_node])] ** beta)
if x > best_x:
best_node = next_node
best_x = x
self.path.append(best_node)
else:
probability_distribution = []
sum_x = 0
for next_node in next_nodes:
sum_x += (trace[(sensors[current_node], sensors[next_node])] ** alpha) * \
(visibility[(sensors[current_node], sensors[next_node])] ** beta)
for next_node in next_nodes:
probability_distribution.append(
(trace[(sensors[current_node], sensors[next_node])]**alpha)
* (visibility[(sensors[current_node], sensors[next_node])]
**beta) / sum_x)
draw = np_choice(next_nodes, 1, probability_distribution)
self.path.append(draw[0])
return True
def gen_paths(self):
length = len(self.distances_table)
L = []
for k in range(0, length):
L.append(k)
ret=[]
start=0
L.pop(0)
ret=L
paths=[]
paths1=[]
leng=len(ret)
for k in range(0,self.ants_number):
path=[]
path.append(start)
while ret!=[]:
ind=np_choice(np.array(ret),1)
indice=ret.index(ind)
path.append(ret[indice])
ret.pop(indice)
path.append(start)
paths.append(path)
for j in range (1,len(self.distances_table)):
ret.append(j)
for k in paths:
path1=[]
l=len(k)
for j in range (0,l-1):
kha=(k[j],k[j+1])
path1.append(kha)
paths1.append(path1)
print('the paths of the '+str(self.ants_number) +' that exist :')
print(paths1)
return paths1
def pick_next_path(self, prev, visited_edges):
"""
:param self:
:param prev:
:param visited_edges:
:return: path given by np_choice
Get all the neighbouring_unvisited_paths and take the most attractive
"""
paths = get_neighbouring_unvisited_paths(prev, visited_edges)
if len(paths) == 0:
return None
if len(paths) == 1:
return paths[0]
proba_incoming = np.array([
path.pheromone**(self.alpha) * ((1.0 / path.length)**self.beta)
for path in paths
])
s = proba_incoming.sum()
proba = proba_incoming / s
# for i in range(len(proba)):
#
for i in range(len(paths)):
print(f'{i}: p = {proba[i]} {paths[i]}')
choice = np_choice(len(paths), 1, p=proba)[0]
res_path = paths[choice]
print(f"Result path: ({choice}) {res_path}")
return res_path
def sample(self):
"""Samples a random batch of experiences."""
all_weights = [e.weight for e in self.memory if e is not None]
s = sum(all_weights)
idxs = np_choice(len(self.memory),
self.batch_size,
p=[w / s for w in all_weights],
replace=False)
experiences = [self.memory[i] for i in idxs]
# let's convert that into tensors, sent into the appropriate device.
states = torch.from_numpy(
np.vstack([e.state for e in experiences
if e is not None])).float().to(self.device)
actions = torch.from_numpy(
np.vstack([e.action for e in experiences
if e is not None])).float().to(self.device)
rewards = torch.from_numpy(
np.vstack([e.reward for e in experiences
if e is not None])).float().to(self.device)
dones = torch.from_numpy(
np.vstack([e.done for e in experiences
if e is not None])).float().to(self.device)
next_states = torch.from_numpy(
np.vstack([e.next_state for e in experiences
if e is not None])).float().to(self.device)
return (states, actions, rewards, next_states, dones)
def travel(self):
"""Makes ant travel to next vertex.
Generates allowed moves and their probabilities. Makes choice.
"""
# Generate valid vertices.
self.generate_allowed_moves()
probabilities = np.array(list(
map(self.get_probability, self.allowed_moves)),
dtype='float64')
self.validate_probabilities(probabilities)
next_vertex = np_choice(self.allowed_moves, p=probabilities)
'''Add next edge value to total cost. If previous value was bigger,
then subtract the previous value, and add it 10 times bigger.'''
