def merge_results_d(result_a, result_b, repository):
if result_a.label_map != result_b.label_map:
print('error, label mappings dont match, intervention required')
Tracer()()
class_to_idx = result_a.label_map
samples_1 = repository.get_samples_by_uid(result_a.ids)
samples_2 = repository.get_samples_by_uid(result_b.ids)
templates_1 = repository.get_templates_by_uid(result_a.template_order,
samples_1)
templates_2 = repository.get_templates_by_uid(result_b.template_order,
samples_2)
Tracer()()
X_a, y_a, ids_a = extract_features(samples_1, templates_2, class_to_idx)
X_b, y_b, ids_b = extract_features(samples_2, templates_1, class_to_idx)
# take special notice to the flipped use of feature vectors due to
# template matching results concatentation
for i, x in enumerate(result_a.X):
X_a[i].extend(x)
for i, x in enumerate(X_b):
result_b.X[i].extend(x)
data = FeatureData()
data.X = X_a + result_b.X
data.y = result_a.y + result_b.y
data.ids = result_a.ids + result_b.ids
data.label_map = class_to_idx
data.template_order = result_a.template_order + result_b.template_order
def run(batch_size,nb_classes,nb_epoch,data_augmentation,img_rows, img_cols,img_channels,weightfilepath,logfilepath,bestweightfilepath,Load_model ):
#function to run the actual test
# the data, shuffled and split between train and test sets
X_train,y_train,X_test,y_test,X_val,y_val = load_data()
X_train,Y_train,X_test,Y_test,X_val,Y_val = normalize_date(X_train,y_train,X_test,y_test,X_val,y_val,nb_classes)
print('Loading and formatiing of data complete...')
#load the model defined in model_architecture function
model = model_architecture(img_rows,img_cols,img_channels,nb_classes)
# training the model using SGD + momentum
adm = Adam(lr=0.00001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
model.compile(loss='categorical_crossentropy',
optimizer=adm,
metrics=[get_categorical_accuracy_keras])
if Load_model:
Tracer()()
model.load_weights(weightfilepath )
save_model_per_epoch = ModelCheckpoint1(weightfilepath,logfilepath,bestweightfilepath, monitor='val_loss', verbose=1, save_best_only=False)
model.fit(X_train,Y_train,
batch_size=batch_size,
nb_epoch=nb_epoch,
validation_data=(X_val, Y_val),
shuffle=True,
verbose=2,
callbacks=[save_model_per_epoch])
def coco_modify_segs(self):
for count, im_idx in enumerate(self.D.image_index):
if count % 100 == 0:
print 'Image {:d} of {:d}'.format(count,
len(self.D.image_index))
seg_filename = self.D.seg_path_from_index(im_idx)
im = cv2.imread(im_filename)
if im is None:
print 'Could not read ', im_filename
sys.exit(-1)
seg = self.get_seg(im)
# get annotations
c = self.D._COCO
ann_ids = c.getAnnIds(imgIds=im_idx, iscrowd=False)
anns = c.loadAnns(ann_ids)
mask = np.zeros(seg.shape, dtype=np.int)
for ann in anns:
if 'segmentation' not in ann:
continue
cat_id = ann['category_id'] + 9
if type(ann['segmentation']) == list:
for s in ann['segmentation']:
poly = np.array(s).reshape((len(s) / 2, 2))
poly = Polygon(poly)
pth = path.Path(poly.get_xy(), closed=True)
y, x = np.mgrid[:seg.shape[0], :seg.shape[1]]
points = np.transpose((x.ravel(), y.ravel()))
m = pth.contains_points(points).reshape(mask.shape)
