def __iter__(self):
batch = []
for idx in range(len(self.arrays[0])):
batch.append(idx)
if len(batch) == self.batch_size:
yield [tensor(arr[batch].copy()) for arr in self.arrays]
batch = []
if batch:
yield [tensor(arr[batch].copy()) for arr in self.arrays]
def spike(x):
time_range = 2247
truth, shape = loader.loader(
'D:\\document\\体外网络发放数据\\a195e390b707a65cf3f319dbadbbc75f_6b245c0909b1a21072dd559d4203e15b_8.txt')
ture = truth[200::, :]
start =tf.tensor(truth[0:200, :],dtype=tf.float32)
ture = tf.tensor(ture, dtype=tf.float32)
spike = tf.zeros(size=(shape[0]-200,shape[1]),dtype=tf.float32)
spike = tf.concat([start,spike],axis=0)
print(spike.size())
def condition(self,time):
time <time_range
def __iter__(self):
for idxs in self.sampler(self.val):
hvg_input = tensor(self.hvg_input[idxs].copy())
hvg_target = tensor(self.hvg_target[idxs].copy())
p_target = tensor(self.p_target[idxs].copy())
if (self.lvg_input is not None) and (self.lvg_target is not None):
lvg_input = tensor(self.lvg_input[idxs].copy())
lvg_target = tensor(self.lvg_target[idxs].copy())
else:
lvg_input = None
lvg_target = None
yield [hvg_input, lvg_input], hvg_target, lvg_target, p_target
def encode(sm_list, pad_size=50):
"""
Encoder list of smiles to tensor of tokens
"""
res = []
lens = []
for s in sm_list:
tokens = ([1] + [__t2i[tok]
for tok in smiles_tokenizer(s)])[:pad_size - 1]
lens.append(len(tokens))
tokens += (pad_size - len(tokens)) * [2]
res.append(tokens)
#zz return torch.tensor(res).long(), lens
tf.tensor(res).long(), lens
def gen_context_from_nrp(self, prev_context, nrp_sym, position):
if nrp_sym in self.sym2char:
rapper = self.sym2char[nrp_sym]
rap_vec = self.rapper_vectors[rapper]
prev_context["rapper" + str(position)] = tf.cast(
tf.tensor(rap_vec), tf.int32)
return prev_context
def __init__(
self,
initial_conditions=[0, 2, 20],
model_parameters=[28., 10., 8. / 3.],
final_time=50,
time_steps=5000):
self.initial_conditions = np.array(initial_conditions)
self.model_parameters = model_parameters
self.final_time = final_time
self.time_steps = time_steps
self.state = tf.tensor()
def pad_neighbors(self, mesh, size):
""" extracts one-ring neighbors (4x) -> mesh.edge_to_neighbors
which is of size #edges x 4
add the edge_id itself to make #edges x 5
then pad to desired size e.g., size x 5
"""
padded_neighbors = tf.tensor(mesh.edge_to_neighbors).float()
padded_neighbors = tf.concat((tf.expand_dims(tf.range(len(mesh.edges)).float(), 1), padded_neighbors), axis=1)
padded_neighbors = tf.pad(padded_neighbors, [0, 0, 0, size-len(mesh.edges)], 'CONSTANT')
padded_neighbors = tf.expand_dims(padded_neighbors, 0)
return padded_neighbors
def __iter__(self):
batch = []
for idx in range(len(self.array)):
batch.append(idx)
if len(batch) == self.batch_size:
yield tensor(self.array[batch].copy())
batch = []
if batch:
yield self.array[batch].copy()
def step(self, closure=None):
"""Performs a single optimization step.
Arguments:
closure (callable, optional): A closure that reevaluates the model
and returns the loss.
"""
loss = None
if closure is not None:
with tf.GradientTape():
loss = closure()
for group in self.param_groups:
for p in group['params']:
param_norm = tf.max(unitwise_norm(p),
tf.Variable(group['eps']).to(p.device))
grad_norm = unitwise_norm(p.grad)
max_norm = param_norm * group['clipping']
trigger = grad_norm > max_norm
clipped_grad = p.grad * \
(max_norm / tf.max(grad_norm,
tf.tensor(1e-6).to(grad_norm.device)))
p.grad.data.copy_(tf.where(trigger, clipped_grad, p.grad))
for group in self.param_groups:
weight_decay = group['weight_decay']
momentum = group['momentum']
dampening = group['dampening']
nesterov = group['nesterov']
for p in group['params']:
if p.grad is None:
continue
d_p = p.grad
if weight_decay != 0:
d_p = d_p.add(p, alpha=weight_decay)
if momentum != 0:
param_state = self.state[p]
if 'momentum_buffer' not in param_state:
buf = param_state['momentum_buffer'] = d_p.numpy()
else:
buf = param_state['momentum_buffer']
buf.mul_(momentum).add_(d_p, alpha=1 - dampening)
if nesterov:
d_p = d_p.add(buf, alpha=momentum)
else:
d_p = buf
p.add_(d_p, alpha=-group['lr'])
return loss
def loss(self, f, y):
"""The average loss across batch examples.
Computes the average hinge loss.
Args:
f: Tensor containing the output of the forward operation.
y(tf.placeholder): Tensor containing the ground truth label.
Returns:
(1): Returns the loss function tensor.
"""
if -tf.matmul(y, f) > 0:
return -tf.matmul(y, f)
else:
return tf.tensor(0.0)
def _intr_reward(self, seq):
inputs = seq['feat']
if self.config.disag_action_cond:
action = tf.cast(seq['action'], inputs.dtype)
inputs = tf.concat([inputs, action], -1)
preds = [head(inputs).mode() for head in self._networks]
disag = tf.tensor(preds).std(0).mean(-1)
if self.config.disag_log:
disag = tf.math.log(disag)
reward = self.config.expl_intr_scale * self.intr_rewnorm(disag)[0]
if self.config.expl_extr_scale:
reward += self.config.expl_extr_scale * self.extr_rewnorm(
self.reward(seq))[0]
return reward
def __iter__(self):
for idxs in self.sampler(self.val):
yield [tensor(arr[idxs].copy()) for arr in self.arrays]
def __iter__(self):
for idxs in self.sampler(self.val):
yield (tensor(self.embedding[idxs].copy()),
tensor(self.sizefactor[idxs].copy())), tensor(
self.target[idxs].copy())
def get_press_time_array(press_time_file):
time_press_array = np.load(press_time_file.eval())
# del train_path
# del press_time_file
return tf.tensor(time_press_array)
def gen_context_from_nrp(self, prev_context, nrp_sym, position):
if nrp_sym in self.sym2char:
rapper = self.sym2char[nrp_sym]
rap_vec = self.rapper_vectors[rapper]
prev_context["rapper"+str(position)] = tf.cast(tf.tensor(rap_vec), tf.int32)
return prev_context