if self.previous_vertex is not None:
if self.aco.graph.matrix[self.previous_vertex, self.current_vertex] > \
self.aco.graph.matrix[self.current_vertex, next_vertex]:
self.total_cost -= self.aco.graph.matrix[self.previous_vertex,
self.current_vertex]
self.total_cost += self.aco.graph.matrix[
self.previous_vertex, self.current_vertex] * 10
self.total_cost += self.aco.graph.matrix[self.current_vertex,
next_vertex]
# On next move ant can't go to previous and current.
if self.previous_vertex is not None:
self.tabu_moves.append(self.previous_vertex)
self.tabu_moves.append(self.current_vertex)
# Set a new current vertex, update previous.
self.visited_vertices.append(next_vertex)
self.previous_vertex = self.current_vertex
self.current_vertex = next_vertex
def pick_move(self, pheromone, dist, visited):
pheromone = np.copy(pheromone)
pheromone[list(
visited)] = 0 # посещенные города не должны посещаться снова
row = pheromone**self.alpha * ((1.0 / dist)**self.beta)
norm_row = row / row.sum()
move = np_choice(self.all_inds, 1, p=norm_row)[0]
return move
def _chose_path(self):
dirs = self._labirynth.dir_list(self._pos)
neigbour_fields = self._labirynth.neigbour_fields(self._pos)
neigbour_fields_probability = [((self._labirynth._labirynth[pos.y()][pos.x(
)]+1))/(self._labirynth.find_max_pheromon()+1) for pos in neigbour_fields]
neigbour_fields_probability_with_adjusted_p = [
prob/sum(neigbour_fields_probability) for prob in neigbour_fields_probability]
return np_choice(dirs, p=neigbour_fields_probability_with_adjusted_p)
def next(self, left):
if left is None:
return r_choice(tuple(self._fragments.keys()))
probabilities = self.probabilities(left)
rights = tuple(probabilities.keys())
rights_probabilities = tuple(map(lambda r: probabilities[r], rights))
return np_choice(rights, p=rights_probabilities)
def pickWeightedRand(self, sampleArray, weightsArray, pick_n, np=False):
picks = []
if (np == True):
picks = np_choice(sampleArray, pick_n, weightsArray)
else:
picks = r_choices(sampleArray, weightsArray, k=pick_n)
return picks
def pickRand(self, pick_n, np=False):
sampleArray = self.getArrayNums();
picks = [];
if (np==True):
picks = np_choice(sampleArray, pick_n)
else: picks = r.choices(sampleArray, k=pick_n)
picks.sort();
return picks;
def pick_move(self, pheromone, dist, visited):
# here we use this line to avoid any change in the reference array
pheromone = np.copy(pheromone)
#to make a probabilty being zero for those nodes which is already visited
pheromone[list(visited)] = 0
row = pheromone**self.alpha * ((1.0 / dist)**self.beta)
norm_row = row / row.sum()
# here we pick a random number from the nodes ,according to it's probapilities and using direct access as the choice function return list not element
move = np_choice(self.all_inds, 1, p=norm_row)[0]
return move
def get_weather(self):
occurrence = np_choice(['rare', 'uncommon', 'common'],
1,
p=[0.1, 0.3, 0.6])[0]
terrain = choice(self.terrain)
weather_options = [
weather for weather in terrain_tags[terrain]['weather']
if weather['occurrence'] == occurrence
]
return choice(weather_options)
def genRobSol(self):
while True:
cmpltBool, updateRobLst = self._mpdaUpdater.updateState()
# print(cmpltBool)
if cmpltBool:
break
# print(updateRobLst)
nextUpdateRobTaskPairLst = []
random.shuffle(updateRobLst)
self.updateRobDic = dict()
for robID in updateRobLst:
self.updateRobDic[robID] = np.inf
for robID in updateRobLst:
rob = self.robotLst[robID]
pheromoneLst = np.copy(
self.taskPheromoneLst[robID][rob.taskID])
pheromoneLst[self.encode[robID]] = 0
row = np.zeros([self._taskNum])
for taskID, pheromone in enumerate(pheromoneLst):
# print(pheromone)
if pheromone == 0:
continue
else:
eta = self.calHeuristic(robID, taskID, dummy=False)
row[taskID] = pheromone**self.alpha * (eta**self.beta)
row_sum = row.sum()
if row_sum == 0:
rob.stopBool = True
break
norm_row = row / row_sum
next_taskID = np_choice(range(self._taskNum), 1, p=norm_row)[0]
nextUpdateRobTaskPairLst.append(
RobTaskPair(robID=robID, taskID=next_taskID))
self.updateRobDic[robID] = next_taskID
