Dosidicus

.gitignore

@@ -22,4 +22,5 @@ dist/*
 logs/*
 .pytest_cache/*
 .vscode/*
+.idea
 data/*

bindsnet/network/topology.py

@@ -395,6 +395,7 @@ def __init__(
         pipeline: list = [],
         manual_update: bool = False,
         traces: bool = False,
+        mask: torch.Tensor = None,
         **kwargs,
     ) -> None:
         # language=rst
@@ -408,11 +409,13 @@ def __init__(
         :param manual_update: Set to :code:`True` to disable automatic updates (applying learning rules) to connection features.
             False by default, updates called after each time step
         :param traces: Set to :code:`True` to record history of connection activity (for monitors)
+        :param mask: A mask to zero out weights
         """
 
         super().__init__(source, target, device, pipeline, **kwargs)
         self.traces = traces
         self.manual_update = manual_update
+        self.mask = mask
         if self.traces:
             self.activity = None
 
@@ -487,6 +490,8 @@ def update(self, **kwargs) -> None:
             # Pipeline learning
             for f in self.pipeline:
                 f.update(**kwargs)
+                if type(f).__name__ == 'Weight' and self.mask is not None:
+                    f.value.masked_fill_(self.mask, 0)
 
     def normalize(self) -> None:
         # language=rst

examples/dosidicus/network.py

@@ -0,0 +1,37 @@
+import torch
+
+from bindsnet.network import Network
+from bindsnet.network.nodes import LIFNodes
+from bindsnet.network.topology import MulticompartmentConnection
+from bindsnet.network.topology_features import Weight, Bias
+from bindsnet.learning.MCC_learning import PostPre
+
+
+network = Network(dt=1.0)
+neurons_number = 7
+layer = LIFNodes(n=neurons_number, traces=True)
+network.add_layer(layer, name="input")
+network.add_layer(layer, name="output")
+mask = ~torch.tril(torch.ones((neurons_number, neurons_number)), diagonal=-1).bool()
+weight = Weight(
+    name='weight_feature',
+    value=torch.rand(neurons_number, neurons_number) * mask,
+    learning_rule=PostPre,
+    nu=(1e-4, 1e-2)
+)
+bias = Bias(
+    name='bias_feature',
+    value=torch.rand(neurons_number, neurons_number)
+)
+connection = MulticompartmentConnection(
+    source=layer,
+    target=layer,
+    pipeline=[weight, bias],
+    mask=mask,
+    device='cpu'
+)
+network.add_connection(connection, source="input", target="output")
+network.run(
+    inputs={"input": torch.bernoulli(torch.rand(250, neurons_number)).byte()},
+    time=250
+)