def get_people(terminal: Terminal):
people = []
known_folders = os.listdir(OUTPUT_CSVS_FRIENDS_DIR)
try:
progress_bar = ProgressBar(
'Loading knowns from .csv files', max=len(known_folders))
for folder in known_folders:
knowns_file = '%s/%s/known_people.csv' % (
OUTPUT_CSVS_FRIENDS_DIR, folder)
if os.path.exists(knowns_file):
with open(knowns_file, 'r') as f:
reader = csv.DictReader(f)
for individual in reader:
people.append(individual)
f.close()
progress_bar.next()
print()
with open('%s/known_people.csv' % OUTPUT_CSVS_DIR, 'r') as f:
reader = csv.DictReader(f)
progress_bar_knowns = ProgressBar(
'Loading knowns from .csv files', max=sum(1 for row in reader) - 1)
for individual in reader:
people.append(individual)
progress_bar_knowns.next()
f.close()
except FileNotFoundError as ex:
terminal.error(str(ex))
return uniquify(people)
def saveAverageImage(kitti_base, pos_labels, shape, fname, avg_num=None):
num_images = float(len(pos_labels))
avg_num = min(avg_num, num_images)
if avg_num is None:
avg_num = num_images
# avg_img = np.zeros((shape[0],shape[1],3), np.float32)
avg_img = np.zeros(shape, np.float32)
progressbar = ProgressBar('Averaging ' + fname, max=len(pos_labels))
num = 0
for label in pos_labels:
if num >= avg_num:
break
num += 1
progressbar.next()
sample = getCroppedSampleFromLabel(kitti_base, label)
# sample = np.float32(sample)
resized = resizeSample(sample, shape, label)
resized = auto_canny(resized)
resized = np.float32(resized)
avg_img = cv2.add(avg_img, resized / float(avg_num))
progressbar.finish()
cv2.imwrite(fname, avg_img)
def compete(strategies, num_rounds=1000):
"""Create and run No Thanks competition."""
# Play 3, 4, and 5 player games
game_sizes = [3, 4, 5]
# A dictionary of scores
results = {}
for num_players in game_sizes:
results[num_players] = {}
for strategy in strategies:
results[num_players][strategy] = 0
for num_players in game_sizes:
for _ in ProgressBar(
'Playing {}-player games'.format(num_players)).iter(
range(num_rounds)):
selected_strategies = choices(strategies, k=num_players)
players = [import_module(s).Player() for s in selected_strategies]
winners, _ = nothanks.Game(players).run()
for strategy, player in zip(selected_strategies, players):
results[num_players][strategy] -= 1 / num_players / num_rounds
if id(player) in winners:
results[num_players][strategy] += 1 / len(
winners) / num_rounds
results = pd.DataFrame(results)
results['combined'] = results.sum(axis=1)
results.loc['total', :] = results.sum(axis=0)
return results
def average_image(pos_region_generator, shape, avg_num=None):
pos_regions = list(pos_region_generator)
num_images = float(len(pos_regions))
if avg_num is None:
avg_num = num_images
else:
avg_num = min(avg_num, num_images)
window_dims = (shape[1], shape[0])
# avg_img = np.zeros((shape[0],shape[1],3), np.float32)
avg_img = np.zeros(shape, np.float32)
progressbar = ProgressBar('Averaging ', max=avg_num)
num = 0
for reg in pos_regions:
if num >= avg_num:
break
num += 1
progressbar.next()
resized = reg.load_cropped_resized_sample(window_dims)
resized = auto_canny(resized)
resized = np.float32(resized)
avg_img = cv2.add(avg_img, resized / float(avg_num))
progressbar.finish()
return avg_img
def save_regions(reg_gen, num_regions, window_dims, save_dir):
progressbar = ProgressBar('Saving regions', max=num_regions)
index = 0