# print(nextUpdateRobTaskPairLst)
if len(nextUpdateRobTaskPairLst) == 0:
continue
if False not in self._mpdaUpdater.cmpltLst:
break
encode, fitness, nextUpdateRobTaskPairLst = self.fixSol(
nextUpdateRobTaskPairLst, dummy=False)
if fitness != 0:
return encode, fitness, []
self._mpdaUpdater.updateEncode(nextUpdateRobTaskPairLst)
# print(self.encode)
fitness = self.calFitness()
encode = self.encode
arrCmpltTaskLst = self._mpdaUpdater._arrCmpltTaskLst
if fitness == sys.float_info.max:
pass
return encode, fitness, arrCmpltTaskLst
def pickWeightedRand(self, pick_n, score_type="overall", np=False):
sampleArray = self.getArrayNums();
weightsArray = self.getArrayScores(score_type)
picks = [];
if (np==True):
picks = np_choice(sampleArray, pick_n, weightsArray)
else: picks = r.choices(sampleArray, weightsArray, k=pick_n, replace=False)
picks.sort();
return picks;
def pick_move(self, pheromone, dist, visited):
pheromone = np.copy(pheromone)
pheromone[list(visited)] = 0
# adjust pheromone level acc to weight
row = pheromone ** self.alpha * ((1.0 / dist) ** self.beta)
norm_row = row / row.sum()
move = np_choice(self.all_inds, 1, p=norm_row)[
0] # pick 1 random index for move
return move
def pick_move(self, pheromone_exclude: np.ndarray,
pheromone_include: np.ndarray, i: int) -> int:
excluded = pheromone_exclude * (self.times_taken[0][i] /
(self.n_ants * self.current_iteration))
included = pheromone_include * (self.times_taken[1][i] /
(self.n_ants * self.current_iteration))
rows = np.array([excluded, included]) / (excluded + included)
move = np_choice([0, 1], 1, p=rows)[0]
return move
def _get_optimized_learning_step(self, partial_semantics):
"""Calculates optimized learning step."""
""" bootstrap samples; compute OLS for each; use desired criterion to select the final LS """
if self.bootstrap_ols:
weights = []
size = self.target_vector.shape[0]
for sample in range(self.bootstrap_ols_samples):
idx = np_choice(arange(size), size, replace=True)
bootstrap_delta_target = copy(
self.target_vector[idx]).astype(float)
if self.champion:
full_predictions = self.champion.neural_network.get_predictions(
)
bootstrap_delta_target -= full_predictions[idx]
bootstrap_partial_semantics = partial_semantics[idx]
inverse = array(
pinv(
resize(bootstrap_partial_semantics,
(1, bootstrap_partial_semantics.size))))
ols = dot(inverse.transpose(), bootstrap_delta_target)[0]
weights += [ols]
ols_median = median(weights)
ols_mean = mean(weights)
ols = self._compute_ols(partial_semantics)
abs_dif = abs(ols_median - ols_mean)
if abs_dif >= self.high_absolute_ls_difference:
self.high_absolute_differences_history.append(
[abs_dif, ols_median, ols_mean, ols])
#===============================================================
# print('Absolute difference: %.3f, median vs. mean: %.3f vs. %.3f' % (abs_dif, ols_median, ols_mean))
#===============================================================
#===============================================================
# print('Absolute difference: %.3f, median vs. mean vs. OLS: %.3f vs. %.3f vs. %.3f' % (abs_dif, ols_median, ols_mean, ols))
# print()
#===============================================================
if self.bootstrap_ols_criterion == 'median':
return median(weights)
else:
return mean(weights)
else:
return self._compute_ols(partial_semantics)
def pick_move(self, pheromone, dist, visited):
pheromone = np.copy(pheromone)
# Sets pheromone of visited nodes to zero so the ants do not go backwards
pheromone[list(visited)] = 0
# Careful with alpha and beta values. Large exponents may cause overflow or underflow (i.e. e+16 == inf and e-16 == 0)
row = pheromone**self.alpha * ((1.0 / dist)**self.beta)
# Ensures that we return to start_node after reaching an End of path
if (row.sum() == 0):
move = self.start_node
return move
norm_row = row / row.sum()
# This returns 0 if all probabilities are zero (i.e. no more places to go)
move = np_choice(self.all_inds, 1, p=norm_row)[0]
return move
def sample_multicomb(seq, l):
"""
Sample uniformly from the set of all multisets of elements of seq with
size l. Multisets are represented as sorted tuples.
