Marc record tensornetwork entropy

CI/unit_tests/observables/test_fisher_trace_calculation.py

@@ -0,0 +1,59 @@
+"""
+ZnNL: A Zincwarecode package.
+
+License
+-------
+This program and the accompanying materials are made available under the terms
+of the Eclipse Public License v2.0 which accompanies this distribution, and is
+available at https://www.eclipse.org/legal/epl-v20.html
+
+SPDX-License-Identifier: EPL-2.0
+
+Copyright Contributors to the Zincwarecode Project.
+
+Contact Information
+-------------------
+email: [email protected]
+github: https://github.com/zincware
+web: https://zincwarecode.com/
+
+Citation
+--------
+If you use this module please cite us with:
+
+Summary
+-------
+This module tests the implementation of the fisher trace computation module.
+"""
+
+import numpy as np
+
+from znnl.observables.fisher_trace import compute_fisher_trace
+
+
+class TestFisherTrace:
+    """
+    Class for testing the implementation of the fisher trace calculation
+    """
+
+    def test_fisher_trace_computation(self):
+        """
+        Function tests if the fisher trace computation works correctly for an
+        example which was calculated by hand before.
+
+        Returns
+        -------
+        Asserts the calculated fisher trace for the manually defined inputs
+        is what it should be.
+        """
+
+        ntk = np.array(
+            [
+                [[[1, 2, 3], [4, 5, 6], [7, 8, 9]], np.random.rand(3, 3)],
+                [np.random.rand(3, 3), [[2, 1, 3], [1, 2, 3], [3, 2, 1]]],
+            ]
+        )
+        loss_derivative = np.array([[5, 4, 3], [2, 1, 0]])
+
+        trace = compute_fisher_trace(loss_derivative=loss_derivative, ntk=ntk)
+        assert trace == 638 / 2

CI/unit_tests/observables/test_tensornetwork_matrix.py

@@ -0,0 +1,57 @@
+"""
+ZnNL: A Zincwarecode package.
+
+License
+-------
+This program and the accompanying materials are made available under the terms
+of the Eclipse Public License v2.0 which accompanies this distribution, and is
+available at https://www.eclipse.org/legal/epl-v20.html
+
+SPDX-License-Identifier: EPL-2.0
+
+Copyright Contributors to the Zincwarecode Project.
+
+Contact Information
+-------------------
+email: [email protected]
+github: https://github.com/zincware
+web: https://zincwarecode.com/
+
+Citation
+--------
+If you use this module please cite us with:
+
+Summary
+-------
+This module tests the implementation of the tensornetwork matrix computation module.
+"""
+
+import numpy as np
+from numpy.testing import assert_almost_equal
+
+from znnl.observables.tensornetwork_matrix import compute_tensornetwork_matrix
+
+
+class TestTensornetworkMatrix:
+    """
+    Class for testing the implementation of the tensornetwork matrix calculation
+    """
+
+    def test_tensornetwork_matrix_computation(self):
+        """
+        Function tests if the fisher trace computation works correctly for an
+        example which was calculated by hand before.
+
+        Returns
+        -------
+        Asserts the calculated fisher trace for the manually defined inputs
+        is what it should be.
+        """
+
+        ntk = np.array([[1, 2, 3, 4], [1, 1, 1, 1], [0, 1, 2, 4], [1, 0, 0, 8]])
+        targets = np.array([2, 2, 1, 1])
+
+        matrix = compute_tensornetwork_matrix(ntk=ntk, targets=targets)
+        correctmatrix = [[3.5, 0.5], [2.25, 1.25]]
+
+        assert_almost_equal(matrix, correctmatrix)

CI/unit_tests/training_recording/test_training_recording.py

@@ -64,6 +64,10 @@ def test_instantiation(self):
             eigenvalues=True,
             trace=True,
             loss_derivative=True,
+            fisher_trace=True,
+            tensornetwork_entropy=True,
+            tensornetwork_covariance_entropy=True,
+            tensornetwork_magnitude_entropy=True,
         )
         recorder.instantiate_recorder(data_set=self.dummy_data_set)
         _exclude_list = [