for img_region in itertools.islice(reg_gen, 0, num_regions):
fname = os.path.join(save_dir, '{:06d}.png'.format(index))
index += 1
sample = img_region.load_cropped_resized_sample(window_dims)
cv2.imwrite(fname, sample)
progressbar.next()
progressbar.finish()
def train(self, points, labels, Niter=1e6, eta=0.05):
# weights, biases, number of layers, number of classes
W, b, L, C = self.W, self.b, len(self.b), len(self.b[-1])
# declare arrays
a, δ, D = [None] * L, [None] * L, [None] * L
# value of cost function at each iteration
costvals = []
# let the random sequence (indices) of points selected for stochastic gradient descent be
seq = np.random.randint(low=0, high=len(points), size=int(Niter))
# loop over the seq with a progress bar
for itr in ProgressBar("Training").iter(range(len(seq))):
# the random training point
k = seq[itr]
x, y = points[k], int(labels[k])
# 1-of-K representation for label
yy = [int(y == c) for c in range(C)]
# --- forward pass ---
a[0] = x
# For l = 1,...,L-1
for l in range(1, L):
# a[l] = σ( z[l] ) = σ( W[l] a[l-1] + b[l] )
a[l] = sigmoid(W[l] @ a[l - 1] + b[l])
# D[l] = diag( σ'(z[l]) ) = diag( a[l]*(1-a[l]) )
D[l] = np.diag(a[l] * (1 - a[l]))
# --- backward pass ---
# δ[L] = D[L]( a[L] − y(x{k}) )
δ[L - 1] = D[L - 1] @ (a[L - 1] - yy)
# For l = L-2,...,1
for l in reversed(range(1, L - 1)):
# δ[l] = D[l] * ( W[l+1]^T * δ[l+1] )
δ[l] = D[l] @ (W[l + 1].T @ δ[l + 1])
# --- gradient step ---
for l in range(1, L):
# W[l] <= W[l] - η δ[l] a[l-1]^T
self.W[l] = W[l] - eta * np.outer(δ[l], a[l - 1])
# b[l] <= b[l] - η δ[l]
self.b[l] = b[l] - eta * δ[l]
# --- cost function (calc every kth iteration) ---
if itr % costskip == 0:
currcost = self.total_cost(points, labels)
costvals.append([itr, currcost])
# return cost values
return np.array(costvals)
def __run_list(self, net, dataloader, output_dir):
pbar = ProgressBar('Processing', max=len(dataloader), width=48)
for _, batch_data in enumerate(dataloader):
imgs_input, codenames, sizes = batch_data
imgs_recon = self.__infer_step(imgs_input, net)
for idx in range(0, imgs_recon.shape[0]):
orig_img_size = sizes[idx].numpy()
recon_img = cv2.cvtColor(imgs_recon[idx], cv2.COLOR_RGB2BGR)
recon_img = utils.cropping_center(recon_img, orig_img_size)
cv2.imwrite('%s/%s.jpg' % (output_dir, codenames[idx]), recon_img)
pbar.next()
pbar.finish()
return
def fetch_photos(terminal):
terminal.write('-' * 44)
knowns = []
try:
knowns = facebook_api.get_knowns()
except IOError as ex:
token_not_generated(terminal, ex)
progress_bar = ProgressBar('Saving knowns photos', max=len(knowns))
for known in knowns:
picture = facebook_api.get_profile_picture(known['id'])
known['picture'] = picture
progress_bar.next()
save_list_of_dicts('%s/known_people.csv' %
OUTPUT_CSVS_DIR, knowns, ['id', 'name', 'picture'])
terminal.info('%s knowns found' % len(knowns))
def find_label_clusters(kitti_base,
kittiLabels,
shape,
num_clusters,
descriptors=None):
if descriptors is None:
progressbar = ProgressBar('Computing descriptors',
max=len(kittiLabels))
descriptors = []
for label in kittiLabels:
progressbar.next()
img = getCroppedSampleFromLabel(kitti_base, label)
# img = cv2.resize(img, (shape[1], shape[0]), interpolation=cv2.INTER_AREA)
img = resizeSample(img, shape, label)
hist = get_hog(img)
descriptors.append(hist)
progressbar.finish()
else:
print 'find_label_clusters,', 'Using supplied descriptors.'
