Connections to ensembles have proper synapse

This fixes (for example) neuron->ensemble connections (of which
we currently have no test cases).

Unfortunately, this also changes the behaviour of node->ensemble
connections, which caused node_ens_ens to fail. The synapse set on
the connection (None) overrides the default INTER_TAU synapse,
removing the filtering on the ReLU neurons used to turn the node
values into spikes.

To fix this, I lowered the frequency of the input signal.

docs/examples/node_ens_ens.py

@@ -18,7 +18,7 @@ def a_fn(x):
 nengo_loihi.set_defaults()
 with nengo.Network(seed=seed) as model:
     u = nengo.Node(
-        output=nengo.processes.WhiteSignal(tend, high=5, seed=seed + 1),
+        output=nengo.processes.WhiteSignal(tend, high=2, seed=seed + 1),
         size_out=d)
     up = nengo.Probe(u, synapse=None)
 

nengo_loihi/builder.py

@@ -282,6 +282,7 @@ def build_ensemble(model, ens):
         group.bias[:] = bias
         model.build(ens.neuron_type, ens.neurons, group)
 
+    # set default filter just in case no other filter gets set
     group.configure_filter(INTER_TAU, dt=model.dt, default=True)
 
     if ens.noise is not None:
@@ -469,7 +470,7 @@ def build_connection(model, conn):
     transform = get_samples(
         conn.transform, conn.size_out, d=conn.size_mid, rng=rng)
 
-    tau_s = 0.0
+    tau_s = 0.0  # `synapse is None` gets mapped to `tau_s = 0.0`
     if isinstance(conn.synapse, nengo.synapses.Lowpass):
         tau_s = conn.synapse.tau
     elif conn.synapse is not None:
@@ -526,6 +527,7 @@ def build_connection(model, conn):
 
     mid_cx = pre_cx
     mid_axon_inds = slice(None)
+    post_tau = tau_s
     if needs_interneurons and not isinstance(conn.post_obj, Neurons):
         # --- add interneurons
         assert weights.ndim == 2
@@ -578,7 +580,9 @@ def build_connection(model, conn):
             weights2 = 0.5 * gain * np.vstack([weights,
                                                -weights] * INTER_N).T
 
+        # use tau_s for filter into interneurons, and INTER_TAU for filter out
         dec_cx.configure_filter(tau_s, dt=model.dt)
+        post_tau = INTER_TAU
 
         dec_syn.set_full_weights(weights2)
         dec_cx.add_synapses(dec_syn)
@@ -632,7 +636,7 @@ def build_connection(model, conn):
         ax.target = syn
         mid_cx.add_axons(ax)
 
-        post_cx.configure_filter(tau_s, dt=model.dt)
+        post_cx.configure_filter(post_tau, dt=model.dt)
 
         if conn.learning_rule_type is not None:
             raise NotImplementedError()
@@ -654,7 +658,7 @@ def build_connection(model, conn):
         ax.target = syn
         mid_cx.add_axons(ax)
 
-        post_cx.configure_filter(tau_s, dt=model.dt)
+        post_cx.configure_filter(post_tau, dt=model.dt)
 
         if conn.learning_rule_type is not None:
             raise NotImplementedError()
@@ -667,6 +671,8 @@ def build_connection(model, conn):
         mid_ax.target_inds = mid_axon_inds
         mid_cx.add_axons(mid_ax)
         model.objs[conn]['mid_axons'] = mid_ax
+
+        post_cx.configure_filter(post_tau, dt=model.dt)
     elif isinstance(conn.post_obj, Node):
         raise NotImplementedError()
     else:

nengo_loihi/tests/test_examples.py

@@ -100,15 +100,15 @@ def test_node_ens_ens(allclose, plt):
     plt.subplot(413)
     plt.plot(t, a[:, 0] ** 2, c="b", label="a[0]**2")
     plt.plot(t, b[:, 0], c="g", label="b[0]")
-    plt.ylim([-1, 1])
+    plt.ylim([-0.05, 1])
     plt.legend(loc=0)
 
     plt.subplot(414)
-    plt.plot(t, a[:, 0] ** 2, c="b", label="a[1]**2")
-    plt.plot(t, b[:, 0], c="g", label="b[1]")
-    plt.ylim([-1, 1])
+    plt.plot(t, a[:, 1] ** 2, c="b", label="a[1]**2")
+    plt.plot(t, b[:, 1], c="g", label="b[1]")
+    plt.ylim([-0.05, 1])
     plt.legend(loc=0)
 
     tmask = t > 0.1  # ignore transients at the beginning
-    assert allclose(a[tmask], np.clip(u[tmask], -1, 1), atol=0.4, rtol=0.25)
-    assert allclose(b[tmask], a[tmask]**2, atol=0.35, rtol=0.0)
+    assert allclose(a[tmask], np.clip(u[tmask], -1, 1), atol=0.1, rtol=0.1)
+    assert allclose(b[tmask], a[tmask]**2, atol=0.1, rtol=0.2)