def calc_ged(recepie1, recepie2, timeout_val=600):
start_time = time.time()
G1 = make_graph(recepie1)
G2 = make_graph(recepie2)
ged = None
try:
status = "OK"
with timeout(Quota(timeout_val), exception=RuntimeError):
for ged in nx.optimize_graph_edit_distance(G1, G2, lambda n1, n2: n1['op'] == n2['op']):
pass
except RuntimeError as e:
status = "Timeout"
except Exception as e:
status = "Exception: " + str(e)
return {
"recepie_i": recepie1,
"recepie_j": recepie2,
"ged": ged,
"time": time.time() - start_time,
"status": status
}
def main(distribution, count):
"""dnolul"""
past_days, past_values, fieldnames = datasets.btc_csv_data()
model = models.linear_regression(past_days, past_values)
future_days = [[x * distribution + past_days[-1][0]]
for x in xrange(count)]
future_values = model.predict(future_days)
total_days = past_days + future_days
total_values = past_values + future_values
utils.make_graph(total_days, total_values, fieldnames[0], fieldnames[1])
def goto(start: int, end: int) -> Dict[str, Any]:
sess = Session(bind=engine)
graph = make_graph(sess)
try:
paths = bf_paths(graph, start)
path = paths[(start, end)]
except KeyError as e:
try:
paths = df_paths(graph, start)
path = paths[(start, end)]
except KeyError as e:
print("search for path failed. ")
return "search for path failed"
print(f"it will take roughly under {9 * len(path) / 60} minutes to make {len(path)} moves")
for direction in path:
moved_to = move(direction)
curr_room_id = moved_to['room_id']
exits = moved_to['exits']
curr_room = sess.query(Room).filter(Room.room_id==curr_room_id).all()[0]
directions = dict()
for exit in exits:
directions[exit] = eval(f'curr_room.{exit}_to')
for direction, room_id in directions.items():
if not room_id:
goto_then_move_dir.append((curr_room_id, direction))
print('appending a tuple! ', end='')
print(f"backtracking... {moved_to['room_id']}: {moved_to['title']}")
return moved_to
def select_isomer(self, mother, scaffold, latent_vector):
"""\
Return an isomer-selection vector and the answer one-hot.
Returns
-------
retval: isomer-selection latent vector of shape (len(isomers),)
target: answer one-hot of shape (len(isomers),)
where `isomers` are the isomers of `mother`.
"""
#sample possible isomer
m_mother = Chem.MolFromSmiles(mother)
isomer_candidates = utils.enumerate_molecule(
mother) # list of isomer SMILESs
isomers = []
for s in isomer_candidates:
m = Chem.MolFromSmiles(s)
if m.HasSubstructMatch(Chem.MolFromSmiles(scaffold),
useChirality=True):
isomers.append(s)
graph_vectors = []
#make graph for each isomer
for s in isomers:
g, h = utils.make_graph(s,
extra_atom_feature=True,
extra_bond_feature=True)
self.embede_graph(g, h)
for k in range(len(self.prop_select_isomer_U)):
self.mpnn(g, h, self.prop_select_isomer_U[k],
self.prop_select_isomer_C[k], latent_vector)
graph_vectors.append(utils.average_node_state(h))
graph_vectors = torch.cat(graph_vectors, 0)
# -> (len(isomers), dim_of_node_vector)
latent_vectors = latent_vector.repeat(len(isomers), 1)
# -> (len(isomers), dim_of_node_vector + N_conditions)
retval = torch.cat([graph_vectors, latent_vectors], -1)
# -> (len(isomers), 2*dim_of_node_vector + N_conditions)
#FC layer
retval = F.relu(self.select_isomer1(retval))
retval = F.relu(self.select_isomer2(retval))
retval = self.select_isomer3(retval)
retval = retval.view(-1) # (len(isomers),)
retval = torch.sigmoid(retval)
target = []
#check which isomer is same as mother
for s in isomers:
if m_mother.HasSubstructMatch(Chem.MolFromSmiles(s),
useChirality=True):
target.append(1)
else:
target.append(0)
target = utils.create_var(torch.Tensor(target)) # (len(isomers),)
return retval, target, isomers
def callback_inline(call):
date_str, today = call.data.split(' ')
# proxies = settings.proxy
data_tb = utils.get_data_tb(today) #, proxies=proxies)
plt = utils.make_graph(data_tb, date_str, save=False)
plot_buffer = BytesIO()
with plot_buffer as plot:
plt.savefig(plot, format='png')