"""
# sample a pattern according to the number of possible instantiations
pats = compute_patterns(len(seq), l)
ps = [p for p, n in pats]
ns = [n for p, n in pats]
divisor = sum(ns)
ns = list(map(lambda x: x / divisor, ns))
draw = np_choice(range(len(ps)), 1, p=ns)[0]
res_pat = ps[draw]
# res_pat = random.choices(ps, weights=ns, k=1)[0]
res = instantiate_pattern(seq, res_pat)
return tuple(sorted(res))
def pick_next(self, prev_pheromone, prev_row, visited):
"""
:param prev_pheromone (1D array): Previous row of the matrix of pheromone.
:param prev_row (1D array): Previous row of the matrix of distances.
:param visited (set): Set of visited nodes.
:return (int): Next node by using probability.
"""
row_pheromone = np.copy(prev_pheromone)
row_pheromone[list(visited)] = 0
row = row_pheromone**self.alpha * ((1.0 / prev_row)**self.beta)
norm_row = row / row.sum()
next = np_choice(range(self.n), 1, p=norm_row)[0]
return next
def pick_move(pheromone, dist, visited):
pheromone = np.copy(pheromone)
#Make zero if the path has been visited
pheromone[list(visited)] = 0
#Ant makes a decision on what city to go using this formula
row = pheromone**alpha * ((1.0 / dist)**beta)
#Probability formula
norm_row = row / row.sum()
#Move randomly using probability (select path to go using probability)
#p=probability
#Get index of an element that has bigger probability
move = np_choice(all_inds, 1, p=norm_row)[0]
#print(move)
#Return path that randomly selected
return move
def genRobFirstActTrad(self):
'''
is used to generate first event.
:return the first event:
'''
robInitVisitLst = []
robSeq = [x for x in range(self._robNum)]
random.shuffle(robSeq)
robInitVisitLst = [np.inf for x in range(self._robNum)]
self.robInitVisitLst = [np.inf for x in range(self._robNum)]
for robID in robSeq:
row = np.zeros([self._taskNum])
for taskID, pheromone in enumerate(
self.robTaskPheromoneLst[robID]):
eta = self.calHeuristic(robID, taskID, dummy=True)
row[taskID] = pheromone**self.alpha * (eta**self.beta)
# roadDur = self._rob2taskDisMat[robID][taskID] / self._robVelLst[robID]
row_sum = row.sum()
norm_row = row / row_sum
firstAct = np_choice(range(self._taskNum), 1, p=norm_row)[0]
robInitVisitLst[robID] = firstAct
self.robInitVisitLst[robID] = firstAct
'''
此处需要和后续的生成方法一致
'''
# print('robInitVisitLst = ',robInitVisitLst)
nextUpdateRobTaskPairLst = []
for robID, taskID in enumerate(robInitVisitLst):
nextUpdateRobTaskPairLst.append(RobTaskPair(robID, taskID))
encode, fitness, nextUpdateRobTaskPairLst = self.fixSol(
nextUpdateRobTaskPairLst, dummy=True)
robInitVisitLst = [x.taskID for x in nextUpdateRobTaskPairLst]
# print('robInitVisitLst = ',robInitVisitLst)
return robInitVisitLst
def updatingLimited(self, nextUpdateRobTaskPairLst):
updateRobLst = []
for robID, taskID in nextUpdateRobTaskPairLst:
updateRobLst.append(robID)
curAccRobAbiLst = [0 for x in self._taskRateLst]
for robID in range(self._robNum):
if robID in updateRobLst:
nextVisitTaskID = nextUpdateRobTaskPairLst[updateRobLst.index(
robID)].taskID
curAccRobAbiLst[nextVisitTaskID] = curAccRobAbiLst[
nextVisitTaskID] + self._robAbiLst[robID]
else:
nextVisitTaskID = self._mpdaUpdater.encode[robID][-1]
curAccRobAbiLst[nextVisitTaskID] = curAccRobAbiLst[
nextVisitTaskID] + self._robAbiLst[robID]
curTaskRateLst = []
for taskID in range(self._taskNum):
if curAccRobAbiLst[taskID] == 0:
continue
if self._mpdaUpdater.cmpltLst[taskID]:
continue
curTaskRateLst.append(
(taskID, self._taskRateLst[taskID] - curAccRobAbiLst[taskID]))
# print('length curTaskRateLst = ',len(curTaskRateLst))
if len(curTaskRateLst) <= self.limitedNum:
return nextUpdateRobTaskPairLst
minTaskID, minCurTaskRate = min(curTaskRateLst, key=lambda x: x[1])
limitedSet = set()
for robID in range(self._robNum):
if robID in updateRobLst:
pass
else:
taskID = self.encode[robID][-1]
if self._mpdaUpdater.cmpltLst[taskID] == False:
limitedSet.add(taskID)
newVisitedTaskLst = []
for robID, taskID in nextUpdateRobTaskPairLst:
if taskID not in limitedSet:
newVisitedTaskLst.append(taskID)
# print(limitedSet)
# print(newVisitedTaskLst)
'''
此处可以修正 如何将 newVisitedTaskLst 的元素加入 limitedSet 中
'''
random.shuffle(newVisitedTaskLst)
# print(newVisitedTaskLst)
while len(limitedSet) < self.limitedNum:
limitedSet.add(newVisitedTaskLst[0])
newVisitedTaskLst.remove(newVisitedTaskLst[0])
# print(limitedSet)
nextUpdateRobTaskPairLst = []
for robID in updateRobLst:
rob = self.robotLst[robID]
pheromoneLst = np.copy(self.taskPheromoneLst[robID][rob.taskID])
pheromoneLst[self.encode[robID]] = 0
row = np.zeros([self._taskNum])
for taskID, pheromone in enumerate(pheromoneLst):
# print(pheromone)
if taskID not in limitedSet:
continue
if pheromone == 0:
continue
else:
eta = self.calHeuristic(robID, taskID, dummy=False)
roadDur = self._taskDisMat[
rob.taskID][taskID] / self._robVelLst[robID]
predictArrTime = self.robotLst[robID].leaveTime + roadDur
if predictArrTime > self.taskLst[taskID].cmpltTime:
continue
row[taskID] = pheromone**self.alpha * (eta**self.beta)
row_sum = row.sum()
if row_sum == 0:
rob.stopBool = True
break
norm_row = row / row_sum
# print(norm_row)
next_taskID = np_choice(range(self._taskNum), 1, p=norm_row)[0]
nextUpdateRobTaskPairLst.append(
RobTaskPair(robID=robID, taskID=next_taskID))
return nextUpdateRobTaskPairLst
def play(self, board, v=0):
# print "PLAYING"
chosen = None
if v:
print "playing..."
if board.pos['turn'] == "player_1":
playable = [1, 2, 3, 4, 5, 6]
else:
playable = [8, 9, 10, 11, 12, 13]
playable = filter(lambda bin: board.pos[bin], playable)
if len(playable) == 1:
return playable[0]
elif playable:
odds = []
for bin in playable:
b_board = Board()
b_board.pos = dict(board.pos)
b_board.play(bin)
pos = [b_board.pos[i] for i in range(14)]
if board.pos['turn'] == "player_1":
p = [0, 1]
else: # board.pos['turn'] == "player_2"
p = [1, 0]
# odds of winning - odds of losing
pos = array(pos).reshape(1, -1)
# print self.model.predict_proba(pos)
my_odds = self.model.predict_proba(pos)[0][p[0]]
opponents_odds = self.model.predict_proba(pos)[0][p[1]]
if opponents_odds > 0.0:
win_odds = my_odds / opponents_odds
elif (my_odds - opponents_odds) <= 0.0:
win_odds = 0.0
else:
win_odds = my_odds - opponents_odds
odds.append(win_odds)
# normalize odds to be from 0 to 1
min_, max_ = min(odds), max(odds)
odds = map(lambda x: scale(x, (min_, max_), (
0.00,
1.0,
)), odds)
# so their sum == 1
raw_sum = sum(odds)
weights = [(float(i) / float(raw_sum)) for i in odds]
# weighted choice
if not self.monte_carlo:
chosen = np_choice(playable, p=weights)
# TODO: add monte carlo choice
# Monte Carlo playouts
if self.monte_carlo:
# print "monte_carlo", self.v
end = time() + self.time_per_move
monte_counts = defaultdict(list)
mc_count = 0
while time() < end:
mc_count += 1
# make a monte_dojo with save=False and p1=LearnerBot(model=self.model, monte_carlo=False) and
# p2=LearnerBot(model=self.model, monte_carlo=False)
p1 = LearnerBot(model=self.model, monte_carlo=False)
p2 = LearnerBot(model=self.model, monte_carlo=False)
monte_dojo = Dojo(p1=p1,
p2=p2,
csv_name="raw_games.csv",
save=False)
# replace monte_dojo.board.pos with dict(board.pos)
monte_dojo.board.pos = dict(board.pos)
# choose a move and play it out
# print weights
chosen = np_choice(playable, p=weights)
monte_dojo.board.play(chosen)
# have dojo finish the game
monte_dojo.play_game()
# determine winner with monte_dojo.board.pos[0] and monte_dojo.board.pos[7]
if monte_dojo.board.pos[0] < monte_dojo.board.pos[7]:
winner = 'player_1'
elif monte_dojo.board.pos[0] > monte_dojo.board.pos[7]:
winner = 'player_2'
else:
winner = "tie"
# add winner to monte_weights[chosen]
monte_counts[chosen].append(winner)
# player = board.pos['turn'] # which is 'player_1' or 'player_2' or None
monte_weights = dict(monte_counts)
player = board.pos['turn']
opponent = list({'player_1', 'player_2'} - {player})[0]
for bin, winner_list in monte_weights.iteritems():
opponent_wins = float(winner_list.count(opponent))
my_wins = float(winner_list.count(player))
if opponent_wins:
monte_weights[bin] = my_wins / opponent_wins
else:
monte_weights[bin] = None
max_odds = max(monte_weights.values())
for bin, odds in monte_weights.iteritems():
if odds == None:
monte_weights[bin] = max_odds
if set(monte_weights.values()) == {None}:
for bin, odds in monte_weights.iteritems():
if odds == None:
monte_weights[bin] = 1.0
# convert monte_weights from dict(list) to list(float) of wins-losses
possible_bins = monte_weights.keys()
weight_list = monte_weights.values()
# normalize odds to be from 0 to 1
min_, max_ = min(weight_list), max(weight_list)
if min_ == max_:
min_ = max_ - 1
odds = map(lambda x: scale(x, (min_, max_), (
0.0,
1.0,
)), weight_list)
# so their sum == 1
raw_sum = sum(odds)
weights = [(float(i) / float(raw_sum)) for i in odds]
# chosen = np_choice(playable, p=weights)
chosen = np_choice(possible_bins, p=weights)
if self.v:
print mc_count, 'game playouts'
return chosen
else:
return None
if v:
print chosen
return chosen
def get_word(word_list):
"""
Choosing a random word from a given list
"""
choice = np_choice(word_list)
return choice, len(choice)
def do_run (run, trials):
#present instructions
#if fMRI, wait for scanner trigger
instruct1.draw()
win.flip()
logging.data('** START RUN %s **' % specific_run)
event.waitKeys(keyList=('space'))
instruct2.draw()
win.flip()
event.waitKeys(keyList=('space'))
keyPress.draw()
win.flip()
event.waitKeys(keyList=('space'))
instruct3.draw()
win.flip()
event.waitKeys(keyList=('space'))
if run == 1:
ITI = ITI_1
sampling_durList = sampling_dur1
if run == 2:
ITI = ITI_2
sampling_durList = sampling_dur2
for trial in trials:
#add trial logic: show stimuli, get resp, add data to 'trial'
idx = trials.thisIndex
sampling_dur = sampling_durList[idx]
ITItime=ITI[idx]
core.checkPygletDuringWait = False
### RANDOMIZATION
#also it should be noted that in the fmriProblemSet_20180802 file I was dumb and the OUTCOME and PROBABILITIES have different naming conventions SMH. They are re-streamlined here.
#as of 20190114 I have renamed the PROBABILITIES naming conventions so that they match with the OUTCOME.
