[Stable] Prepare 0.2.1 release (#193)

* Update default batch_size in quantum_kernel (#150) The max circuits per job for most/all backend devices is 900. Setting batch_size=900 fixes an inefficient jobs being created. Co-authored-by: Anton Dekusar <adekusar@ie.ibm.com> Co-authored-by: Anton Dekusar <62334182+adekusar-drl@users.noreply.github.com> * This adds the callback feature for NN Classifier and Regressor (#112) (#151) * Added a wrapper `get_objective` method in `TrainableModel`, and `callback` argument to NN classifier `get_objective` returns a callable which is passed as objective function to NN classifier. It checks for a callable `callback` argument, which if not `None` can access the intermediate data during the optimization. * style changes * Updated callback tests for `NeuralNetworkClassifier` * Added callback for `NeuralNetworkRegressor` * Added tests for callback * Refactored `get_objective` * Added release note * Updated `callback` argument docstring * Shifted `callback` parameter to parent class `TrainableModel` * Added assertions for number of weights in `TestNeuralNetworkClassifier` `TestNeuralNetworkRegressor` * Rephrased release note and debugged tests for callback Co-authored-by: Anton Dekusar <62334182+adekusar-drl@users.noreply.github.com> Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com> Co-authored-by: Manoel Marques <Manoel.Marques@ibm.com> * Added ability to automatically handle categorical labels (#62) (#143) Added categorical label encoding * Added documentation * Update qiskit_machine_learning/algorithms/classifiers/neural_network_classifier.py * Update releasenotes/notes/cateforical-output-049e682adc9d1e28.yaml * Update releasenotes/notes/cateforical-output-049e682adc9d1e28.yaml * Update releasenotes/notes/cateforical-output-049e682adc9d1e28.yaml * Cleaned and organized code * Update qiskit_machine_learning/algorithms/classifiers/neural_network_classifier.py * Update qiskit_machine_learning/algorithms/classifiers/neural_network_classifier.py * Update qiskit_machine_learning/algorithms/classifiers/neural_network_classifier.py * Split categorical encoding into two functions for classifier's fit and score methods * Set sparse=False in one-hot encoder * removed unnecessary copy of arrays * Renamed methods and fixed label reshape * Added explanatory comments Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com> Co-authored-by: Manoel Marques <Manoel.Marques@ibm.com> Co-authored-by: Anton Dekusar <adekusar@ie.ibm.com> Co-authored-by: Anton Dekusar <62334182+adekusar-drl@users.noreply.github.com> * Add hybrid qnn unit tests for TorchConnector (#156) * Add hybrid qnn unit tests * Fix spelling * Add docstrings * Retry * Fix style * Register weight name * Update test * Separate hybrid tests * Fix spelling * Fix lint * Add comments * Remove unused functions * Add register weight param. reno * Clarify procedure * Remove unused import Co-authored-by: Anton Dekusar <62334182+adekusar-drl@users.noreply.github.com> Co-authored-by: Manoel Marques <Manoel.Marques@ibm.com> * Fixed variable name in file 03_quantum_kernel.ipynb . (#179) Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com> * Added getter/setter, exception handling for OpflowQNN, CircuitQNN (#79) * added getter and setter, raises exception if no qi * set sampler in qi setter, added tests * added logic to setters, fixed tests * fix black * calling setter with instantiation * setters no longer accept optional qi * undid changes to docstrings * reset output shape within qi setter * revised implementation * fix spell * fix mypy * Update