dodox/petriDish
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Made progress, but not working

Browse Source
This commit is contained in:
Dominik Moritz Roth 2021-10-24 18:07:11 +02:00
parent 2c9de61fee
commit ba26666329
2 changed files with 14 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
train.py
View File

@ -59,16 +59,16 @@ def evaluate(dataloader, module):
y_true_batch = y_true_batch.to(device, dtype) # (batch_size,) y_true_batch = y_true_batch.to(device, dtype) # (batch_size,)
y_pred_batch, p, halting_step = module(x_batch) y_pred_batch, p, halting_step = module(x_batch)
y_halted_batch = y_pred_batch.gather( y_halted_batch = y_pred_batch.index_select(
dim=0, dim=0,
index=halting_step[None, :] - 1, index=halting_step - 1,
)[ )[
0 0
] # (batch_size,) ] # (batch_size,)
# Computing single metrics (mean over samples in the batch) # Computing single metrics (mean over samples in the batch)
accuracy_halted = ( accuracy_halted = (
((y_halted_batch > 0) == y_true_batch).to(torch.float32).mean() abs(y_halted_batch - y_true_batch).to(torch.float32).mean()
) )
metrics_single_["accuracy_halted"].append(accuracy_halted) metrics_single_["accuracy_halted"].append(accuracy_halted)
@ -78,9 +78,9 @@ def evaluate(dataloader, module):
# Computing per step metrics (mean over samples in the batch) # Computing per step metrics (mean over samples in the batch)
accuracy = ( accuracy = (
((y_pred_batch > 0) == y_true_batch[None, :]) (abs(y_pred_batch - y_true_batch[None, :]))
.to(torch.float32) .to(torch.float32)
.mean(dim=1) .mean(dim=1).mean(dim=1)
) )
metrics_per_step_["accuracy"].append(accuracy) metrics_per_step_["accuracy"].append(accuracy)

11
utils.py
View File

@ -102,17 +102,21 @@ class PetriDishNet(nn.Module):
self.allow_halting = allow_halting self.allow_halting = allow_halting
self.synapses = SparseLinear(n_neurons, n_neurons, self.synapses = SparseLinear(n_neurons, n_neurons,
sparsity=0.9, sparsity=0.8,
#sparsity=0.1,
small_world=True, small_world=True,
dynamic=True, dynamic=True,
deltaT=6000, #deltaT=6000,
deltaT=250,
Tend=150000, Tend=150000,
#Tend=5000,
alpha=0.1, alpha=0.1,
max_size=1e8) max_size=1e8)
self.ionDucts = ActivationSparsity(alpha=0.1, self.ionDucts = ActivationSparsity(alpha=0.1,
beta=1.5, beta=1.5,
act_sparsity=0.65) act_sparsity=0.75)
#act_sparsity=0.10)
def forward(self, x): def forward(self, x):
"""Run forward pass. """Run forward pass.
@ -164,6 +168,7 @@ class PetriDishNet(nn.Module):
else: else:
lambda_n = torch.sigmoid(state[:, 0]) lambda_n = torch.sigmoid(state[:, 0])
# Store releavant outputs # Store releavant outputs
y_list.append(state[:, (self.n_in+1):(self.n_in+self.n_out+1)]) y_list.append(state[:, (self.n_in+1):(self.n_in+self.n_out+1)])
p_list.append(un_halted_prob * lambda_n) p_list.append(un_halted_prob * lambda_n)