print len(kittiLabels), len(descriptors)
assert (len(kittiLabels) == len(descriptors))
# X = np.random.randint(25,50,(25,2))
# Y = np.random.randint(60,85,(25,2))
# Z = np.vstack((X,Y))
# convert to np.float32
Z = np.float32(descriptors)
# define criteria and apply kmeans()
K = num_clusters
print 'find_label_clusters,', 'kmeans:', K
criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 10, 1.0)
attempts = 10
ret, label, center = cv2.kmeans(Z, K, None, criteria, attempts,
cv2.KMEANS_RANDOM_CENTERS)
# ret,label,center=cv2.kmeans(Z,2,criteria,attempts,cv2.KMEANS_PP_CENTERS)
print 'ret:', ret
# print 'label:', label
# print 'center:', center
# # Now separate the data, Note the flatten()
# A = Z[label.ravel()==0]
# B = Z[label.ravel()==1]
clusters = partition(kittiLabels, label)
return clusters
def fetch_emails(terminal):
terminal.write('-' * 44)
emails = []
try:
knowns = facebook_api.get_knowns()
except IOError as ex:
token_not_generated(terminal, ex)
progress_bar = ProgressBar('Fetching emails', max=len(knowns))
for known in knowns:
known_data = facebook_api.get_profile_data(known['id'])
known['email'] = known_data.get('email', None)
emails.append(known['email'])
progress_bar.next()
save_list_of_dicts('%s/known_emails.csv' %
OUTPUT_CSVS_DIR, knowns, ['id', 'name', 'email'])
terminal.info('%s/%s emails found' % (len(knowns), len(emails)))
def fetch_all(terminal):
terminal.write('-' * 44)
try:
knowns = facebook_api.get_knowns()
except IOError as ex:
token_not_generated(terminal, ex)
progress_bar = ProgressBar('Fetching basic data', max=len(knowns))
for known in knowns:
known_data = facebook_api.get_profile_data(known['id'])
picture = facebook_api.get_profile_picture(known['id'])
known['picture'] = picture.name
known['phone'] = known_data.get('mobile_phone', None)
known['email'] = known_data.get('email', None)
known['username'] = known_data.get('username', None)
progress_bar.next()
file_name = '%s/known_all.csv' % OUTPUT_CSVS_DIR
fields = ['id', 'username', 'name', 'email', 'phone', 'picture']
save_list_of_dicts(file_name, knowns, fields)
terminal.info('%s knowns found' % len(knowns))
def compute_hog_descriptors(hog, image_regions, window_dims, label):
# hog = get_hog_object(window_dims)
progressbar = ProgressBar('Computing descriptors', max=len(image_regions))
reg_descriptors = []
for reg in image_regions:
progressbar.next()
img = reg.load_cropped_resized_sample(window_dims)
grey = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
winStride = (8, 8)
padding = (0, 0)
locations = [] # (10, 10)# ((10,20),)
hist = hog.compute(grey, winStride, padding, locations)
reg_desc = utils.RegionDescriptor(reg, hist, label)
reg_descriptors.append(reg_desc)
progressbar.finish()
return reg_descriptors
def start(self, download_dir: str):
"""Starts the download process.
Arguments:
download_dir {str} -- Download directory
"""
self.logger.info(
f"Attempting to download {self.comic.title}, {len(self.chapters)} chapter(s).")
download_dir = path.normpath(download_dir)
comic_dir = path.join(
download_dir, clean_filename(self.comic.title))
self.comic_dir = comic_dir
self.logger.debug(f"Trying to create folders: {comic_dir}")
create_folders(comic_dir)
total_pages = 0
for chapter in self.chapters:
total_pages += len(chapter.get_pages())
with ProgressBar(f"Downloading {self.comic.title}", max=total_pages) as bar:
start_time = time.time()
if self.number_of_threads > 1:
self.logger.debug("Starting multithreaded download...")