qiskit_machine_learning/neural_networks/circuit_qnn.py Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com> * Update qiskit_machine_learning/neural_networks/circuit_qnn.py Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com> * Update qiskit_machine_learning/neural_networks/opflow_qnn.py Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com> * code review * updated docstrings * Update qiskit_machine_learning/neural_networks/circuit_qnn.py Co-authored-by: dlasecki <dal@zurich.ibm.com> * Update qiskit_machine_learning/neural_networks/circuit_qnn.py Co-authored-by: dlasecki <dal@zurich.ibm.com> * code review * docstrings Co-authored-by: Anton Dekusar <adekusar@ie.ibm.com> Co-authored-by: Anton Dekusar <62334182+adekusar-drl@users.noreply.github.com> Co-authored-by: Manoel Marques <Manoel.Marques@ibm.com> Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com> Co-authored-by: dlasecki <dal@zurich.ibm.com> * Global random seed (#178) * Made TrainableModel use global seed (issue #158) Made TrainableModel use RandomState object seeded by algorithm_globals.random_seed to choose an initial point for the optimizer outside of warm start. * Global random seed for QSVC constructor (issue #158) The value of algorithm_globals.random_seed is being used as random_state parameter for SVC constructor during construction of QSVC object. * Added Qiskit into requirements.txt (issu #158) Added Qiskit~=0.26.0 into requirements.txt, as qiskit.utils contain algorithm_globals object. * Added Qiskit into requirements.txt & improved code formatting (issue #158) Added Qiskit~=0.26.0 into requirements.txt, as qiskit.utils contain algorithm_globals object. * Removed unnecessary explicit requirement of Qiskit package (issue #158). * Simplified random number generation using global seed (issue #158). * random_state is now being randomly initialized only if not passed as parameter (issue #158). * Parameter 'random_state' now initialized by 'random_seed' in kwargs (issue #158). * Update qiskit_machine_learning/algorithms/trainable_model.py * Reverted incorrect changes to requirements.txt (issue #158). Co-authored-by: Anton Dekusar <62334182+adekusar-drl@users.noreply.github.com> Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com> * Fix `RawFeatureVector` copy (#182) * Fix num_qubits * Remove _num_qubits * Fix style * Add tests for copy, bind_parameters * Add Fix RawFeatVec reno * Fix typing * Fix test * Add reset registers * Fix num_qubits>0 * Fix mypy conflict * Fix black * Update if check Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com> * Add if check * Add comment * Modify reno * code review * code review * updated docstrings * code review Co-authored-by: Anton Dekusar <62334182+adekusar-drl@users.noreply.github.com> Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com> Co-authored-by: Anton Dekusar <adekusar@ie.ibm.com> * Prepare 0.2.1 release Co-authored-by: Anna Phan <9410731+attp@users.noreply.github.com> Co-authored-by: Anton Dekusar <adekusar@ie.ibm.com> Co-authored-by: Anton Dekusar <62334182+adekusar-drl@users.noreply.github.com> Co-authored-by: Darsh Kaushik <70013230+darshkaushik@users.noreply.github.com> Co-authored-by: Steve Wood <40241007+woodsp-ibm@users.noreply.github.com> Co-authored-by: Matt Wright <44040188+mattwright99@users.noreply.github.com> Co-authored-by: ElePT <57907331+ElePT@users.noreply.github.com> Co-authored-by: Martin Beseda <martinbeseda@seznam.cz> Co-authored-by: Stephen Murray <stephenandrewmurray@gmail.com> Co-authored-by: dlasecki <dal@zurich.ibm.com> Co-authored-by: Martin Beseda <martin.beseda@vsb.cz>
qiskit-community · Aug 24, 2021 · ec3cbcc · ec3cbcc
1 parent 03cbe77
commit ec3cbcc
Show file tree

