W2D1 Post-Course Update (Superset)

tutorials/W2D1_Macrocircuits/W2D1_Tutorial1.ipynb

@@ -580,7 +580,7 @@
     "\n",
     "    return nn.Sequential(*layers)\n",
     "\n",
-    "net = make_MLP(10, 3, 2)"
+    "net = make_MLP(n_in = 10, W = 3, D = 2)"
    ]
   },
   {
@@ -710,7 +710,7 @@
     "        nn.init.normal_(param, std = ...)\n",
     "\n",
     "initialize_layers(net, 1)\n",
-    "np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.6093441247940063, err_msg = \"Expected value of parameter is different!\")"
+    "np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.609, err_msg = \"Expected value of parameter is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -737,7 +737,7 @@
     "        nn.init.normal_(param, std = sigma/np.sqrt(n_in))\n",
     "\n",
     "initialize_layers(net, 1)\n",
-    "np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.6093441247940063, err_msg = \"Expected value of parameter is different!\")"
+    "np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.609, err_msg = \"Expected value of parameter is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -803,7 +803,7 @@
     "    return X, ...\n",
     "\n",
     "X, y = make_data(net, 10, 10000000)\n",
-    "np.testing.assert_allclose(X[0][0].item(), 1.9269152879714966, err_msg = \"Expected value of data is different!\")"
+    "np.testing.assert_allclose(X[0][0].item(), 1.927, err_msg = \"Expected value of data is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -835,7 +835,7 @@
     "    return X, y\n",
     "\n",
     "X, y = make_data(net, 10, 10000000)\n",
-    "np.testing.assert_allclose(X[0][0].item(), 1.9269152879714966, err_msg = \"Expected value of data is different!\")"
+    "np.testing.assert_allclose(X[0][0].item(), 1.927, err_msg = \"Expected value of data is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -916,7 +916,7 @@
     "    return Es\n",
     "\n",
     "Es = train_model(net, X, y, 10, 1e-3)\n",
-    "np.testing.assert_allclose(Es[0], 0.0, err_msg = \"Expected value of loss is different!\")"
+    "np.testing.assert_allclose(Es[0], 0.0, err_msg = \"Expected value of loss is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -964,7 +964,7 @@
     "    return Es\n",
     "\n",
     "Es = train_model(net, X, y, 10, 1e-3)\n",
-    "np.testing.assert_allclose(Es[0], 0.0, err_msg = \"Expected value of loss is different!\")"
+    "np.testing.assert_allclose(Es[0], 0.0, err_msg = \"Expected value of loss is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -1037,7 +1037,7 @@
     "    return loss\n",
     "\n",
     "loss = compute_loss(net, X, y)\n",
-    "np.testing.assert_allclose(loss, 0.0, err_msg = \"Expected value of loss is different!\")"
+    "np.testing.assert_allclose(loss, 0.0, err_msg = \"Expected value of loss is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -1070,7 +1070,7 @@
     "    return loss\n",
     "\n",
     "loss = compute_loss(net, X, y)\n",
-    "np.testing.assert_allclose(loss, 0.0, err_msg = \"Expected value of loss is different!\")"
+    "np.testing.assert_allclose(loss, 0.0, err_msg = \"Expected value of loss is different!\", atol = 1e-3)"
    ]
   },
   {

tutorials/W2D1_Macrocircuits/W2D1_Tutorial2.ipynb

@@ -1197,7 +1197,7 @@
     "    Outputs:\n",
     "    - (float): mean error for train data.\n",
     "    \"\"\"\n",
-    "    return np.mean(...)\n",
+    "    return np.mean(np.array([fit_relu(..., ..., ..., ..., n_hidden=n_hidden, reg = reg)[1] for _ in range(n_reps)]))\n",
     "\n",
     "test_errs = [sweep_test(x_train, y_train, x_test, y_test, n_hidden=..., n_reps=100, reg = 0.0) for n_hid in ...]\n",
     "\n",
@@ -1238,7 +1238,7 @@
     "    Outputs:\n",
     "    - (float): mean error for train data.\n",
     "    \"\"\"\n",
-    "    return np.mean(np.array([fit_relu(x_train, y_train, x_test, y_test, n_hidden=n_hidden, reg = reg)[1] for i in range(n_reps)]))\n",
+    "    return np.mean(np.array([fit_relu(x_train, y_train, x_test, y_test, n_hidden=n_hidden, reg = reg)[1] for _ in range(n_reps)]))\n",
     "\n",
     "test_errs = [sweep_test(x_train, y_train, x_test, y_test, n_hidden=n_hid, n_reps=100, reg = 0.0) for n_hid in n_hids]\n",
     "\n",