self._multi_threaded_download(download_dir, bar=bar)
else:
self.logger.debug("Starting singlethreaded download...")
self._single_threaded_download(download_dir, bar=bar)
end_time = time.time()
print()
self.logger.info(
f"Download complete! Time elapsed: {end_time - start_time}")
self._convert()
self.logger.info(f"Operation done!")
def save_list_of_dicts(file_name: str, l: list, fields: list):
exists = os.path.exists(file_name)
mode = 'w+'
if exists:
mode = 'a'
with open(file_name, mode) as f:
writer = csv.DictWriter(f, fieldnames=fields)
if not exists:
writer.writeheader()
progress_bar = ProgressBar('Saving %s' %
relative_path(file_name), max=len(l))
for el in l:
row = {}
for field in fields:
try:
row[field] = el[field]
except KeyError as ex:
raise ex
if row.keys():
writer.writerow(row)
progress_bar.next()
f.close()
print()
return f
def _run_epoch(self,
model,
dataloader,
optimize=False,
save_activations=False,
reweight=None,
bit_pretrained=False,
adv_metrics=False):
"""Runs the model on a given dataloader.
Note:
The latter item in the returned tuple is what is necessary to run
GEORGECluster.train and GEORGECluster.evaluate.
Args:
model(nn.Module): A PyTorch model.
dataloader(DataLoader): The dataloader. The dataset within must
subclass GEORGEDataset.
optimize(bool, optional): If True, the model is trained on self.criterion.
save_activations(bool, optional): If True, saves the activations in
`outputs`. Default is False.
bit_pretrained(bool, optional): If True, assumes bit_pretrained and does not evaluate
performance metrics
Returns:
metrics(Dict[str, Any]) A dictionary object that stores the metrics defined
in self.config['metric_types'].
outputs(Dict[str, Any]) A dictionary object that stores artifacts necessary
for model analysis, including labels, activations, and predictions.
"""
dataset = dataloader.dataset
self._check_dataset(dataset)
type_to_num_classes = {
label_type: dataset.get_num_classes(label_type)
for label_type in LABEL_TYPES
if label_type in dataset.Y_dict.keys()
}
outputs = {
'metrics': None,
'activations': [],
'superclass': [],
'subclass': [],
'true_subclass': [],
'alt_subclass': [],
'targets': [],
'probs': [],
'preds': [],
'losses': [],
'reweight': [],
}
activations_handle = self._init_activations_hook(
model, outputs['activations'])
if optimize:
progress_prefix = 'Training'
model.train()
else:
progress_prefix = 'Evaluation'
model.eval()
per_class_meters = self._init_per_class_meters(type_to_num_classes)
metric_meters = {
k: AverageMeter()
for k in ['loss', 'acc', 'loss_rw', 'acc_rw']
}
progress = self.config['show_progress']
if progress:
bar = ProgressBar(progress_prefix, max=len(dataloader), width=50)
for batch_idx, (inputs, targets) in enumerate(dataloader):
batch_size = len(inputs)
if self.use_cuda:
inputs, targets = move_to_device([inputs, targets],
device=self.device)
type_to_labels = {}
for label_type in type_to_num_classes.keys():
type_to_labels[label_type] = targets[label_type]
outputs[label_type].append(targets[label_type])
if optimize and not bit_pretrained:
logits = model(inputs)
loss_targets = targets['superclass']
co = self.criterion(logits, loss_targets, targets['subclass'])
loss, (losses, corrects), _ = co
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
else:
with torch.no_grad():
logits = model(inputs)
loss_targets = targets['superclass']
if bit_pretrained:
if progress:
bar.suffix = PROGRESS_BAR_SUFFIX.format(
batch=batch_idx + 1,
size=len(dataloader),
total=format_timedelta(bar.elapsed_td),
eta=format_timedelta(bar.eta_td),
**{k: 0
for k in prog_metric_names})
bar.next()
continue