Hide file tree

Showing 26 changed files with 1,082 additions and 232 deletions.
diff --git a/.pylintdict b/.pylintdict
@@ -162,6 +162,7 @@ sy
 tensored
 th
 toctree
+todo
 traceback
 transpilation
 uncompiled

diff --git a/docs/tutorials/03_quantum_kernel.ipynb b/docs/tutorials/03_quantum_kernel.ipynb
@@ -232,7 +232,7 @@
     "qsvc.fit(train_features, train_labels)\n",
     "qsvc_score = qsvc.score(test_features, test_labels)\n",
     "\n",
-    "print(f'QSVC classification test score: {adhoc_score}')"
+    "print(f'QSVC classification test score: {qsvc_score}')"
    ]
   },
   {

diff --git a/qiskit_machine_learning/VERSION.txt b/qiskit_machine_learning/VERSION.txt
@@ -1 +1 @@
-0.2.0
+0.2.1
diff --git a/qiskit_machine_learning/algorithms/classifiers/neural_network_classifier.py b/qiskit_machine_learning/algorithms/classifiers/neural_network_classifier.py
@@ -11,11 +11,12 @@
 # that they have been altered from the originals.
 """An implementation of quantum neural network classifier."""
 
-from typing import Union, Optional
+from typing import Union, Optional, Callable
 
 import numpy as np
 from qiskit.algorithms.optimizers import Optimizer
 from sklearn.base import ClassifierMixin
+from sklearn.preprocessing import OneHotEncoder, LabelEncoder
 
 from ..objective_functions import (
     BinaryObjectiveFunction,
@@ -43,6 +44,7 @@ def __init__(
         optimizer: Optional[Optimizer] = None,
         warm_start: bool = False,
         initial_point: np.ndarray = None,
+        callback: Optional[Callable[[np.ndarray, float], None]] = None,
     ):
         """
         Args:
@@ -65,18 +67,27 @@ def __init__(
                 one-hot-encoded sample (e.g. for 'CrossEntropy' loss function), and if False
                 as a set of individual predictions with occurrence probabilities (the index would be
                 the prediction and the value the corresponding frequency, e.g. for absolute/squared
-                loss). This option is ignored in case of a one-dimensional output.
+                loss). In case of a one-dimensional categorical output, this option determines how
+                to encode the target data (i.e. one-hot or integer encoding).
             optimizer: An instance of an optimizer to be used in training. When `None` defaults to SLSQP.
             warm_start: Use weights from previous fit to start next fit.
             initial_point: Initial point for the optimizer to start from.
-
+            callback: a reference to a user's callback function that has two parameters and
+                returns ``None``. The callback can access intermediate data during training.
+                On each iteration an optimizer invokes the callback and passes current weights
+                as an array and a computed value as a float of the objective function being
+                optimized. This allows to track how well optimization / training process is going on.
         Raises:
             QiskitMachineLearningError: unknown loss, invalid neural network
         """
-        super().__init__(neural_network, loss, optimizer, warm_start, initial_point)
+        super().__init__(neural_network, loss, optimizer, warm_start, initial_point, callback)
         self._one_hot = one_hot
+        # encodes the target data if categorical
+        self._target_encoder = OneHotEncoder(sparse=False) if one_hot else LabelEncoder()
 
     def fit(self, X: np.ndarray, y: np.ndarray):  # pylint: disable=invalid-name
+        y = self._fit_and_encode_labels(y)
+
         # mypy definition
         function: ObjectiveFunction = None
         if self._neural_network.output_shape == (1,):
@@ -91,9 +102,11 @@ def fit(self, X: np.ndarray, y: np.ndarray):  # pylint: disable=invalid-name
             else:
                 function = MultiClassObjectiveFunction(X, y, self._neural_network, self._loss)
 