mask[m > 0] = cat_id * m[m > 0]
else:
Tracer()()
# superimpose mask on seg
seg[mask > 0] = mask[mask > 0]
cv2.imwrite(seg_filename, seg)
def store_media_data(self,broadcast_data,entry_name,broadcast_key):
self.download_media_data.update(broadcast_data)
create_broadcast_view(self.download_media_data,self.left_box)
broadcast_data[broadcast_key]['episodes'][entry_name] = self.download_dict[entry_name]
Tracer()()
create_episode_view(self.download_media_data["episodes"][entry_name],self.middle_box)
def map_FirstHaloInFOFgroup(self,tree):
first_halo = np.empty(len(tree),np.int32)
first_halo.fill(-1)
host_inds = (np.where(tree['pid'] == -1))[0]
if len(host_inds) > 0:
## Host halos points to themselves
first_halo[host_inds] = host_inds
## Now set the subhalos
hostids = tree['id'][host_inds]
for host_loc,hostid in zip(host_inds,hostids):
ind_subs = (np.where(tree['pid'] == hostid))[0]
if len(ind_subs) > 0:
assert np.max(first_halo[ind_subs]) == -1,'First halo inds should not already have been assigned'
first_halo[ind_subs] = host_loc
if min(first_halo) < 0:
Tracer()()
return first_halo
print("tree = {}".format(tree))
print("returning none. expect code to break")
return None
def breakpoint():
try:
from IPython.core.debugger import Tracer
Tracer()()
except:
import pdb
pdb.set_trace()
def main():
file_names = get_data_file_names()
game_logs_created = 0
start_connection()
for fname in file_names:
with open('pickles/{}'.format(fname), 'rb') as f:
season, season_type, team = get_info_from_filename(fname)
lines = map(lambda x: x.split(),
filter(lambda x: len(x) > 10, f.readlines()))
for gl in lines[1:]:
gl[0] = gl[0][
2:] # removes 'aV' from first element of each row
try:
create_gamelog(gl, season, season_type, team)
game_logs_created += 1
except:
Tracer()()
pass
print fname + ': ' + str(len(lines[1:]))
print str(game_logs_created) + ' game logs created.'
def TracerFactory():
"""
Returns a callable `Tracer` object.
When this object is called it starts the ipython commandline debugger
in that place.
"""
return Tracer(colors='Linux')
def canonical_cumsum(array, n, axis, crop=True):
from IPython.core.debugger import Tracer
debug_here = Tracer()
''' Input:
> array : np-array or list
> n : windowsize (n>0)
> axis : axis along which the cumsum will be taken
> crop : bool | will crop the beginning of the output array,
where the sum runs over uncomplete window.
note: if ndim of array is >1 then the output will always be cropped.
--> did not find a function like np.take to /access/ an array.
'''
ndims = len(array.shape)
# pandas doesn't know negative axes-allocation
if axis < 0:
axis = ndims + axis
if n <= 0:
sys.exis('The windowsize n must be chosen greater than 0!')
if ndims == 1:
b = utils_ana.nancumsum(array)
b[n:] = b[n:] - b[:-n]
if crop == True:
b = b[n - 1:]
return (b)
else:
lenax = array.shape[axis]
b = utils_ana.nancumsum(array, axis=axis)
b_first = np.expand_dims(np.take(b, n - 1, axis=axis), axis=axis)
c = np.take(b, np.arange(n, lenax), axis=axis) - np.take(
b, np.arange(lenax - n), axis=axis)
d = np.concatenate((b_first, c), axis=axis)
return (d)
def train_step(self, data, **kwargs):
