ci:spirosChv-patch-1

tutorials/W3D5_NetworkCausality/W3D5_Tutorial2.ipynb

@@ -328,7 +328,8 @@
         "\n",
         "    # set the timescale of the dynamical system to about 100 steps\n",
         "    _, s_vals, _ = np.linalg.svd(A_0)\n",
-        "    A = A_0 / (1.01 * s_vals[0])\n",
+        "    epsilon = np.finfo(float).eps  # add a small number to avoid division with zero\n",
+        "    A = A_0 / (1.01 * s_vals[0] + epsilon)\n",
         "\n",
         "    return A\n",
         "\n",
@@ -356,7 +357,6 @@
         "                                                np.eye(n_neurons))\n",
         "        X[:, t + 1] = sigmoid(A.dot(X[:, t]) + epsilon)\n",
         "\n",
-        "        assert epsilon.shape == (n_neurons,)\n",
         "    return X\n",
         "\n",
         "\n",
@@ -1405,7 +1405,6 @@
       "outputs": [],
       "source": [
         "# to_remove solution\n",
-        "\n",
         "def get_coarse_corr(n_groups, X):\n",
         "    \"\"\"\n",
         "    A wrapper function for our correlation calculations between coarsely sampled\n",
@@ -1544,7 +1543,7 @@
         "id": "3ND-mbdo27rZ"
       },
       "source": [
-        "We know that for large systems correlation $\\neq$ causation. Here, we have looked at what happens when we coarsely sample a large system. Do we get better at estimating the *effective* causal interaction between groups (=average of weights) from the correlation between the groups?\n",
+        "We know that for large systems **correlation $\\neq$ causation**. Here, we have looked at what happens when we coarsely sample a large system. Do we get better at estimating the *effective* causal interaction between groups (i.e., average of weights) from the correlation between the groups?\n",
         "\n",
         "From our simulation above, the answer appears to be no: as the number of neurons per group increases, we don't see any significant increase in our ability to estimate the causal interaction between groups."
       ]