+        objective = self._get_objective(function)
+
         self._fit_result = self._optimizer.optimize(
             self._neural_network.num_weights,
-            function.objective,
+            objective,
             function.gradient,
             initial_point=self._choose_initial_point(),
         )
@@ -119,4 +132,32 @@ def predict(self, X: np.ndarray) -> np.ndarray:  # pylint: disable=invalid-name
     def score(
         self, X: np.ndarray, y: np.ndarray, sample_weight: Optional[np.ndarray] = None
     ) -> float:
+        y = self._encode_labels(y)
         return ClassifierMixin.score(self, X, y, sample_weight)
+
+    def _fit_and_encode_labels(self, y: np.ndarray) -> np.ndarray:
+        """Fits label or one-hot encoder and converts categorical target data."""
+
+        if isinstance(y[0], str):
+            # string data is assumed to be categorical
+
+            # OneHotEncoder expects data with shape (n_samples, n_features) but
+            # LabelEncoder expects shape (n_samples,) so set desired shape
+            y = y.reshape(-1, 1) if self._one_hot else y
+            self._target_encoder.fit(y)
+            y = self._target_encoder.transform(y)
+
+        return y
+
+    def _encode_labels(self, y: np.ndarray) -> np.ndarray:
+        """Converts categorical target data using label or one-hot encoding."""
+
+        if isinstance(y[0], str):
+            # string data is assumed to be categorical
+
+            # OneHotEncoder expects data with shape (n_samples, n_features) but
+            # LabelEncoder expects shape (n_samples,) so set desired shape
+            y = y.reshape(-1, 1) if self._one_hot else y
+            y = self._target_encoder.transform(y)
+
+        return y
diff --git a/qiskit_machine_learning/algorithms/classifiers/qsvc.py b/qiskit_machine_learning/algorithms/classifiers/qsvc.py
@@ -16,6 +16,8 @@
 
 from sklearn.svm import SVC
 
+from qiskit.utils.algorithm_globals import algorithm_globals
+
 from qiskit_machine_learning.kernels.quantum_kernel import QuantumKernel
 
 
@@ -47,7 +49,14 @@ def __init__(self, *args, quantum_kernel: Optional[QuantumKernel] = None, **kwar
 
         self._quantum_kernel = quantum_kernel if quantum_kernel else QuantumKernel()
 
-        super().__init__(kernel=self._quantum_kernel.evaluate, *args, **kwargs)
+        if "random_state" not in kwargs:
+            kwargs["random_state"] = algorithm_globals.random_seed
+
+        super().__init__(
+            kernel=self._quantum_kernel.evaluate,
+            *args,
+            **kwargs,
+        )
 
     @property
     def quantum_kernel(self) -> QuantumKernel:

diff --git a/qiskit_machine_learning/algorithms/classifiers/vqc.py b/qiskit_machine_learning/algorithms/classifiers/vqc.py
@@ -11,7 +11,7 @@
 # that they have been altered from the originals.
 """An implementation of quantum neural network classifier."""
 
-from typing import Union, Optional, cast
+from typing import Union, Optional, Callable, cast
 import numpy as np
 
 from qiskit import QuantumCircuit
@@ -39,6 +39,7 @@ def __init__(
         warm_start: bool = False,
         quantum_instance: QuantumInstance = None,
         initial_point: np.ndarray = None,
+        callback: Optional[Callable[[np.ndarray, float], None]] = None,
     ) -> None:
         """
         Args:
@@ -50,6 +51,11 @@ def __init__(
             optimizer: An instance of an optimizer to be used in training. When `None` defaults to SLSQP.
             warm_start: Use weights from previous fit to start next fit.
             initial_point: Initial point for the optimizer to start from.
+            callback: a reference to a user's callback function that has two parameters and
+                returns ``None``. The callback can access intermediate data during training.
+                On each iteration an optimizer invokes the callback and passes current weights
+                as an array and a computed value as a float of the objective function being
+                optimized. This allows to track how well optimization / training process is going on.
         Raises:
             QiskitMachineLearningError: Needs at least one out of num_qubits, feature_map or
                 ansatz to be given.
@@ -119,6 +125,7 @@ def __init__(
             optimizer=optimizer,
             warm_start=warm_start,
             initial_point=initial_point,
+            callback=callback,
         )
 
     @property

diff --git a/qiskit_machine_learning/algorithms/regressors/neural_network_regressor.py b/qiskit_machine_learning/algorithms/regressors/neural_network_regressor.py
@@ -39,9 +39,11 @@ def fit(self, X: np.ndarray, y: np.ndarray):  # pylint: disable=invalid-name
         else:
             function = MultiClassObjectiveFunction(X, y, self._neural_network, self._loss)
 
+        objective = self._get_objective(function)
+
         self._fit_result = self._optimizer.optimize(
             self._neural_network.num_weights,
-            function.objective,
+            objective,
             function.gradient,
             initial_point=self._choose_initial_point(),
         )

diff --git a/qiskit_machine_learning/algorithms/regressors/vqr.py b/qiskit_machine_learning/algorithms/regressors/vqr.py
@@ -11,7 +11,7 @@
 # that they have been altered from the originals.
 """An implementation of quantum neural network regressor."""
 
