physbo 3.0.0__tar.gz → 3.1.0__tar.gz

{physbo-3.0.0 → physbo-3.1.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: physbo
-Version: 3.0.0
+Version: 3.1.0
 Summary: optimization tool for PHYSics based on Bayesian Optimization
 Author-email: PHYSBO developers <physbo-dev@issp.u-tokyo.ac.jp>
 License-Expression: MPL-2.0
@@ -12,6 +12,8 @@ Requires-Python: >=3.9
 Description-Content-Type: text/markdown
 Requires-Dist: numpy
 Requires-Dist: scipy
+Provides-Extra: odat-se
+Requires-Dist: odat-se; extra == "odat-se"
 
 # optimization tools for PHYsics based on Bayesian Optimization ( PHYSBO )
 
@@ -35,13 +37,6 @@ PHYSBO was developed based on [COMBO](https://github.com/tsudalab/combo) for aca
 - NumPy
 - SciPy
 
-### About Cython
-
-From v3.0.0, PHYSBO no longer uses Cython in order to simplify installation process particularly on Windows.
-This means that the performance of PHYSBO is slightly degraded from older versions.
-If you need more performance, you can install ``physbo-core-cython`` additionally.
-This package offers Cythonized version of some functions of PHYSBO.
-
 ## Install
 
 - From PyPI (recommended)
@@ -67,12 +62,6 @@ python3 -m pip install physbo
         python3 -m pip install ./PHYSBO
         ```
 
-- To install ``physbo-core-cython`` ::
-
-  ```bash
-  python3 -m pip install physbo-core-cython
-  ```
-
 ## Uninstall
 
 ```bash

{physbo-3.0.0 → physbo-3.1.0}/README.md

@@ -20,13 +20,6 @@ PHYSBO was developed based on [COMBO](https://github.com/tsudalab/combo) for aca
 - NumPy
 - SciPy
 
-### About Cython
-
-From v3.0.0, PHYSBO no longer uses Cython in order to simplify installation process particularly on Windows.
-This means that the performance of PHYSBO is slightly degraded from older versions.
-If you need more performance, you can install ``physbo-core-cython`` additionally.
-This package offers Cythonized version of some functions of PHYSBO.
-
 ## Install
 
 - From PyPI (recommended)
@@ -52,12 +45,6 @@ python3 -m pip install physbo
         python3 -m pip install ./PHYSBO
         ```
 
-- To install ``physbo-core-cython`` ::
-
-  ```bash
-  python3 -m pip install physbo-core-cython
-  ```
-
 ## Uninstall
 
 ```bash

{physbo-3.0.0 → physbo-3.1.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "physbo"
-version = "3.0.0"
+version = "3.1.0"
 description = "optimization tool for PHYSics based on Bayesian Optimization"
 readme = { file = "README.md", content-type = "text/markdown" }
 license = "MPL-2.0"
@@ -16,6 +16,9 @@ dependencies = [
     "scipy",
 ]
 
+[project.optional-dependencies]
+odat-se = [ "odat-se" ]
+
 [project.urls]
 homepage = "https://www.pasums.issp.u-tokyo.ac.jp/physbo/en"
 source = "https://github.com/issp-center-dev/PHYSBO"

{physbo-3.0.0 → physbo-3.1.0}/src/physbo/__init__.py

@@ -26,7 +26,7 @@ def variable(*args, **kwargs):
     return Variable(*args, **kwargs)
 
 
-__version__ = "3.0.0"
+__version__ = "3.1.0"
 
 # __all__ is used to specify the public API of the package
 # and is used by sphinx to generate the API documentation

{physbo-3.0.0 → physbo-3.1.0}/src/physbo/gp/cov/_gauss.py

@@ -8,9 +8,38 @@
 # -*- coding:utf-8 -*-
 import numpy as np
 from scipy import spatial
-from ._src import grad_width64
 
 
+def grad_width(X, width, G):
+    """
+    Gradiant along width direction.
+
+    Parameters
+    ----------
+    X: numpy.ndarray[numpy.float64_t, ndim = 2]
+
+    width: numpy.ndarray[numpy.float64_t, ndim = 1]
+        The grid width
+    G: numpy.ndarray[numpy.float64_t, ndim = 2]
+        The gram matrix
+    Returns
+    -------
+    numpy.ndarray
+    """
+
+    ## The meaning of the following einsum is:
+    # N = X.shape[0]
+    # D = X.shape[1]
+    # gradG = np.zeros([D, N, N], dtype=DTYPE64)
+    # for d in range(D):
+    #     for i in range(N):
+    #         for j in range(i + 1, N):
+    #             gradG[d, i, j] = (X[i, d] - X[j, d]) / width[d]
+    #             gradG[d, i, j] = gradG[d, i, j] ** 2 * G[i, j]
+    #             gradG[d, j, i] = gradG[d, i, j]
+    gradG = np.einsum("ijd,ij->dij", ((X[:, None, :] - X) / width)**2, G)
+    return gradG
+
 class Gauss:
     """gaussian kernel"""
 