examples/Computing-Collective-Variables.ipynb

@@ -1,6 +1,7 @@
 {
  "cells": [
   {
+   "attachments": {},
    "cell_type": "markdown",
    "id": "1155f508",
    "metadata": {},
@@ -11,6 +12,7 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "id": "c2f5d943",
    "metadata": {},
@@ -33,6 +35,7 @@
     "os.environ['CUDA_VISIBLE_DEVICES'] = '-1'\n",
     "\n",
     "import znnl as nl\n",
+    "from znnl.utils.matrix_utils import calculate_l_pq_norm\n",
     "from neural_tangents import stax\n",
     "import optax\n",
     "\n",
@@ -45,6 +48,7 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "id": "9d379c99",
    "metadata": {},
@@ -111,6 +115,7 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "id": "38fa1b9d",
    "metadata": {
@@ -154,6 +159,7 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "id": "a6a9fbe0-def2-4bab-a808-8858ab2aa5e9",
    "metadata": {
@@ -207,6 +213,7 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "id": "612896b2-e9e2-4917-b01b-1ff56da2b142",
    "metadata": {
@@ -233,6 +240,7 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "id": "38bda516-9feb-4ab1-b3d3-3a552b46e28c",
    "metadata": {
@@ -259,6 +267,7 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "id": "99646c3f-95ed-4c4e-b01e-5646b3767ff4",
    "metadata": {},
@@ -331,8 +340,9 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "loss_derivative = calculate_l_pq_norm(train_report.loss_derivative)\n",
     "plt.plot(\n",
-    "    train_report.loss_derivative, \n",
+    "    loss_derivative, \n",
     "    'o', \n",
     "    mfc='None', \n",
     "    label=r\"$|| \\frac{\\partial L }{ \\partial f}||_R$\"\n",

znnl/observables/__init__.py

@@ -0,0 +1,40 @@
+"""
+ZnNL: A Zincwarecode package.
+
+License
+-------
+This program and the accompanying materials are made available under the terms
+of the Eclipse Public License v2.0 which accompanies this distribution, and is
+available at https://www.eclipse.org/legal/epl-v20.html
+
+SPDX-License-Identifier: EPL-2.0
+
+Copyright Contributors to the Zincwarecode Project.
+
+Contact Information
+-------------------
+email: [email protected]
+github: https://github.com/zincware
+web: https://zincwarecode.com/
+
+Citation
+--------
+If you use this module please cite us with:
+
+Summary
+-------
+Module for the observables.
+"""
+from znnl.observables.covariance_entropy import compute_covariance_entropy
+from znnl.observables.entropy import compute_entropy
+from znnl.observables.fisher_trace import compute_fisher_trace
+from znnl.observables.magnitude_entropy import compute_magnitude_density
+from znnl.observables.tensornetwork_matrix import compute_tensornetwork_matrix
+
+__all__ = [
+    compute_fisher_trace.__name__,
+    compute_tensornetwork_matrix.__name__,
+    compute_entropy.__name__,
+    compute_magnitude_density.__name__,
+    compute_covariance_entropy.__name__,
+]

znnl/observables/covariance_entropy.py

@@ -0,0 +1,53 @@
+"""
+ZnNL: A Zincwarecode package.
+
+License
+-------
+This program and the accompanying materials are made available under the terms
+of the Eclipse Public License v2.0 which accompanies this distribution, and is
+available at https://www.eclipse.org/legal/epl-v20.html
+
+SPDX-License-Identifier: EPL-2.0
+
+Copyright Contributors to the Zincwarecode Project.
+
+Contact Information
+-------------------
+email: [email protected]
+github: https://github.com/zincware
+web: https://zincwarecode.com/
+
+Citation
+--------
+If you use this module please cite us with:
+
+Summary
+-------
+Module for the computation of the matrix covariance entropy.
+"""
+
+import numpy as np
+
+from znnl.analysis.entropy import EntropyAnalysis
+from znnl.utils.matrix_utils import normalize_gram_matrix
+
+
+def compute_covariance_entropy(matrix: np.ndarray):
+    """
+    Function to compute the covariance entropy of a matrix.
+
+    Parameters
+    ----------
+    Matrix of which to compute the covariance entropy.
+
+    Returns
+    -------
+    Covariance entropy of the matrix."""
+
+    cov_matrix = normalize_gram_matrix(matrix)
+    calculator = EntropyAnalysis(matrix=cov_matrix)
+    entropy = calculator.compute_von_neumann_entropy(
+        effective=False, normalize_eig=True
+    )
+
+    return entropy

znnl/observables/entropy.py

@@ -0,0 +1,50 @@
+"""
+ZnNL: A Zincwarecode package.
+
+License
+-------
+This program and the accompanying materials are made available under the terms
+of the Eclipse Public License v2.0 which accompanies this distribution, and is
+available at https://www.eclipse.org/legal/epl-v20.html
+
+SPDX-License-Identifier: EPL-2.0
+
+Copyright Contributors to the Zincwarecode Project.
+
+Contact Information
+-------------------
+email: [email protected]
+github: https://github.com/zincware
+web: https://zincwarecode.com/
+
+Citation
+--------
+If you use this module please cite us with:
+
+Summary
+-------
+Module for the computation of the matrix entropy.
+"""
+
+import numpy as np
+
+from znnl.analysis.entropy import EntropyAnalysis
+
+
+def compute_entropy(matrix: np.ndarray):
+    """
+    Function to compute the entropy of a matrix.
+
+    Parameters
+    ----------
+    Matrix of which to compute the entropy.
+
+    Returns
+    -------
+    Entropy of the matrix."""
+    calculator = EntropyAnalysis(matrix=matrix)
+    entropy = calculator.compute_von_neumann_entropy(
+        effective=False, normalize_eig=True
+    )
+
+    return entropy