prob1_1 = float(trial['P1_1']) #probability1 for option1
prob2_1 = float(trial['P2_1']) #probability2 for option1
prob1_2 = float(trial['P1_2']) #probability1 for option2
prob2_2 = float(trial['P2_2']) #probability2 for option2
if '-' not in trial['O1_1'] and trial['O1_1'] != '0': #O1_1
O1_1 = '+%s' % (trial['O1_1'])
else:
O1_1 = trial['O1_1']
if '-' not in trial['O1_2'] and trial['O1_2'] != '0': #O1_2
O1_2 = '+%s' % (trial['O1_2'])
else:
O1_2 = trial['O1_2']
if '-' not in trial['O2_1'] and trial['O2_1'] != '0': #O2_1
O2_1 = '+%s' % (trial['O2_1'])
else:
O2_1 = trial['O2_1']
if '-' not in trial['O2_2'] and trial['O2_2'] != '0': #O2_2
O2_2 = '+%s' % (trial['O2_2'])
else:
O2_2 = trial['O2_2']
Prob1 = {
'p': [prob1_1, prob2_1],
'out': [O1_1, O2_1]
}
Prob2 = {
'p': [prob1_2, prob2_2],
'out': [O1_2, O2_2]
}
sampling_RT_list = []
sampling_resp_list = []
sampling_lr_list= []
sampling_outcome_list = []
sampling_respOnset_list = []
samplingCount = 0
safeCount = 0
riskyCount = 0
switchCount = 0
leftRight = random.randint(0,1) #if 0, P1 is on the left, P2 is on the right
trialStartTime = globalClock.getTime()
trials.addData('trialStartTime', trialStartTime)
logging.data('Trial Onset - ProblemType: %s, ProblemNumber: %s, leftRight: %s, samplingDur: %s, assignedITI: %s' % (trial['ProbType'], trial['ProbNumber'], leftRight, sampling_dur, ITItime))
timer.reset()
#https://github.com/alishir/IGT_net/blob/master/igt_psychtoolbox/igt_mri.m
logging.data('Sampling Onset')
while timer.getTime() < sampling_dur:
samplingOnset = timer.getTime()
#SET PROBABILITIES FOR EACH SUBTRIAL
draw1 = np_choice(a=Prob1['out'], size=1, p=Prob1['p'])
draw2 = np_choice(a=Prob2['out'], size=1, p=Prob2['p'])
box1.draw()
box2.draw()
press1.draw()
press2.draw()
win.flip()
if DEBUG:
resp_val = random.randint(1,2)
rt_onset = globalClock.getTime()
logging.data('DEBUG MODE Sampled Response: %s' % resp_val)
core.wait(.5)
else:
resp = event.getKeys(keyList = responseKeys, timeStamped=globalClock)
resp_val = None
prev_resp = None
#prev_rt = None
if len(resp) > 0:
resp_val = int(resp[0][0])
rt_onset = resp[0][1]
sampling_respOnset_list.append(rt_onset)
if samplingCount == 0: #and idx == 0:
sampling_RT_list.append(rt_onset-trialStartTime)
else:
sampling_RT_list.append(rt_onset-prev_rt)
if samplingCount != 0: #won't have a previous response for the first sample
prev_resp = sampling_resp_list[-1]
prev_rt = sampling_respOnset_list[-1]
#for adding in SwitchCount later
#save prev_response
#if resp_val != prev_resp:
#switchCount += 1
#make new response replace previous response
if resp_val == 1:
samplingCount += 1
#logic to decide which outcome to display
if leftRight == 0: #if 0, P1 is on the left, P2 is on the right
out1.setText(draw1[0])
response = 1
else: #if 1, P1 is on the right, P2 is on the left
out1.setText(draw2[0])
response = 2
if response == int(trial['SafeOption']):
safeCount += 1
else:
riskyCount += 1
if prev_resp and response != prev_resp: #won't have a previous response for the first sample
switchCount += 1
box1.draw()
box2.draw()
press1.draw()
press2.draw()
out1.draw()
win.flip()
core.wait(feedback_dur)
logging.data('Sampled Response: %s, Outcome Shown: %s' % (response, draw1[0]))
sampling_resp_list.append(response)
sampling_lr_list.append(leftRight)
sampling_outcome_list.append(draw1[0])
if resp_val == 2:
samplingCount += 1
if leftRight == 0:
out2.setText(draw2[0])
response = 2
else:
out2.setText(draw1[0])
response = 1
if response == int(trial['SafeOption']):
safeCount += 1
else:
riskyCount += 1
if prev_resp and response != prev_resp: #won't have a previous response for the first sample
switchCount += 1
box1.draw()
box2.draw()
press1.draw()
press2.draw()
out2.draw()
win.flip()
core.wait(feedback_dur)
logging.data('Sampled Response: %s, Outcome Shown: %s' % (response, draw2[0]))
sampling_resp_list.append(response)
sampling_lr_list.append(leftRight)
sampling_outcome_list.append(draw2[0])
event.clearEvents()
if timer.getTime() >= sampling_dur-.5 :
logging.data('** broke sampling loop (<.5s left in trial) **')
break
for idx in range(samplingCount):
sampling_file.writerow([
trials.thisIndex+1,
sampling_resp_list[idx],
sampling_lr_list[idx],
sampling_outcome_list[idx],
sampling_RT_list[idx],
sampling_respOnset_list[idx]])
samplingEndTime = globalClock.getTime()
sampling_pad = sampling_dur-timer.getTime()
trials.addData('samplingCount_total', samplingCount) #jk #I starting samplingCount at 1 to use it as an index for correctly calculating response time. We subtract that here to log it accurately.