-from typing import Union, Optional, cast
+from typing import Union, Optional, Callable, cast
 
 import numpy as np
 
@@ -39,6 +39,7 @@ def __init__(
         warm_start: bool = False,
         quantum_instance: QuantumInstance = None,
         initial_point: np.ndarray = None,
+        callback: Optional[Callable[[np.ndarray, float], None]] = None,
     ) -> None:
         r"""
         Args:
@@ -54,7 +55,11 @@ def __init__(
             optimizer: An instance of an optimizer to be used in training. When `None` defaults to SLSQP.
             warm_start: Use weights from previous fit to start next fit.
             initial_point: Initial point for the optimizer to start from.
-
+            callback: a reference to a user's callback function that has two parameters and
+                returns ``None``. The callback can access intermediate data during training.
+                On each iteration an optimizer invokes the callback and passes current weights
+                as an array and a computed value as a float of the objective function being
+                optimized. This allows to track how well optimization / training process is going on.
         Raises:
             QiskitMachineLearningError: Neither num_qubits, nor feature_map, nor ansatz given.
         """
@@ -75,6 +80,7 @@ def __init__(
             optimizer=optimizer,
             warm_start=warm_start,
             initial_point=initial_point,
+            callback=callback,
         )
 
     @property

diff --git a/qiskit_machine_learning/algorithms/trainable_model.py b/qiskit_machine_learning/algorithms/trainable_model.py
@@ -12,10 +12,11 @@
 """A base ML model with a Scikit-Learn like interface."""
 
 from abc import abstractmethod
-from typing import Union, Optional
+from typing import Union, Optional, Callable
 
 import numpy as np
 from qiskit.algorithms.optimizers import Optimizer, SLSQP
+from qiskit.utils import algorithm_globals
 
 from qiskit_machine_learning import QiskitMachineLearningError
 from qiskit_machine_learning.neural_networks import NeuralNetwork
@@ -28,6 +29,8 @@
 )
 from qiskit_machine_learning.deprecation import deprecate_values
 
+from .objective_functions import ObjectiveFunction
+
 
 class TrainableModel:
     """Base class for ML model. This class defines Scikit-Learn like interface to implement."""
@@ -42,6 +45,7 @@ def __init__(
         optimizer: Optional[Optimizer] = None,
         warm_start: bool = False,
         initial_point: np.ndarray = None,
+        callback: Optional[Callable[[np.ndarray, float], None]] = None,
     ):
         """
         Args:
@@ -62,7 +66,11 @@ def __init__(
             optimizer: An instance of an optimizer to be used in training. When `None` defaults to SLSQP.
             warm_start: Use weights from previous fit to start next fit.
             initial_point: Initial point for the optimizer to start from.
-
+            callback: a reference to a user's callback function that has two parameters and
+                returns ``None``. The callback can access intermediate data during training.
+                On each iteration an optimizer invokes the callback and passes current weights
+                as an array and a computed value as a float of the objective function being
+                optimized. This allows to track how well optimization / training process is going on.
         Raises:
             QiskitMachineLearningError: unknown loss, invalid neural network
         """
@@ -94,6 +102,7 @@ def __init__(
         self._warm_start = warm_start
         self._fit_result = None
         self._initial_point = initial_point
+        self._callback = callback
 