plot.seek(0)
bot.send_photo(call.message.chat.id, plot.getvalue())
r = precision_recall_fscore_support(dev_x, pred_x, average='micro')[1]
f = precision_recall_fscore_support(dev_x, pred_x, average='micro')[2]
total_precision.append(p)
total_recall.append(r)
total_f1.append(f)
print("intermediate metrics: ", p, r, f)
X = []
y = []
X_pred = []
y_pred = []
graph = defaultdict(dict)
graph2 = defaultdict(dict)
data_val = val
make_graph(data_val, val_y, X, y, pred_x, X_pred, y_pred)
characters = entities_2_graph(X, y)
characters2 = entities_2_graph(X, y_pred)
graph = prepare_eval(characters, y, graph, param='gold')
graph2 = prepare_eval(characters2, y_pred, graph2, param='pred')
if setting != "2class":
eval(graph, graph2, graph_total_precision, graph_total_recall,
graph_total_f1)
else:
eval_2class(graph, graph2, graph_total_precision, graph_total_recall,
graph_total_f1)
print(setting, 'indicators:', indicators, 'window_size:', window_size)
print(np.mean(total_precision), np.mean(total_recall), np.mean(total_f1))
import sys
sys.path.append('midterm_presentation/scripts')
from utils import make_graph
make_graph(with_red_circle=True)
def __init__(self, pathToPlugin, pathToFolder, parameterString, osSystem):
self.pathToFolder = pathToFolder
self.parameterString = parameterString
self.osSystem = int(osSystem)
self.originalData = []
self.randomData = []
if self.osSystem == 1:
self.pathToPlugin = '\\'.join(pathToPlugin.split('\\')[:-1])
else:
self.pathToPlugin = '/'.join(pathToPlugin.split('/')[:-1])
# set parameters
self.randn, self.sigma, self.block, self.small, self.factr = self.parameterString.split(
",")
self.randn = int(self.randn)
self.sigma = float(self.sigma)
self.block = float(self.block)
self.small = float(self.small)
self.factr = float(self.factr)
self.get_parameters()
# go to selected folder and search filtered image
os.chdir(self.pathToFolder)
self.filterImg = glob.glob('*_filter.tif')
if len(self.filterImg) == 0:
print(
"ERROR: No pre-processed image ('*_filter.tif') was found. Select pre-processing to create the pre-processed image."
)
else:
self.imgRaw = skimage.io.imread(self.filterImg[0],
plugin='tifffile')
if len(self.imgRaw.shape) > 2:
if self.imgRaw.shape[2] in (3, 4):
self.imgRaw = np.swapaxes(self.imgRaw, -1, -3)
self.imgRaw = np.swapaxes(self.imgRaw, -1, -2)
self.slices = len(self.imgRaw)
else:
self.slices = 1
# find and open image mask
self.maskImg = glob.glob('*_mask.tif')
self.mask = skimage.io.imread(self.maskImg[0],
plugin='tifffile') > 0
print('Start extraction of image', self.filterImg[0])
for i in range(self.slices):
print('\n', i + 1, 'of', self.slices)
if self.slices == 1:
self.imgSlice = self.imgRaw.copy()
else:
self.imgSlice = self.imgRaw[i]
self.imgGaussian = skimage.filters.gaussian(
self.imgSlice, self.sigma)
self.imgTube, self.imgSkeleton = utils.skeletonize_graph(
self.imgGaussian, self.mask, self.sigma, self.block,
self.small, self.factr)
if np.sum(self.imgSkeleton) == 0:
print(
"ERROR: No skeleton was extracted from the selected image. Check the parameters and try again."
)
else:
self.imgNodes = utils.node_graph(self.imgSkeleton > 0,
self.imgGaussian)
self.originalGraph, self.originalPosition = utils.make_graph(
self.imgNodes, self.imgGaussian)
self.originalNormalizedGraph, self.originalProperties, self.unifiedGraph = self.processGraph(
self.originalGraph, self.originalPosition,
self.imgGaussian, self.mask)
self.originalData.append([
i, self.originalNormalizedGraph, self.originalPosition,
self.originalProperties
])
for r in range(self.randn):
self.randomGraph, self.randomPosition = utils.randomize_graph(
self.unifiedGraph, self.originalPosition,
self.mask)
self.randomNormalizedGraph, self.randomProperties, _ = self.processGraph(
self.randomGraph, self.randomPosition,
self.imgGaussian, self.mask)
self.randomData.append([
i, self.randomNormalizedGraph, self.randomPosition,
self.randomProperties
])
if i == 0:
print('Export plot.')