co = self.criterion(logits, loss_targets,
targets['subclass'])
loss, (losses, corrects), _ = co
if not save_activations:
outputs['activations'].pop() # delete activations
reweight_vec = None if reweight is None else reweight[
targets['true_subclass']]
metrics = self._compute_progress_metrics(losses,
corrects,
type_to_labels,
type_to_num_classes,
per_class_meters,
reweight=reweight_vec)
acc, preds = compute_accuracy(logits.data,
loss_targets.data,
return_preds=True)
outputs['probs'].append(
F.softmax(logits, dim=1).detach().cpu()[:, 1])
outputs['preds'].append(preds)
outputs['losses'].append(losses.detach().cpu())
outputs['targets'].append(loss_targets.detach().cpu())
if reweight_vec is not None:
outputs['reweight'].append(reweight_vec.cpu())
self._update_metrics(metric_meters, acc, loss, losses, corrects,
batch_size, reweight_vec)
PROGRESS_BAR_STR = PROGRESS_BAR_SUFFIX
if self.compute_auroc:
sub_map = dataloader.dataset.get_class_map('subclass')
assert (set(sub_map.keys()) == {0,
1}) # must be a binary problem
targets_cat, probs_cat = torch.cat(
outputs['targets']), torch.cat(outputs['probs'])
auroc = compute_roc_auc(targets_cat, probs_cat)
metrics['auroc'] = auroc
has_alt_subclass = 'alt_subclass' in dataloader.dataset.Y_dict
for key in ['subclass', 'true_subclass'
] + ['alt_subclass'] * has_alt_subclass:
sub_map = dataloader.dataset.get_class_map(key)
neg_subclasses = sub_map[0]
pos_subclasses = sub_map[1]
if len(neg_subclasses) == len(pos_subclasses) == 1:
# only one subclass in each superclass
rob_auroc = auroc
else:
subclass_labels = torch.cat(outputs[key])
paired_aurocs = []
for neg_subclass in neg_subclasses:
for pos_subclass in pos_subclasses:
inds = ((subclass_labels == neg_subclass) |
(subclass_labels
== pos_subclass)).cpu()
subset_pair_auroc = compute_roc_auc(
targets_cat[inds], probs_cat[inds])
paired_aurocs.append(subset_pair_auroc)
rob_auroc = min(paired_aurocs)
metrics[f'{key}_rob_auroc'] = rob_auroc
if not has_alt_subclass:
metrics[alt_subclass_rob_auroc] = auroc
PROGRESS_BAR_STR += ' | AUROC: {auroc:.4f} | R AUROC: {subclass_rob_auroc:.4f} | ' \
'TR AUROC: {true_subclass_rob_auroc:.4f} | AR AUROC: {alt_subclass_rob_auroc:.4f}'
if progress:
bar.suffix = PROGRESS_BAR_STR.format(
batch=batch_idx + 1,
size=len(dataloader),
total=format_timedelta(bar.elapsed_td),
eta=format_timedelta(bar.eta_td),
**{
**metrics,
**{k: v.avg
for k, v in metric_meters.items()}
})
bar.next()
if progress:
bar.finish()
if activations_handle:
activations_handle.remove()
for k, v in outputs.items():
if type(v) == list and len(v) > 0:
outputs[k] = concatenate_iterable(v)
if bit_pretrained:
return outputs['metrics'], outputs
outputs['metrics'] = metrics
outputs['metrics'].update(
{k: float(v.avg)
for k, v in metric_meters.items()})
outputs['metrics'].update(self._compute_aggregate_metrics(outputs))
self._print_output_metrics(outputs)
if adv_metrics:
scaa = np.mean([
ga.avg * 100
for ga in np.array(per_class_meters[f'per_true_subclass_accs'])
])
self.logger.info(
f'All accs: {[ga.avg * 100 for ga in np.array(per_class_meters[f"per_true_subclass_accs"])]}'
)
self.logger.info(f'SCAA: {scaa:.3f}')
ap = sklearn.metrics.average_precision_score(
outputs['targets'],
outputs['probs'],
sample_weight=outputs['reweight']
if reweight_vec is not None else None)
self.logger.info(f'MaP: {ap:.4f}')
return outputs['metrics'], outputs
def process(self,
cpu: interface.Processor,
selector: Tuple[str, str] = None,
source: tsdb.Database = None,
fieldmapper: FieldMapper = None,
gzip: bool = False,
buffer_size: int = 1000) -> None:
"""
Process each item in a [incr tsdb()] test suite.
