Merge pull request #51 from ChitambarLab/behavior_fn-bug

Behavior fn bug and support for Adam

Author: bdoolittle

src/qnetvo/gradient_descent.py

@@ -12,6 +12,8 @@ def gradient_descent(
     grad_fn=None,
     verbose=True,
     interface="autograd",
+    optimizer=None,
+    optimizer_kwargs={},
 ):
     """Performs a numerical gradient descent optimization on the provided ``cost`` function.
     The optimization is seeded with (random) ``init_settings`` which are then varied to
@@ -42,6 +44,13 @@ def gradient_descent(
     :param interface: Specifies the optimizer software either ``"autograd"`` or ``"tf"`` (TensorFlow).
     :type interface: string, default ``"autograd``"
 
+    :param optimizer: Specifies the PennyLane optimizer to use. Default ``qml.GradientDescentOptimizer``.
+                      Set to ``"adam"`` to use the ``qml.AdamOptimizer``, note that ``interface="autograd"`` must be set.
+    :type optimizer: String
+
+    :param optimizer_kwargs: Keyword arguments to pass to the specified optimizer.
+    :type optimizer_kwargs: Dict
+
     :return: Data regarding the gradient descent optimization.
     :rtype: dictionary, contains the following keys:
 
@@ -69,7 +78,10 @@ def gradient_descent(
     """
 
     if interface == "autograd":
-        opt = qml.GradientDescentOptimizer(stepsize=step_size)
+        if optimizer == "adam":
+            opt = qml.AdamOptimizer(stepsize=step_size, **optimizer_kwargs)
+        else:
+            opt = qml.GradientDescentOptimizer(stepsize=step_size)
     elif interface == "tf":
         from .lazy_tensorflow_import import tensorflow as tf
 

src/qnetvo/information.py

@@ -1,6 +1,6 @@
 from pennylane import math
 from .qnodes import joint_probs_qnode
-from .utilities import mixed_base_num
+from .utilities import mixed_base_num, ragged_reshape
 from pennylane import numpy as np
 
 
@@ -43,11 +43,22 @@ def behavior_fn(network_ansatz, postmap=np.array([]), qnode_kwargs={}):
               behavior matrix for a given set of settings.
     :rtype: function
     """
-    num_in_prep_nodes = [node.num_in for node in network_ansatz.layers[0]]
-    num_in_meas_nodes = [node.num_in for node in network_ansatz.layers[-1]]
+    # num_in_prep_nodes = [node.num_in for node in network_ansatz.layers[0]]
+    # num_in_meas_nodes = [node.num_in for node in network_ansatz.layers[-1]]
 
-    base_digits = num_in_prep_nodes + num_in_meas_nodes
-    net_num_in = math.prod(base_digits)
+    # base_digits = num_in_prep_nodes + num_in_meas_nodes
+    # net_num_in = math.prod(base_digits)
+
+    # raw_net_num_out = 2 ** len(network_ansatz.layers_wires[-1])
+
+    probs_qnode = joint_probs_qnode(network_ansatz, **qnode_kwargs)
+
+    net_num_in = math.prod(network_ansatz.layers_total_num_in)
+    num_inputs_list = math.concatenate(network_ansatz.layers_node_num_in).tolist()
+    node_input_ids = [
+        ragged_reshape(mixed_base_num(i, num_inputs_list), network_ansatz.layers_num_nodes)
+        for i in range(net_num_in)
+    ]
 
     raw_net_num_out = 2 ** len(network_ansatz.layers_wires[-1])
 
@@ -56,18 +67,14 @@ def behavior_fn(network_ansatz, postmap=np.array([]), qnode_kwargs={}):
         if postmap.shape[1] != raw_net_num_out:
             raise ValueError("The `postmap` must have " + str(raw_net_num_out) + " columns.")
 
-    node_input_ids = [mixed_base_num(i, base_digits) for i in range(net_num_in)]
+    # node_input_ids = [mixed_base_num(i, base_digits) for i in range(net_num_in)]
 