# torch.autograd.set_detect_anomaly(True)
loss, data_dict, r = self.model(
data['X'],
actions=data['action'].cuda().float(),
rewards=data['reward'].cuda().float().squeeze(-1),
mask=data['done'].cuda().float().squeeze(-1),
pretrain=kwargs['pretrain'])
self.optimizer.zero_grad()
torch.mean(-1. * loss).backward()
if self.clip_gradient:
torch.nn.utils.clip_grad_norm_(self.model.parameters(),
self.clip_gradient_value)
for n, p in self.model.named_parameters():
if p.grad is not None:
# print(f'{n}: {p.grad.max()} {p.grad.min()} {p.grad.mean()}')
if not torch.isfinite(p.grad.mean()):
from IPython.core.debugger import Tracer
Tracer()()
self.optimizer.step()
# if self.num_steps % 150 == 0 and kwargs['visdom']:
# if kwargs['pretrain']:
# self.visdom_logger.log_visdom(data['X'].flatten(end_dim=1), data_dict['masks'], data_dict['reconstruction'], 6)
# else:
# self.visdom_logger.log_visdom(data['X'][0], data_dict['masks'][0], data_dict['reconstruction'][0], 6)
self.num_steps += 1
return data_dict
def parse_tweet(self, tweet, response):
tweet_item = items.TwitterscraperItem()
tweet_item['session_id'] = self.session_id
tweet_item['tweet_id'] = tweet['tweet_id']
tweet_item['text'] = tweet['text']
tweet_item['user_id'] = tweet['user_id']
tweet_item['user_screen_name'] = tweet['user_screen_name']
tweet_item['user_name'] = tweet['user_name']
tweet_item['created_at_ts'] = tweet['created_at_ts']
tweet_item['created_at_iso'] = tweet['created_at_iso']
# tweet_item['convo_url'] = tweet['convo_url']
# tweet_item['image_url'] = tweet['image_url']
tweet_item['num_retweets'] = tweet['num_retweets']
tweet_item['num_favorites'] = tweet['num_favorites']
tweet_item['keyword'] = tweet['keyword']
tweet_item['query'] = self.query
# referring_url = response.request.headers.get('Referer', None)
tweet_item['referring_url'] = response.request.headers.get(
'Referer', None) or self.start_urls[0]
tweet_item['request_url'] = response.url
tweet_item['quote_tweet_id'] = tweet['quote_tweet_userid']
tweet_item['quote_tweet_userid'] = tweet['quote_tweet_userid']
tweet_item['quote_tweet_username'] = tweet['quote_tweet_username']
tweet_item['quote_tweet_screenname'] = tweet['quote_tweet_screenname']
tweet_item['quote_tweet_text'] = tweet['quote_tweet_text']
try:
tweet_item['html'] = tweet['html']
except Exception, e:
Tracer()()
pass
def button_press_callback(self, event):
"""Left button: add point; middle button: delete point;
right button: fill polygon and stop interaction.
"""
from IPython.core.debugger import Tracer
Tracer()()
if event.inaxes:
x, y = event.xdata, event.ydata
ax = event.inaxes
if event.button == 1: # If you press the left button
if self.circ == None: # if there is no line, create a line
self.circ = plt.Circle((x, y),
0.5,
facecolor='none',
edgecolor='r')
ax.add_artist(self.circ)
# add a segment
else: # if there is a line, create a segment
self.circ.set_color('b')
self.circ.set_alpha(0.3)
self.disconnect()
elif event.button == 3 and self.circ != None: # middle button: remove last segment
# ax.artists.remove(self.circ)
self.circ.remove()
self.circ = None
self.fig.canvas.draw()
def iou_calc(self, boxes1, boxes2): #x, y, w, h