tutorials/W2D1_Macrocircuits/W2D1_Tutorial3.ipynb

@@ -2312,6 +2312,23 @@
     "1. Is there any difference between the trajectories for the modular and holistic agents? If so, what does it imply?"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "execution": {}
+   },
+   "outputs": [],
+   "source": [
+    "#to_remove explanation\n",
+    "\n",
+    "\"\"\"\n",
+    "Discussion: Is there any difference between the trajectories for the modular and holistic agents? If so, what does it imply?\n",
+    "\n",
+    "The holistic agent's trajectory has a higher curvature and length than that of the modular agent, suggesting that the modular agent's trajectory is more optimal. This is because, based on the RL objective with a discount factor smaller than 1, the trajectory should be as efficient (involving fewer steps) as possible.\n",
+    "\"\"\""
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,

tutorials/W2D1_Macrocircuits/instructor/W2D1_Tutorial1.ipynb

@@ -580,7 +580,7 @@
     "\n",
     "    return nn.Sequential(*layers)\n",
     "\n",
-    "net = make_MLP(10, 3, 2)"
+    "net = make_MLP(n_in = 10, W = 3, D = 2)"
    ]
   },
   {
@@ -712,7 +712,7 @@
     "        nn.init.normal_(param, std = ...)\n",
     "\n",
     "initialize_layers(net, 1)\n",
-    "np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.6093441247940063, err_msg = \"Expected value of parameter is different!\")\n",
+    "np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.609, err_msg = \"Expected value of parameter is different!\", atol = 1e-3)\n",
     "\n",
     "```"
    ]
@@ -741,7 +741,7 @@
     "        nn.init.normal_(param, std = sigma/np.sqrt(n_in))\n",
     "\n",
     "initialize_layers(net, 1)\n",
-    "np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.6093441247940063, err_msg = \"Expected value of parameter is different!\")"
+    "np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.609, err_msg = \"Expected value of parameter is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -807,7 +807,7 @@
     "    return X, ...\n",
     "\n",
     "X, y = make_data(net, 10, 10000000)\n",
-    "np.testing.assert_allclose(X[0][0].item(), 1.9269152879714966, err_msg = \"Expected value of data is different!\")\n",
+    "np.testing.assert_allclose(X[0][0].item(), 1.927, err_msg = \"Expected value of data is different!\", atol = 1e-3)\n",
     "\n",
     "```"
    ]
@@ -841,7 +841,7 @@
     "    return X, y\n",
     "\n",
     "X, y = make_data(net, 10, 10000000)\n",
-    "np.testing.assert_allclose(X[0][0].item(), 1.9269152879714966, err_msg = \"Expected value of data is different!\")"
+    "np.testing.assert_allclose(X[0][0].item(), 1.927, err_msg = \"Expected value of data is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -922,7 +922,7 @@
     "    return Es\n",
     "\n",
     "Es = train_model(net, X, y, 10, 1e-3)\n",
-    "np.testing.assert_allclose(Es[0], 0.0, err_msg = \"Expected value of loss is different!\")\n",
+    "np.testing.assert_allclose(Es[0], 0.0, err_msg = \"Expected value of loss is different!\", atol = 1e-3)\n",
     "\n",
     "```"
    ]
@@ -972,7 +972,7 @@
     "    return Es\n",
     "\n",
     "Es = train_model(net, X, y, 10, 1e-3)\n",
-    "np.testing.assert_allclose(Es[0], 0.0, err_msg = \"Expected value of loss is different!\")"
+    "np.testing.assert_allclose(Es[0], 0.0, err_msg = \"Expected value of loss is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -1045,7 +1045,7 @@
     "    return loss\n",
     "\n",
     "loss = compute_loss(net, X, y)\n",
-    "np.testing.assert_allclose(loss, 0.0, err_msg = \"Expected value of loss is different!\")\n",
+    "np.testing.assert_allclose(loss, 0.0, err_msg = \"Expected value of loss is different!\", atol = 1e-3)\n",
     "\n",
     "```"
    ]
@@ -1080,7 +1080,7 @@
     "    return loss\n",
     "\n",
     "loss = compute_loss(net, X, y)\n",
-    "np.testing.assert_allclose(loss, 0.0, err_msg = \"Expected value of loss is different!\")"
+    "np.testing.assert_allclose(loss, 0.0, err_msg = \"Expected value of loss is different!\", atol = 1e-3)"
    ]
   },
   {