The output rows will be flushed to disk when the number of new
rows in a table is *buffer_size*.
Args:
cpu (:class:`~delphin.interface.Processor`): processor
interface (e.g., :class:`~delphin.ace.ACEParser`)
selector: a pair of (table_name, column_name) that specify
the table and column used for processor input (e.g.,
`('item', 'i-input')`)
source (:class:`~delphin.tsdb.Database`): test suite from
which inputs are taken; if `None`, use the current
test suite
fieldmapper (:class:`FieldMapper`): object for
mapping response fields to [incr tsdb()] fields; if
`None`, use a default mapper for the standard schema
gzip: if `True`, compress non-empty tables with gzip
buffer_size (int): number of output rows to hold in memory
before flushing to disk; ignored if the test suite is all
in-memory; if `None`, do not flush to disk
Examples:
>>> ts.process(ace_parser)
>>> ts.process(ace_generator, 'result:mrs', source=ts2)
"""
if selector is None:
assert isinstance(cpu.task, str)
input_table, input_column = _default_task_selectors[cpu.task]
else:
input_table, input_column = selector
if (input_table not in self.schema
or all(f.name != input_column
for f in self.schema[input_table])):
raise ITSDBError('invalid table or column: {!s}, {!s}'
.format(input_table, input_column))
if source is None:
source = self
if fieldmapper is None:
fieldmapper = FieldMapper(source=source)
index = tsdb.make_field_index(source.schema[input_table])
affected = set(fieldmapper.affected_tables).intersection(self.schema)
for name in affected:
self[name].clear()
key_names = [f.name for f in source.schema[input_table] if f.is_key]
bar = None
if not logger.isEnabledFor(logging.INFO):
with tsdb.open(source.path, input_table) as fh:
total = sum(1 for _ in fh)
if total > 0:
bar = ProgressBar('Processing', max=total)
for row in source[input_table]:
datum = row[index[input_column]]
keys = [row[index[name]] for name in key_names]
keys_dict = dict(zip(key_names, keys))
response = cpu.process_item(datum, keys=keys_dict)
logger.info(
'Processed item {:>16} {:>8} results'
.format(tsdb.join(keys), len(response['results']))
)
if bar:
bar.next()
for tablename, data in fieldmapper.map(response):
_add_row(self, tablename, data, buffer_size)
for tablename, data in fieldmapper.cleanup():
_add_row(self, tablename, data, buffer_size)
if bar:
bar.finish()
tsdb.write_database(self, self.path, gzip=gzip)
def detect_objects_in_image(self,
img,
greyscale=False,
resize=True,
return_detection_img=True,
progress=True):
h, w = img.shape[:2]
scaled_img_dims = (w, h)
if resize:
max_w = 1024
# max_w = 200
if img.shape[0] > max_w:
# print 'resize:', img_path, img.shape
# img = cv2.resize(img, dsize=None, fx=0.5, fy=0.5)
h, w = img.shape[:2]
aspect = w / float(h)
new_h = int(max_w / aspect)
img = cv2.resize(img, dsize=(max_w, new_h))
scaled_img_dims = (max_w, new_h)
print 'img.shape:', img.shape
def get_win_gen(only_rects=False):
return detectutils.sliding_window_generator(
img,
window_dims=self.window_dims,
scale_factor=1.1,
strides=(8, 8),
# scale_factor=1.2,
# strides=(16, 16),
only_rects=only_rects)
win_gen = get_win_gen()
num_windows = sum(1 for _ in get_win_gen(only_rects=True))
batch_size = 100
# num_batches = int(np.ceil(num_windows/float(batch_size)))
detected_cars = []
progressbar = None
if progress:
progressbar = ProgressBar('Processing windows:',
max=num_windows,
suffix='%(index)d/%(max)d - %(eta)ds')
while True:
samples_windows = list(itertools.islice(win_gen, 0, batch_size))
if len(samples_windows) == 0:
break
# Unzip the list of tuples into two lists:
samples, windows = zip(*samples_windows)