     @property
     def neural_network(self):
@@ -194,5 +203,29 @@ def _choose_initial_point(self) -> np.ndarray:
         if self._warm_start and self._fit_result is not None:
             self._initial_point = self._fit_result[0]
         elif self._initial_point is None:
-            self._initial_point = np.random.rand(self._neural_network.num_weights)
+            self._initial_point = algorithm_globals.random.random(self._neural_network.num_weights)
         return self._initial_point
+
+    def _get_objective(
+        self,
+        function: ObjectiveFunction,
+    ) -> Callable:
+        """
+        Wraps the given `ObjectiveFunction` to add callback calls, if `callback` is not None, along
+        with evaluating the objective value. Returned objective function is passed to
+        `Optimizer.optimize()`.
+        Args:
+            function: The objective function whose objective is to be evaluated.
+
+        Returns:
+            Objective function to evaluate objective value and optionally invoke callback calls.
+        """
+        if self._callback is None:
+            return function.objective
+
+        def objective(objective_weights):
+            objective_value = function.objective(objective_weights)
+            self._callback(objective_weights, objective_value)
+            return objective_value
+
+        return objective
diff --git a/qiskit_machine_learning/circuit/library/raw_feature_vector.py b/qiskit_machine_learning/circuit/library/raw_feature_vector.py
@@ -12,7 +12,7 @@
 
 """The raw feature vector circuit."""
 
-from typing import Optional
+from typing import Optional, List
 import numpy as np
 from qiskit.exceptions import QiskitError
 from qiskit.circuit import QuantumRegister, QuantumCircuit, ParameterVector, Instruction
@@ -27,7 +27,11 @@ class RawFeatureVector(BlueprintCircuit):
     placeholder instruction that can only be synthesized/defined when all parameters are bound.
 
     In ML, this circuit can be used to load the training data into qubit amplitudes. It does not
-    apply an kernel transformation. (Therefore, it is a "raw" feature vector.)
+    apply an kernel transformation (therefore, it is a "raw" feature vector).
+
+    Since initialization is implemented via a ``QuantumCircuit.initialize()`` call, this circuit
+    can't be used with gradient based optimizers, one can see a warning that gradients can't be
+    computed.
 
     Examples:
 
@@ -65,15 +69,14 @@ class RawFeatureVector(BlueprintCircuit):
     def __init__(self, feature_dimension: Optional[int]) -> None:
         """
         Args:
-            feature_dimension: The feature dimension and number of qubits.
+            feature_dimension: The feature dimension from which the number of
+                                qubits is inferred as ``n_qubits = log2(feature_dim)``
 
         """
         super().__init__()
 
-        self._num_qubits = None
         self._ordered_parameters = ParameterVector("x")
-
-        if feature_dimension:
+        if feature_dimension is not None:
             self.feature_dimension = feature_dimension
 
     def _build(self):
@@ -103,7 +106,7 @@ def num_qubits(self) -> int:
         Returns:
             The number of qubits.
         """
-        return self._num_qubits if self._num_qubits is not None else 0
+        return super().num_qubits
 
     @num_qubits.setter
     def num_qubits(self, num_qubits: int) -> None:
@@ -112,11 +115,12 @@ def num_qubits(self, num_qubits: int) -> None:
         Args:
             The new number of qubits.
         """
-        if self._num_qubits != num_qubits:
+        if self.num_qubits != num_qubits:
             # invalidate the circuit
             self._invalidate()
-            self._num_qubits = num_qubits
-            self.add_register(QuantumRegister(self.num_qubits, "q"))
+            self.qregs: List[QuantumRegister] = []
+            if num_qubits is not None and num_qubits > 0:
+                self.qregs = [QuantumRegister(num_qubits, name="q")]
 
     @property
     def feature_dimension(self) -> int:
@@ -141,14 +145,10 @@ def feature_dimension(self, feature_dimension: int) -> None:
         if int(num_qubits) != num_qubits:
             raise ValueError("feature_dimension must be a power of 2!")
 
-        if self._num_qubits is None or num_qubits != self._num_qubits:
+        if num_qubits != self.num_qubits:
             self._invalidate()
             self.num_qubits = int(num_qubits)
 
-    def _invalidate(self):
-        super()._invalidate()
-        self._num_qubits = None
-
 
 class ParameterizedInitialize(Instruction):
     """A normalized parameterized initialize instruction."""