boxes1 = torch.stack([boxes1[:, :, :, 0] - boxes1[:, :, :, 2] / 2.0,
boxes1[:, :, :, 1] - boxes1[:, :, :, 3] / 2.0,
boxes1[:, :, :, 0] + boxes1[:, :, :, 2] / 2.0,
boxes1[:, :, :, 1] + boxes1[:, :, :, 3] / 2.0])
boxes1 = boxes1.permute(1, 2, 3, 0)
boxes2 = torch.stack([boxes2[:, :, :, 0] - boxes2[:, :, :, 2] / 2.0,
boxes2[:, :, :, 1] - boxes2[:, :, :, 3] / 2.0,
boxes2[:, :, :, 0] + boxes2[:, :, :, 2] / 2.0,
boxes2[:, :, :, 1] + boxes2[:, :, :, 3] / 2.0])
boxes2 = boxes2.permute(1, 2, 3, 0)
# calculate the left up point & right down point
lu = torch.max(boxes1[:, :, :, :2], boxes2[:, :, :, :2])
rb = torch.min(boxes1[:, :, :, 2:], boxes2[:, :, :, 2:])
# intersection
intersection = torch.max((rb - lu), Variable(torch.zeros(rb.size())))
inter_square = intersection[:, :, :, 0] * intersection[:, :, :, 1]
square1 = (boxes1[:, :, :, 2] - boxes1[:, :, :, 0]) * \
(boxes1[:, :, :, 3] - boxes1[:, :, :, 1])
square2 = (boxes2[:, :, :, 2] - boxes2[:, :, :, 0]) * \
(boxes2[:, :, :, 3] - boxes2[:, :, :, 1])
square1 = torch.clamp(square1, 0.00001, self.img_size*self.img_size)
square2 = torch.clamp(square2, 0.00001, self.img_size*self.img_size)
union_square = square1 + square2 - inter_square
result = inter_square / union_square
if math.isnan(torch.sum(result).data.cpu().numpy()):
Tracer()()
return result #shape = (batch_size, 7, 7, 2)
def open_logs(season, season_type, team):
driver = get_page_driver(BASE_URL)
#soup = BS(driver.page_source, 'html.parser')
close_stat_filter(driver)
select_options(driver, season, season_type, team)
press_run_button(driver)
open_all_logs(driver)
rows = get_table_rows(driver)
Tracer()()
def get_U(pulse, wd, gate=None, J_T=None, dissipation=True,
keep_runfolder=None):
"""Propagate pulse in the given rotating frame, using the non-Hermitian
Schrödinger equation, and return the resulting (non-unitary, due to
population loss) gate U"""
assert 5000 < wd < 7000
assert isinstance(pulse, QDYN.pulse.Pulse)
rf = get_temp_runfolder('evaluate_universal_hs')
n_qubit = 5
n_cavity = 6
kappa = list(np.arange(n_cavity) * 0.05)[1:-1] + [10000.0, ] # MHz
gamma = [0.012, 0.024, 0.033, 10000.0] # MHz
if not dissipation:
kappa = list(np.arange(n_cavity) * 0.0)[1:-1] + [0.0, ] # MHz
gamma = [0.0, 0.0, 0.0, 0.0] # MHz
if gate is None:
gate = GATE['BGATE']
assert isinstance(gate, QDYN.gate2q.Gate2Q)
if J_T is None:
J_T = 'sm'
model = transmon_model(
n_qubit, n_cavity, w1, w2, wc, wd, alpha1, alpha2, g, gamma, kappa,
lambda_a=1.0, pulse=pulse, dissipation_model='non-Hermitian',
gate=gate, J_T=J_T, iter_stop=1)
# write to runfolder
model.write_to_runfolder(rf)
np.savetxt(
os.path.join(rf, 'rwa_vector.dat'),
model.rwa_vector, header='rwa vector [MHz]')
gate.write(os.path.join(rf, 'target_gate.dat'), format='array')
# propagate
env = os.environ.copy()
env['OMP_NUM_THREADS'] = '4'
try:
stdout = sp.check_output(
['qdyn_prop_gate', '--internal-units=GHz_units.txt', rf], env=env,
universal_newlines=True)
except sp.CalledProcessError as exc_info:
from IPython.core.debugger import Tracer
Tracer()()
print(exc_info)
# evaluate error
for U_t in get_prop_gate_of_t(os.path.join(rf, 'U_over_t.dat')):
U = U_t
if keep_runfolder is not None:
if os.path.isdir(keep_runfolder):
rmtree(keep_runfolder)
copytree(rf, keep_runfolder)
rmtree(rf)
return U
def translate_nfhs_to_geo(state, district):
res = get_proper_state_name((state, district))
all = []
for r in res:
if r in map_names:
all.append(map_names[r])
if len(all) == 0:
Tracer()()
return all
def iou_calc(self, boxes1, boxes2): #x, y, w, h
boxes1 = torch.stack([boxes1[:, :, :, :, 0] - boxes1[:, :, :, :, 2] / 2.0,
boxes1[:, :, :, :, 1] - boxes1[:, :, :, :, 3] / 2.0,
boxes1[:, :, :, :, 0] + boxes1[:, :, :, :, 2] / 2.0,
boxes1[:, :, :, :, 1] + boxes1[:, :, :, :, 3] / 2.0])
boxes1 = boxes1.permute(1, 2, 3, 4, 0)
boxes2 = torch.stack([boxes2[:, :, :, :, 0] - boxes2[:, :, :, :, 2] / 2.0,
boxes2[:, :, :, :, 1] - boxes2[:, :, :, :, 3] / 2.0,
boxes2[:, :, :, :, 0] + boxes2[:, :, :, :, 2] / 2.0,
boxes2[:, :, :, :, 1] + boxes2[:, :, :, :, 3] / 2.0])
boxes2 = boxes2.permute(1, 2, 3, 4, 0)
# print('boxes1')
# print(boxes1)
# print('boxes2')
# print (boxes2)
# calculate the left up point & right down point
lu = torch.max(boxes1[:, :, :, :, :2], boxes2[:, :, :, :, :2])
rb = torch.min(boxes1[:, :, :, :, 2:], boxes2[:, :, :, :, 2:])
# print('lu')
# print(lu)
# print('rb')
# print(rb)
# intersection
intersection = torch.max((rb - lu), Variable(torch.zeros(rb.size())))
inter_square = intersection[:, :, :, :, 0] * intersection[:, :, :, :, 1]
# print('intersection')
# print(intersection)
# print('inter_square')
# print(inter_square)
square1 = (boxes1[:, :, :, :, 2] - boxes1[:, :, :, :, 0]) * \
(boxes1[:, :, :, :, 3] - boxes1[:, :, :, :, 1])
square2 = (boxes2[:, :, :, :, 2] - boxes2[:, :, :, :, 0]) * \
(boxes2[:, :, :, :, 3] - boxes2[:, :, :, :, 1])
square1 = torch.clamp(square1, 0.00001, self.img_size*self.img_size)
square2 = torch.clamp(square2, 0.00001, self.img_size*self.img_size)
# print('square1')
# print(square1)
# print('square2')
# print(square2)
union_square = square1 + square2 - inter_square
# print('uniou_square')
# print(union_square)
# print('iou')
# print(inter_square / union_square)
result = inter_square / union_square
# for i0 in range(100):
# for i1 in range(7):
# for i2 in range(7):
# for i3 in range(2):
# if math.isnan(result[i0,i1,i2,i3].data.cpu().numpy()):
# Tracer()()
if math.isnan(torch.sum(result).data.cpu().numpy()):
Tracer()()
return result #shape = (batch_size, 7, 7, 2)
def _do_start(_startc):
if _startc != 0:
Tracer()()
if _startc > 3:
try_again = False
print('ERROR COULD NOT START PROCESSES {} TIMES.'.format(_startc))
Tracer()()
if not try_again: return
for pidx in xrange(num_proc):
try:
procs[pidx].start()
except OSError as e:
print('Error: {}'.format(e))
if pidx is 0:
_do_start(_startc + 1)
else:
print('WARNING: pidx != 0: {}'.format(pidx))
Tracer()()
def valid_eval_tfomics(nnt, val_dat, N=14, fragLen=330):
"""
Properly calidates current CNN filters by passing filters over retained validation set N number of times and averaging
the set of predictions for each pair
inputs---
nnt: tfomics nntrainer object
val_dat: 2-D numpy array of downsampled fluorescence traces
N: number of separate start positions for each test fragment to be averaged for each pair