self.plotSkeleton(self.originalData, self.randomData)
if np.sum(self.imgSkeleton) != 0:
print('\nExport data.')
self.saveData(self.originalData, self.randomData)
import sys
sys.path.append('midterm_presentation/scripts')
from utils import make_graph
make_graph(with_red_circle=False)
imO=skimage.io.imread(path+aa,plugin='tifffile')[0] # open first frame of actin image
imT=skimage.io.imread(path+gg,plugin='tifffile')[0] # open Golgi image
mask=skimage.io.imread(path+'mask.tif',plugin='tifffile')>0 # open mask
track=utils.xmlread(path+'track.xml') # read Golgi tracking results
T=len(track) # get number of tracks
print('extract','segment')
imI=utils.im2d3d(imO) # if 2D image convert to 3D
imG=skimage.filters.gaussian(imI,sigma) # apply Gaussian filter
imR,imA=utils.skeletonize_graph(imG,mask,sigma,block,small,factr) # filter and skeletonize actin image
imE=utils.node_graph(imA>0,imG) # detect filaments and network nodes
print('extract','graph')
gBo,pos=utils.make_graph(imE,imG) # construct graph from filament and node image
gBu=utils.unify_graph(gBo) # project multigraph to simple graph
gBc=utils.connect_graph(gBu,pos,imG) # connect disconnected components of graph
gBx=utils.centralize_graph(gBc) # compute edge centrality measures
gBn=utils.normalize_graph(gBx) # normalize total edge capacity to one
#%%############################################################################# plot data
print('export','plot')
aspect=2.0 # set aspect ratio
alpha=1.0 # set transparency
lw=1.5 # set line width
plt.clf()
gs=mpl.gridspec.GridSpec(1,2,width_ratios=[1,1],height_ratios=[1],left=0.01,bottom=0.01,right=0.99,top=0.99,wspace=0.1,hspace=0.1)
def sample(self,
s1=None,
s2=None,
latent_vector=None,
condition1=None,
condition2=None,
stochastic=False):
"""\
Parameters
----------
s1: whole SMILES str
If given, its graph becomes a latent vector to be decoded.
s2: scaffold SMILES str
Must be given other than None.
latent_vector: None | torch.autograd.Variable
A latent vector to be decoded.
Not used if `s1` is given.
If both `latent_vector` and `s1` are None,
a latent vector is sampled from the standard normal.
condition1: list[float] | None
[ target_value1, target_value2, ... ]
If None, target values are sampled from uniform [0, 1].
Can be an empty list for unconditional sampling.
condition2: list[float] | None
[ scaffold_value1, scaffold_value2, ... ]
If None, scaffold values are sampled from uniform [0, 1].
Can be an empty list for unconditional sampling.
stochastic: bool
See `utils.probability_to_one_hot`.
Returns
-------
scaffold_g_save: OrderedDict[int, list[tuple[torch.autograd.Variable, int]]]
A new dict of edge one-hot vectors and partner-node indices
generated from the given scaffold `s2`.
scaffold_h_save: OrderedDict[int, torch.autograd.Variable]
A new dict of node one-hot vectors generated from the given scaffold `s2`.
"""
max_add_nodes = 100
max_add_edges = 5
if s2 is None:
print('when you sample, you must give scaffold')
return None
# Embede the scaffold edge/node vectors.
# If `s1` is given, convert its graph to a latent vector.
if s1 is not None:
g_save, h_save, scaffold_g_save, scaffold_h_save = utils.make_graphs(
s1, s2)
if g_save is None and h_save is None:
return None
g, h, scaffold_g, scaffold_h = utils.make_graphs(
s1, s2, extra_atom_feature=True, extra_bond_feature=True)
self.embede_graph(g, h)
self.embede_graph(scaffold_g, scaffold_h)
self.encode(g, h)
encoded_vector = self.cal_encoded_vector(h)
latent_vector, mu, logvar = self.reparameterize(encoded_vector)
