fix: bound state-conditioned feedback features

1 files changed, 9 insertions, 8 deletions

diff --git a/sdil/core.py b/sdil/core.py
index e66dc39..64c7c79 100644
--- a/sdil/core.py
+++ b/sdil/core.py
@@ -265,10 +265,11 @@ class SDILNet:
     def vectorizer(self, l, c, h_l=None, context=None):
         """Return the learned instructional component of apical activity.
 
-        ``soma_gated`` adds ``h_i G_i c`` independently in every cell.
-        ``context_gated`` adds a linear readout of ``c outer context``. Both
-        vanish when c=0, so neutral-period predictor identification remains
-        unchanged.
+        ``soma_gated`` adds ``tanh(h_i) G_i c`` independently in every cell.
+        ``context_gated`` adds a linear readout of ``c outer tanh(context)``.
+        Bounding the state feature prevents a bilinear positive-feedback loop;
+        both terms vanish when c=0, so neutral-period predictor identification
+        remains unchanged.
         """
         value = c @ self.A[l].t()
         if self.vectorizer_mode == "linear":
@@ -276,10 +277,10 @@ class SDILNet:
         if self.vectorizer_mode == "soma_gated":
             if h_l is None:
                 raise ValueError("soma-gated vectorizer requires somatic activity")
-            return value + h_l * (c @ self.A_gate[l].t())
+            return value + torch.tanh(h_l) * (c @ self.A_gate[l].t())
         if context is None:
             raise ValueError("context-gated vectorizer requires a context state")
-        features = (c.unsqueeze(2) * context.unsqueeze(1)).flatten(1)
+        features = (c.unsqueeze(2) * torch.tanh(context).unsqueeze(1)).flatten(1)
         return value + features @ self.A_gate[l].t()
 
     def apical(self, l, c, h_l=None, context=None):
@@ -551,10 +552,10 @@ def sdil_step(net, x, y, y_onehot, cfg, step, prev_error=None):
                 dA = calibration_error.t() @ c / B     # (n_l, n_classes)
                 net.A[l] += cfg.eta_A * dA
                 if net.vectorizer_mode == "soma_gated":
-                    dgate = (calibration_error * h[l + 1]).t() @ c / B
+                    dgate = (calibration_error * torch.tanh(h[l + 1])).t() @ c / B
                     net.A_gate[l] += cfg.eta_A * dgate
                 elif net.vectorizer_mode == "context_gated":
-                    features = (c.unsqueeze(2) * context.unsqueeze(1)).flatten(1)
+                    features = (c.unsqueeze(2) * torch.tanh(context).unsqueeze(1)).flatten(1)
                     dgate = calibration_error.t() @ features / B
                     net.A_gate[l] += cfg.eta_A * dgate