fragLen: length of trained CNN filter, in time points/samples
outputs---
pred_lbl: 1-D numpy array of predicted connectivity
"""
avg_F = np.mean(val_dat,axis=0)
startgap = np.ceil((val_dat.shape[1] - fragLen)/N).astype('int')
pred_lbl = np.zeros((val_dat.shape[0]*val_dat.shape[0],), dtype='float32')
# Counter for the "pred_lbl" array
cnt_u = 0
for a in range(val_dat.shape[0]):
if a%100 == 0:
print('\r' + 'X'*(a//100))
# Create batch array to send thru network
im_eval = np.empty((N*val_dat.shape[0],3,fragLen,1), dtype='float32')
# Count the number of traces in each batch
cnt = 0
for b in range(val_dat.shape[0]):
for n in range(0, val_dat.shape[1] - fragLen, startgap):
try:
im_eval[cnt,:,:,0] = np.vstack((val_dat[a,n:n+fragLen],
val_dat[b,n:n+fragLen],
avg_F[n:n+fragLen]))
except:
from IPython.core.debugger import Tracer
Tracer()()
cnt += 1
#im_eval = np.tile(im_eval,(100,1,1,1))
# Run batch through network
test = {'inputs': im_eval, 'keep_prob_dense': 1, 'keep_prob_conv': 1, 'is_training': False}
pred_stop = nnt.get_activations(test, layer='output')[:,0]
# Average output over each group of N traces
for u in range(0, len(pred_stop), N):
pred_lbl[cnt_u] = np.mean(pred_stop[u:u+N])
cnt_u += 1
return pred_lbl
def get_halotypes_for_tree_root(self,trees_dir='./',forests=None,locations=None):
if forests is None:
forests = load_forests(trees_dir)
if locations is None:
locations = self.load_locations(trees_dir)
treeparser = TreesDir(trees_dir)
p = treeparser._get_ParseFields([])
tree_types = dict()
## This is an embarrassingly parallel loop
for tree_root_id,(offset,filename) in locations.items():
tree_filename = generate_filename(trees_dir+'/'+filename)
if tree_filename is None:
raise IOError('Could not find tree file'.format(trees_dir+'/'+filename))
with opener(tree_filename,'rb') as f:
f.seek(offset)
## now read the first output for this tree root
line = f.next()
try:
X = p.pack(p.parse_line(line))
tree_types[tree_root_id] = 1 if X['pid'] == -1 else 0
except:
Tracer()()
# nforests = len(forests)
isolated_forest = []
for root_id,tree_ids in forests.items():
try:
num_fofs_max_scale = sum(tree_types[tree_id] for tree_id in tree_ids)
isolated_forest.append(False if (num_fofs_max_scale > 1) else True)
except:
Tracer()()
return tree_types,np.array(isolated_forest,dtype=bool)
def updatecontstat(self, t0 = 36*3600, t1 = 144*3600):
try:
cstat = self.cont_stat[0]
except:
cstat = self.cont_stat
if np.size(np.shape(self.onTime))>1:
onTs = np.min(np.squeeze(self.onTime))
offTs = np.max(np.squeeze(self.offTime))
else:
onTs = np.min(self.onTime)
offTs = np.max(self.offTime)
try:
tcheck = onTs <= t0 and offTs >= t1
except:
Tracer()()
try:
if not tcheck and cstat == 1:
# change this cell to NOT continuous
f1 = h5py.File(self.HDF5_tag[0], 'r+')
data = f1['neurons/neuron_'+str(self.HDF5_tag[1])+'/cont_stat']
data[...] = float(0)
f1.close()
print('Changed this cell from continuous to not.')
elif tcheck and cstat == 0:
# change this cell to continuous
f1 = h5py.File(self.HDF5_tag[0], 'r+')
data = f1['neurons/neuron_'+str(self.HDF5_tag[1])+'/cont_stat']
data[...] = float(1)
f1.close()
print('Changed this cell to continuous.')