# `mu` and `logvar` are not used further.
# If `s1` is None, sample a latent vector from the standard normal.
elif s1 is None:
scaffold_g_save, scaffold_h_save = utils.make_graph(s2)
if scaffold_g_save is None and scaffold_h_save is None:
return None
scaffold_g, scaffold_h = utils.make_graph(s2,
extra_atom_feature=True,
extra_bond_feature=True)
self.embede_graph(scaffold_g, scaffold_h)
if latent_vector is None: # Sampling
latent_vector = utils.create_var(
torch.randn(1, self.dim_of_node_vector))
# Sample condition values if not given.
if condition1 is None or condition2 is None:
assert not self.N_conditions % 2
condition1 = np.random.rand(self.N_conditions // 2)
condition2 = np.random.rand(self.N_conditions // 2)
# A condition torch.FloatTensor of shape (1, N_conditions):
condition = utils.create_var(torch.Tensor(condition1 + condition2))
if condition.shape:
condition = condition.unsqueeze(0)
latent_vector = torch.cat([latent_vector, condition], -1)
# -> (1, dim_of_node_vector + N_conditions)
self.init_scaffold_state(scaffold_g, scaffold_h, condition)
for null_index1 in range(max_add_nodes):
new_node = self.add_node(scaffold_g, scaffold_h,
latent_vector) # (1, N_atom_features)
new_node = utils.probability_to_one_hot(new_node, stochastic)
# Recall our definition of the termination vector:
if np.argmax(new_node.data.cpu().numpy().ravel()
) == N_atom_features - 1:
break
idx = len(scaffold_h)
scaffold_h_save[idx] = new_node
scaffold_h[idx] = self.init_node_state(scaffold_h, new_node)
for null_index2 in range(max_add_edges):
new_edge = self.add_edge(scaffold_g, scaffold_h,
latent_vector) # (1, N_bond_features)
new_edge = utils.probability_to_one_hot(new_edge, stochastic)
# Recall our definition of the termination vector:
if np.argmax(new_edge.data.cpu().numpy().ravel()
) == N_bond_features - 1:
break
selected_node = self.select_node(scaffold_g, scaffold_h,
latent_vector).view(1, -1)
# -> (1, len(scaffold_h)-1)
# Index of the selected node (int)
selected_node = list(scaffold_h.keys())[np.argmax(
utils.probability_to_one_hot(
selected_node, stochastic).data.cpu().numpy().ravel())]
if idx not in scaffold_g_save:
scaffold_g_save[idx] = []
scaffold_g[idx] = []
scaffold_g_save[idx].append((new_edge, selected_node))
scaffold_g[idx].append(
(self.init_edge_state(scaffold_h,
new_edge), selected_node))
# Add the same edge in the opposite direction.
if selected_node not in scaffold_g_save:
scaffold_g_save[selected_node] = []
scaffold_g[selected_node] = []
scaffold_g_save[selected_node].append((new_edge, idx))
scaffold_g[selected_node].append(
(self.init_edge_state(scaffold_h, new_edge), idx))
try:
new_smiles = utils.graph_to_smiles(scaffold_g_save,
scaffold_h_save)
new_smiles = Chem.MolToSmiles(Chem.MolFromSmiles(new_smiles),
isomericSmiles=False)
except:
return None
selected_isomer, target, isomers = self.select_isomer(
new_smiles, s2, latent_vector)
selected_isomer = np.argmax(
utils.probability_to_one_hot(selected_isomer,
stochastic).data.cpu().numpy())
return isomers[selected_isomer]
"""
print('Acc: {:.6f}'.format(WA))
logging.info(
"epoch: {}, train_loss: {:.6f}, valid_loss: {:.6f}, Valid Acc: {:.6f}"
.format(epoch, train_loss, valid_loss, WA))
if valid_loss < loss_min:
torch.save(net.state_dict(), MODEL_PATH + NAME + '.net')
loss_min = valid_loss
best_epoch = epoch
# train fin
print("\nbest_epoch : {}".format(best_epoch))
logging.info("\nbest epoch : {}".format(best_epoch))
print("\nTest")
logging.info("\nTest")
net = model.lld_blstm_attn(hidden_dim=lstm_hidden).to(device)
net.load_state_dict(torch.load(MODEL_PATH + NAME + '.net'))
mat_test, test_loss = test(net, criterion, test_loader)
WA, _ = utils.evaluate(mat_test, 4)
print(np.array(mat_test))
print(WA)
logging.info(np.array(mat_test))
logging.info(WA)
utils.make_graph(tls, vls, vac, GRAPH)