# Perform the required colour conversion and preprocessing:
samples = [self.prepare_sample(sample) for sample in samples]
# Convert shape from [num examples, rows, columns, depth]
# to [num examples, rows*columns*depth]
samples = np.stack(samples)
samples = samples.reshape(
samples.shape[0],
samples.shape[1] * samples.shape[2] * samples.shape[3])
feed = {self.x: samples, self.keep_prob: 1.0}
label_probs = self.sess.run(tf.nn.softmax(self.logits),
feed_dict=feed)
if progress:
progressbar.next(batch_size)
for probs, window in zip(label_probs, windows):
pos_prob, neg_prob = probs
if pos_prob > neg_prob:
detected_cars.append(window.opencv_bbox)
if progress:
progressbar.finish()
if len(detected_cars) > 0:
detected_cars = np.stack(detected_cars)
else:
detected_cars = np.array([])
if return_detection_img:
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
draw_detections(img, detected_cars)
return detected_cars.tolist(), img
else:
return detected_cars.tolist(), scaled_img_dims
def train(self, num_games, print_progress=True):
# Create a set of identical players, each referencing this game tree
players = [
Player(self,
num_players=self.num_players,
starting_coins=self.starting_coins,
low_card=self.low_card,
high_card=self.high_card,
discard=self.discard) for _ in range(self.num_players)
]
# Simulate self-play
if print_progress:
iterable = ProgressBar('Training').iter(range(num_games))
else:
iterable = range(num_games)
for _ in iterable:
winners, _ = nothanks.Game(players,
starting_coins=self.starting_coins,
low_card=self.low_card,
high_card=self.high_card,
discard=self.discard).run()
# Update the shared game tree based on game results.
for player in players:
payoff = -1 / self.num_players
if id(player) in winners:
payoff += 1 / len(winners)
for state_hash, action in player.history.items():
node = self.tree.nodes[state_hash]
state = node['state']
# If the player has no coins, there is no decision to make.
if not node['can_pass']:
continue
# If the expected return from the alternate decision is
# higher than the return from this game, the player regrets
# not having chosen the alternate decision.
take_state = deepcopy(state)
take_state.take()
take_edge = self.tree.edges[(state_hash,
take_state.prehash())]
pass_state = deepcopy(state)
pass_state.pass_turn()
pass_edge = self.tree.edges[(state_hash,
pass_state.prehash())]
if action: # took card and pot when in this state
alternate = pass_state
alt_payoff = self.get_expected_payoff_pass(state)
else: # passed when in this state
alternate = take_state
alt_payoff = self.get_expected_payoff_take(state)
regret = alt_payoff - payoff
# If we have never seen this state before, assign defaults.
if node['visits'] == 0:
logger.debug(
'LOG: State {} was visited for the first time this game.'
.format(state_hash))
# Then update node and edge values
node['visits'] += 1
# Add new regret for the action we DIDN'T take
node['regret'][not action] += regret
# Update strategy based on new cumulative regret
effective_regret = [max(0, r) for r in node['regret']]
total_regret = sum(effective_regret)
if total_regret == 0:
take_edge['weight'] = 1 / 2
pass_edge['weight'] = 1 / 2
else:
take_edge[
'weight'] = effective_regret[True] / total_regret
pass_edge[
'weight'] = effective_regret[False] / total_regret
# Adjust running average strategy
take_edge['avg_weight'] *= (node['visits'] -
1) / node['visits']
take_edge[
'avg_weight'] += take_edge['weight'] / node['visits']
pass_edge['avg_weight'] *= (node['visits'] -
1) / node['visits']
pass_edge[
'avg_weight'] += pass_edge['weight'] / node['visits']