else:
print('Continuous status stays as {}.'.format(cstat))
except:
print('Caught at 535')
Tracer()()
def fillboard(board, topleft, mines_left, rowrem, colrem):
col = len(board[0])
row = len(board)
if min(rowrem, colrem) == 2 and mines_left > 0:
return False
if rowrem == 1 or colrem == 1:
return False
else:
reduct = max(rowrem, colrem)
if reduct == colrem:
tmp = min(col - 2, mines_left + topleft[1])
board[topleft[0]] = ['*'] * tmp + ['.'] * (col - tmp)
mines_left -= tmp - topleft[1]
# Tracer()()
if mines_left > 0:
if topleft[0] + 1 == row:
return False
if topleft[0] + 3 == row:
return False
board[topleft[0] + 1][topleft[1]] = '*'
mines_left -= 1
else:
tmp = min(row - 2, mines_left + topleft[0])
for i in xrange(tmp):
board[i][topleft[1]] = '*'
mines_left -= tmp - topleft[0]
if mines_left > 0:
if topleft[1] + 1 == col:
return False
if topleft[1] + 3 == col:
return False
board[topleft[0]][topleft[1] + 1] = '*'
mines_left -= 1
if mines_left > 0:
Tracer()()
boola = False
boolb = False
boolc = False
if row - 2 >= 0 and board[row - 2][col - 1] == '*':
boola = True
if col - 2 >= 0 and board[row - 1][col - 2] == '*':
boolb = True
if col - 2 >= 0 and row - 2 >= 0 and board[row - 2][col - 2] == '*':
boolc = True
if boola and boolb and boolc:
board[row - 1][col - 1] = 'c'
return True
elif boola or boolb or boolc:
return False
board[row - 1][col - 1] = 'c'
return True
def showResults(data, schedule, results):
players_np = np.array(data)
scores_np = np.array(results['scores'][0])
results_selection = results['x']
days = results['days'][0][0]
start_index = results['start_index'][0][0]
curr_team = players_np[np.nonzero(results['curr_team'][0])].tolist()
curr_team_mask = results['curr_team'][0]
for index in xrange(days):
team_players = players_np[np.nonzero(
results_selection[index])].tolist()
team_player_scores = scores_np[np.nonzero(
results_selection[index])].tolist()
if index == 0:
coming_in = [
int(x < 0) for x in [
x - y
for x, y in zip(curr_team_mask, results_selection[index])
]
]
going_out = [
int(x > 0) for x in [
x - y
for x, y in zip(curr_team_mask, results_selection[index])
]
]
else:
going_out = [
int(x < 0) for x in [
x - y for x, y in zip(results_selection[index],
results_selection[index - 1])
]
]
coming_in = [
int(x > 0) for x in [
x - y for x, y in zip(results_selection[index],
results_selection[index - 1])
]
]
print "These players leaving:\n {0}".format(
players_np[np.nonzero(going_out)].tolist())
print "These players coming in:\n {0}".format(
players_np[np.nonzero(coming_in)].tolist())
print schedule['schedule'][index + start_index]
print team_player_scores
print team_players
# Tracer()()
print "\n\n\n"
Tracer()()
def main():
global hmm
Tracer()()
pwd = os.getcwd()
basepath = pwd + "/corpora/"
# Name of text files in /corpora/ directory
paths = ("ai.txt", "mahabharat.txt")
paths = tuple([basepath + path for path in paths])
texts = [readAndProcess(path) for path in paths]
N = 10
hmm = LetterHMM(texts, N)
Tracer()()
while True:
sample = hmm.randomwalk(2000)
# print sample
# print "\n"
obj = re.findall("[A-Z][A-Za-z:\- ,;]{19,}\.", sample)
if len(obj) < 1:
continue
for sentence in obj:
sentence.translate(string.maketrans("", ""),
string.punctuation).lower()
counts = [0, 0, 0, 0]
for word in sentence.split():
if word in hmm.word_distros[0] and word in hmm.word_distros[1]:
counts[2] += 1
elif word in hmm.word_distros[0]:
counts[0] += 1
elif word in hmm.word_distros[1]:
counts[1] += 1
else:
counts[3] += 1
if counts[0] >= 2 and counts[1] >= 2:
print sentence
print counts
print "\n"
time.sleep(1)
def main():