tutorials/W2D1_Macrocircuits/instructor/W2D1_Tutorial2.ipynb

@@ -1201,7 +1201,7 @@
     "    Outputs:\n",
     "    - (float): mean error for train data.\n",
     "    \"\"\"\n",
-    "    return np.mean(...)\n",
+    "    return np.mean(np.array([fit_relu(..., ..., ..., ..., n_hidden=n_hidden, reg = reg)[1] for _ in range(n_reps)]))\n",
     "\n",
     "test_errs = [sweep_test(x_train, y_train, x_test, y_test, n_hidden=..., n_reps=100, reg = 0.0) for n_hid in ...]\n",
     "\n",
@@ -1244,7 +1244,7 @@
     "    Outputs:\n",
     "    - (float): mean error for train data.\n",
     "    \"\"\"\n",
-    "    return np.mean(np.array([fit_relu(x_train, y_train, x_test, y_test, n_hidden=n_hidden, reg = reg)[1] for i in range(n_reps)]))\n",
+    "    return np.mean(np.array([fit_relu(x_train, y_train, x_test, y_test, n_hidden=n_hidden, reg = reg)[1] for _ in range(n_reps)]))\n",
     "\n",
     "test_errs = [sweep_test(x_train, y_train, x_test, y_test, n_hidden=n_hid, n_reps=100, reg = 0.0) for n_hid in n_hids]\n",
     "\n",

tutorials/W2D1_Macrocircuits/instructor/W2D1_Tutorial3.ipynb

@@ -2322,6 +2322,23 @@
     "1. Is there any difference between the trajectories for the modular and holistic agents? If so, what does it imply?"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "execution": {}
+   },
+   "outputs": [],
+   "source": [
+    "#to_remove explanation\n",
+    "\n",
+    "\"\"\"\n",
+    "Discussion: Is there any difference between the trajectories for the modular and holistic agents? If so, what does it imply?\n",
+    "\n",
+    "The holistic agent's trajectory has a higher curvature and length than that of the modular agent, suggesting that the modular agent's trajectory is more optimal. This is because, based on the RL objective with a discount factor smaller than 1, the trajectory should be as efficient (involving fewer steps) as possible.\n",
+    "\"\"\""
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,

tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_d81fba1b.py → tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_6b3d3e34.py

@@ -18,4 +18,4 @@ def make_data(net, n_in, n_examples):
     return X, y
 
 X, y = make_data(net, 10, 10000000)
-np.testing.assert_allclose(X[0][0].item(), 1.9269152879714966, err_msg = "Expected value of data is different!")
+np.testing.assert_allclose(X[0][0].item(), 1.927, err_msg = "Expected value of data is different!", atol = 1e-3)

tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_4929cd6d.py → tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_6fa930a8.py

@@ -19,4 +19,4 @@ def compute_loss(net, X, y):
     return loss
 
 loss = compute_loss(net, X, y)
-np.testing.assert_allclose(loss, 0.0, err_msg = "Expected value of loss is different!")
+np.testing.assert_allclose(loss, 0.0, err_msg = "Expected value of loss is different!", atol = 1e-3)

tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_a2d6d2f1.py → tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_8c945e68.py

@@ -13,4 +13,4 @@ def initialize_layers(net,sigma):
         nn.init.normal_(param, std = sigma/np.sqrt(n_in))
 
 initialize_layers(net, 1)
-np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.6093441247940063, err_msg = "Expected value of parameter is different!")
+np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.609, err_msg = "Expected value of parameter is different!", atol = 1e-3)

tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_29135326.py → tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_c3274cd4.py

@@ -34,4 +34,4 @@ def train_model(net, X, y, n_epochs, lr, progressbar=True):
     return Es
 
 Es = train_model(net, X, y, 10, 1e-3)
-np.testing.assert_allclose(Es[0], 0.0, err_msg = "Expected value of loss is different!")
+np.testing.assert_allclose(Es[0], 0.0, err_msg = "Expected value of loss is different!", atol = 1e-3)

tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial2_Solution_e6df4a06.py → tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial2_Solution_7717ab4f.py

@@ -18,7 +18,7 @@ def sweep_test(x_train, y_train, x_test, y_test, n_hidden = 10, n_reps = 100, re
     Outputs:
     - (float): mean error for train data.
     """
-    return np.mean(np.array([fit_relu(x_train, y_train, x_test, y_test, n_hidden=n_hidden, reg = reg)[1] for i in range(n_reps)]))
+    return np.mean(np.array([fit_relu(x_train, y_train, x_test, y_test, n_hidden=n_hidden, reg = reg)[1] for _ in range(n_reps)]))
 
 test_errs = [sweep_test(x_train, y_train, x_test, y_test, n_hidden=n_hid, n_reps=100, reg = 0.0) for n_hid in n_hids]
 

tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial3_Solution_02426ac5.py

@@ -0,0 +1,6 @@
+
+"""
+Discussion: Is there any difference between the trajectories for the modular and holistic agents? If so, what does it imply?
+
+The holistic agent's trajectory has a higher curvature and length than that of the modular agent, suggesting that the modular agent's trajectory is more optimal. This is because, based on the RL objective with a discount factor smaller than 1, the trajectory should be as efficient (involving fewer steps) as possible.
+"""

tutorials/W2D1_Macrocircuits/student/W2D1_Tutorial1.ipynb

@@ -580,7 +580,7 @@
     "\n",
     "    return nn.Sequential(*layers)\n",
     "\n",
-    "net = make_MLP(10, 3, 2)"
+    "net = make_MLP(n_in = 10, W = 3, D = 2)"
    ]
   },
   {
@@ -691,7 +691,7 @@
     "        nn.init.normal_(param, std = ...)\n",
     "\n",
     "initialize_layers(net, 1)\n",
-    "np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.6093441247940063, err_msg = \"Expected value of parameter is different!\")"
+    "np.testing.assert_allclose(next(net.parameters())[0][0].item(), 0.609, err_msg = \"Expected value of parameter is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -701,7 +701,7 @@
     "execution": {}
    },
    "source": [
-    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_a2d6d2f1.py)\n",
+    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_8c945e68.py)\n",
     "\n"
    ]
   },
@@ -768,7 +768,7 @@
     "    return X, ...\n",
     "\n",
     "X, y = make_data(net, 10, 10000000)\n",
-    "np.testing.assert_allclose(X[0][0].item(), 1.9269152879714966, err_msg = \"Expected value of data is different!\")"
+    "np.testing.assert_allclose(X[0][0].item(), 1.927, err_msg = \"Expected value of data is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -778,7 +778,7 @@
     "execution": {}
    },
    "source": [
-    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_d81fba1b.py)\n",
+    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_6b3d3e34.py)\n",
     "\n"
    ]
   },
@@ -860,7 +860,7 @@
     "    return Es\n",
     "\n",
     "Es = train_model(net, X, y, 10, 1e-3)\n",
-    "np.testing.assert_allclose(Es[0], 0.0, err_msg = \"Expected value of loss is different!\")"
+    "np.testing.assert_allclose(Es[0], 0.0, err_msg = \"Expected value of loss is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -870,7 +870,7 @@
     "execution": {}
    },
    "source": [
-    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_29135326.py)\n",
+    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_c3274cd4.py)\n",
     "\n"
    ]
   },
@@ -937,7 +937,7 @@
     "    return loss\n",
     "\n",
     "loss = compute_loss(net, X, y)\n",
-    "np.testing.assert_allclose(loss, 0.0, err_msg = \"Expected value of loss is different!\")"
+    "np.testing.assert_allclose(loss, 0.0, err_msg = \"Expected value of loss is different!\", atol = 1e-3)"
    ]
   },
   {
@@ -947,7 +947,7 @@
     "execution": {}
    },
    "source": [
-    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_4929cd6d.py)\n",
+    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial1_Solution_6fa930a8.py)\n",
     "\n"
    ]
   },

tutorials/W2D1_Macrocircuits/student/W2D1_Tutorial2.ipynb

@@ -1112,7 +1112,7 @@
     "    Outputs:\n",
     "    - (float): mean error for train data.\n",
     "    \"\"\"\n",
-    "    return np.mean(...)\n",
+    "    return np.mean(np.array([fit_relu(..., ..., ..., ..., n_hidden=n_hidden, reg = reg)[1] for _ in range(n_reps)]))\n",
     "\n",
     "test_errs = [sweep_test(x_train, y_train, x_test, y_test, n_hidden=..., n_reps=100, reg = 0.0) for n_hid in ...]\n",
     "\n",
@@ -1131,11 +1131,11 @@
     "execution": {}
    },
    "source": [
-    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial2_Solution_e6df4a06.py)\n",
+    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W2D1_Macrocircuits/solutions/W2D1_Tutorial2_Solution_7717ab4f.py)\n",
     "\n",
     "*Example output:*\n",
     "\n",
-    "<img alt='Solution hint' align='left' width=778.0 height=577.0 src=https://raw.githubusercontent.com/neuromatch/NeuroAI_Course/main/tutorials/W2D1_Macrocircuits/static/W2D1_Tutorial2_Solution_e6df4a06_0.png>\n",
+    "<img alt='Solution hint' align='left' width=778.0 height=577.0 src=https://raw.githubusercontent.com/neuromatch/NeuroAI_Course/main/tutorials/W2D1_Macrocircuits/static/W2D1_Tutorial2_Solution_7717ab4f_0.png>\n",
     "\n"
    ]
   },