-    probs_qnode = joint_probs_qnode(network_ansatz, **qnode_kwargs)
+    # probs_qnode = joint_probs_qnode(network_ansatz, **qnode_kwargs)
 
     def behavior(network_settings):
         raw_behavior = np.zeros((raw_net_num_out, net_num_in))
         for i, input_id_set in enumerate(node_input_ids):
-            settings = network_ansatz.qnode_settings(
-                network_settings,
-                [input_id_set[0 : len(num_in_prep_nodes)], input_id_set[len(num_in_prep_nodes) :]],
-            )
-
+            settings = network_ansatz.qnode_settings(network_settings, input_id_set)
             raw_behavior[:, i] += probs_qnode(settings)
 
         return postmap @ raw_behavior if has_postmap else raw_behavior

test/gradient_descent_test.py

@@ -44,3 +44,19 @@ def test_quadratic_cost(self):
                 verbose=False,
                 interface="jax",
             )
+
+        # adam optimizer
+        ad_opt_dict = qnet.gradient_descent(
+            cost,
+            settings,
+            num_steps=50,
+            step_size=0.1,
+            verbose=True,
+            optimizer="adam",
+        )
+
+        assert np.isclose(opt_dict["opt_score"], 0, atol=1e-6)
+        assert np.isclose(opt_dict["opt_settings"], 0, atol=1e-4)
+        assert opt_dict["samples"] == [0, 25, 50]
+        assert len(opt_dict["scores"]) == 3
+        assert len(opt_dict["settings_history"]) == 51

test/information_test.py

@@ -8,12 +8,16 @@
 class TestBehaviorFn:
     def test_simple_settings(self):
         prep_nodes = [
-            qnet.PrepareNode(2, [0], qnet.local_RY, 1),
-            qnet.PrepareNode(2, [1], qnet.local_RY, 1),
+            qnet.PrepareNode(num_in=2, wires=[0], ansatz_fn=qnet.local_RY, num_settings=1),
+            qnet.PrepareNode(num_in=2, wires=[1], ansatz_fn=qnet.local_RY, num_settings=1),
         ]
         meas_nodes = [
-            qnet.MeasureNode(2, 2, [0], qnet.local_RY, 1),
-            qnet.MeasureNode(2, 2, [1], qnet.local_RY, 1),
+            qnet.MeasureNode(
+                num_in=2, num_out=2, wires=[0], ansatz_fn=qnet.local_RY, num_settings=1
+            ),
+            qnet.MeasureNode(
+                num_in=2, num_out=2, wires=[1], ansatz_fn=qnet.local_RY, num_settings=1
+            ),
         ]
         ansatz = qnet.NetworkAnsatz(prep_nodes, meas_nodes)
         P_Net = qnet.behavior_fn(ansatz)
@@ -122,6 +126,61 @@ def test_rand_settings(self):
         assert P_Net.shape == (16, 288)
         assert np.allclose(np.ones(288), [np.sum(P_Net[:, i]) for i in range(288)])
 
+    def test_inputs_from_multiple_layers(self):
+        prep_nodes_a = [
+            qnet.PrepareNode(num_in=2, wires=[0], ansatz_fn=qnet.local_RY, num_settings=1),
+        ]
+        prep_nodes_b = [
+            qnet.PrepareNode(num_in=2, wires=[1], ansatz_fn=qnet.local_RY, num_settings=1),
+        ]
+
+        meas_nodes = [
+            qnet.MeasureNode(
+                num_in=2, num_out=2, wires=[0], ansatz_fn=qnet.local_RY, num_settings=1
+            ),
+            qnet.MeasureNode(
+                num_in=2, num_out=2, wires=[1], ansatz_fn=qnet.local_RY, num_settings=1
+            ),
+        ]
+        ansatz = qnet.NetworkAnsatz(prep_nodes_a, prep_nodes_b, meas_nodes)
+        P_Net = qnet.behavior_fn(ansatz)
+        zero_settings = ansatz.zero_network_settings()
+
+        assert np.all(
+            P_Net(zero_settings)
+            == [
+                [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
+                [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
+                [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
+                [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
+            ]
+        )
+
+        settings = zero_settings
+        settings[1] = np.pi
+
+        assert np.allclose(
+            P_Net(settings),
+            [
+                [1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0],
+                [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
+                [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1],
+                [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
+            ],
+        )
+
+        settings[7] = np.pi / 2
+
+        assert np.allclose(
+            P_Net(settings),
+            [
+                [1, 0.5, 1, 0.5, 1, 0.5, 1, 0.5, 0, 0, 0, 0, 0, 0, 0, 0],
+                [0, 0.5, 0, 0.5, 0, 0.5, 0, 0.5, 0, 0, 0, 0, 0, 0, 0, 0],
+                [0, 0, 0, 0, 0, 0, 0, 0, 1, 0.5, 1, 0.5, 1, 0.5, 1, 0.5],
+                [0, 0, 0, 0, 0, 0, 0, 0, 0, 0.5, 0, 0.5, 0, 0.5, 0, 0.5],
+            ],
+        )
+
 
 class TestShannonEntropy:
     @pytest.mark.parametrize(