# ImageNetData needs path to meta.mat, path to images and path to annotations.
# The images are assumed to be in folders according to their synsets names
#imnet = ImageNetData("ILSVRC2011_devkit-2.0/data", "unpacked", "annotation")
imnet = ImageNetData(
"/nfs3group/chlgrp/datasets/ILSVRC2010/devkit-1.0/data",
bow_path="/nfs3group/chlgrp/datasets/ILSVRC2010")
features, labels = imnet.load_bow()
features_val, labels_val = imnet.load_bow('val')
from IPython.core.debugger import Tracer
tracer = Tracer(colors="LightBG")
tracer()
def forward(self, x):
x = self.pool(F.leaky_relu(self.batchnorm1(self.conv1(x))))
x = self.pool(F.leaky_relu(self.batchnorm2(self.conv2(x))))
x = self.pool(F.leaky_relu(self.batchnorm3(self.conv3(x))))
x = self.pool(F.leaky_relu(self.batchnorm4(self.conv4(x))))
x = self.batchnorm5(self.conv5(x))
x = x.contiguous().view(self.batch_size, -1)
x = self.classifier(x)
x = torch.sigmoid(x)
x = x.contiguous().view(self.batch_size, -1, self.cell_size, self.cell_size)
x = x.permute(0,2,3,1)
if math.isnan(torch.sum(x).data.cpu().numpy()):
Tracer()()
return x
def setbp():
try:
try:
from traitlets.config.configurable import MultipleInstanceError
except ImportError:
from IPython.config.configurable import MultipleInstanceError
try:
import ipdb
return ipdb.set_trace
except (ImportError, MultipleInstanceError):
from IPython.core.debugger import Tracer
return Tracer()
except Exception:
import pdb
return pdb.set_trace
def train_step(self, data, **kwargs):
# torch.autograd.set_detect_anomaly(True)
loss, data_dict, r = self.model(data['X'], actions=data['a'].cuda().float(), pretrain=kwargs['pretrain'])
self.optimizer.zero_grad()
torch.mean(-1. * loss).backward()
if self.clip_gradient:
torch.nn.utils.clip_grad_norm_(self.model.parameters(), self.clip_gradient_value)
for n, p in self.model.named_parameters():
if p.grad is not None:
# print(f'{n}: {p.grad.max()} {p.grad.min()} {p.grad.mean()}')
if not torch.isfinite(p.grad.mean()):
from IPython.core.debugger import Tracer
Tracer()()
self.optimizer.step()
return data_dict
def main():
team = 'BOS'
games = get_season_games_for_team(season='2015-16', team=team)
games = sorted(games, key=lambda x: x.date)
assert_ordered_games(games)
win_points = [0]
wins = 0
for g in games:
wins = update_wins(wins, team, g)
win_points.append(wins)
plt.scatter(x=[i for i in range(0, len(win_points))], y=win_points)
plt.show()
Tracer()()
def check_idempotent_create(expr):
"""Check that an expression is 'idempotent'"""
from qalgebra.core.abstract_algebra import Expression
print("*** CHECKING IDEMPOTENCY of %s" % expr)
if isinstance(expr, Expression):
new_expr = expr.create(*expr.args, **expr.kwargs)
if new_expr != expr:
# noinspection PyPackageRequirements
from IPython.core.debugger import Tracer
Tracer()()
print(expr)
print(new_expr)
print("*** IDEMPOTENCY OK")