trials.addData('samplingCount_risky', riskyCount)
trials.addData('samplingCount_safe', safeCount)
trials.addData('switchCount', switchCount)
trials.addData('samplingDur_assigned', sampling_dur)
trials.addData('samplingDur_total', samplingEndTime-trialStartTime)
trials.addData('samplingEndTime', samplingEndTime)
timer.reset()
box1.setLineColor('white')
box2.setLineColor('white')
thinkOnset = globalClock.getTime()
logging.data('samplingCount: %s' % samplingCount)
logging.data('Think Onset')
trials.addData('thinkOnset', thinkOnset)
while timer.getTime() < think_dur:
box1.draw()
box2.draw()
think.draw()
win.flip()
box1.draw()
box2.draw()
press1.draw()
press2.draw()
decide.draw()
win.flip()
timer.reset()
event.clearEvents()
answer = 0
response = []
decide_onset = globalClock.getTime()
logging.data('Decide Onset')
trials.addData('decideOnset', decide_onset)
timer.reset()
while timer.getTime() < decision_dur:
if DEBUG:
resp = [1]
resp_val = random.randint(1,2)
logging.data('DEBUG MODE Decision Response: %s' % resp_val)
core.wait(.5)
else:
resp = event.getKeys(keyList = responseKeys)
if len(resp) > 0:
resp_onset = globalClock.getTime()
rt = resp_onset-decide_onset
answer=1
if not DEBUG:
resp_val = int(resp[0])
logging.data('Decision Button Press: %s' % resp)
if resp_val == 1:
box1.setLineColor('red')
if leftRight == 0:
response = 1 #response is 1 for P1/option1 and 2 for P2/option2 always
else:
response = 2
if resp_val == 2:
box2.setLineColor('red')
if leftRight == 0:
response = 2 #response is 1 for P1/option1 and 2 for P2/option2 always
else:
response = 1
if response == int(trial['SafeOption']):
safeChoice = 0 #safeChoice is 0 if the safe choice was chosen, 1 if the risky option was chosen
else:
safeChoice = 1
logging.data('Decision Response: %s' % response)
box1.draw()
box2.draw()
press1.draw()
press2.draw()
win.flip()
core.wait(feedback_dur)
decide_pad = decision_dur-rt-.3
decideDur_total = resp_onset-trialStartTime+.3
break
if answer == 0:
response = 'NA'
resp_val = 'NA'
resp_onset = 'NA'
rt = 'NA'
decide_pad = 'NA'
safeChoice = 'NA'
decideDur_total = decision_dur
trials.addData('leftRight', leftRight)
trials.addData('resp', response)
trials.addData('safeChoice', safeChoice) #0 if safe, 1 if risky
trials.addData('resp_onset', resp_onset)
trials.addData('rt', rt)
trials.addData('decideDur_total', decideDur_total)
#reset box colors, reset outcomes
box1.setLineColor('white')
box2.setLineColor('white')
out1.setText()
out2.setText()
timer.reset()
#ITI
ITI_onset=globalClock.getTime()
logging.data('ITI Onset')
if answer == 0:
totalITI = ITItime+sampling_pad
else:
totalITI = ITItime+decide_pad+sampling_pad
while timer.getTime() < totalITI:
fixation.draw()
win.flip()
trials.addData('assignedITI', ITItime)
trials.addData('sampling_pad', sampling_pad)
trials.addData('decide_pad', decide_pad)
trials.addData('totalITI', totalITI)
trials.addData('ITI_onset', ITI_onset)
trialEndTime = globalClock.getTime()
trials.addData('TrialEndTime', trialEndTime)
logging.data('Trial End Time - totalITI: %s' % totalITI)
timer.reset()
event.clearEvents()
trials.saveAsWideText(fileName=log_file.format(subj_id, subj_id, run), delim=',', appendFile=True)
logging.data('*****END RUN %s*****' % specific_run)