@@ -126,7 +155,7 @@ class Gauss:
 
         grad = np.zeros((self.num_params, num_data, num_data))
         if self.ard:
-            grad[0 : self.num_params - 1, :, :] = grad_width64(X, width, G)
+            grad[0 : self.num_params - 1, :, :] = grad_width(X, width, G)
         else:
             pairwise_dists = spatial.distance.pdist(X / width, "euclidean")
             grad[0, :, :] = G * spatial.distance.squareform(pairwise_dists**2)

{physbo-3.0.0 → physbo-3.1.0}/src/physbo/misc/__init__.py

@@ -8,18 +8,15 @@
 from ._centering import centering as centering
 from ._gauss_elim import gauss_elim as gauss_elim
 from ._set_config import SetConfig as SetConfig
-from ._use_cython import use_cython as use_cython
-
-from ._src import cholupdate as cholupdate
-from ._src import diagAB as diagAB
-from ._src import logsumexp64 as logsumexp64
-from ._src import traceAB2 as traceAB2
-from ._src import traceAB3 as traceAB3
 
+from ._cholesky import cholupdate as cholupdate
+from ._matrix import diagAB as diagAB
+from ._matrix import logsumexp as logsumexp
+from ._matrix import traceAB2 as traceAB2
+from ._matrix import traceAB3 as traceAB3
 
 __warned_names = set()
 
-
 def deprecated_warning(old: str, new: str):
     """
     Print a warning message when a deprecated name is used.

physbo-3.1.0/src/physbo/misc/_cholesky.py

@@ -0,0 +1,59 @@
+# SPDX-License-Identifier: MPL-2.0
+# Copyright (C) 2020- The University of Tokyo
+#
+# This Source Code Form is subject to the terms of the Mozilla Public
+# License, v. 2.0. If a copy of the MPL was not distributed with this
+# file, You can obtain one at https://mozilla.org/MPL/2.0/.
+
+import numpy as np
+import scipy
+
+
+def _cholupdate_fastupdate(U, x):
+    """Cholesky update using fast update method."""
+
+    N = x.shape[0]
+    x2 = x.copy()
+
+    for k in range(N):
+        r = np.hypot(U[k, k], x2[k])
+        c = r / U[k, k]
+        s = x2[k] / U[k, k]
+        U[k, k] = r
+
+        U[k, k + 1 :] += s * x2[k + 1 :]
+        U[k, k + 1 :] /= c
+        x2[k + 1 :] *= c
+        x2[k + 1 :] -= s * U[k, k + 1 :]
+
+
+def _cholupdate_direct(U, x):
+    """Cholesky update just by calling scipy.linalg.cholesky directly."""
+
+    A = np.dot(U.T, U) + np.outer(x, x)
+    U[:] = scipy.linalg.cholesky(A, check_finite=False)
+
+
+def cholupdate(U, x, matrix_size_threshold=500):
+    """Cholesky update
+
+    This calculates the Cholesky decomposition of A = U.T @ U + x @ x.T.
+
+    Parameters
+    ----------
+    U: numpy.ndarray
+        Upper triangular matrix of the Cholesky decomposition of the original matrix.
+        U is updated to U' of A in place.
+    x: numpy.ndarray
+        Vector to be added to the original matrix.
+    matrix_size_threshold: int
+        If the size of the matrix is larger than this threshold, the fast update method is used.
+        Otherwise, the direct method is used.
+        Default is 500.
+    """
+
+    N = U.shape[0]
+    if N > matrix_size_threshold:
+        _cholupdate_fastupdate(U, x)
+    else:
+        _cholupdate_direct(U, x)

physbo-3.1.0/src/physbo/misc/_matrix.py

@@ -0,0 +1,74 @@
+# SPDX-License-Identifier: MPL-2.0
+# Copyright (C) 2020- The University of Tokyo
+#
+# This Source Code Form is subject to the terms of the Mozilla Public
+# License, v. 2.0. If a copy of the MPL was not distributed with this
+# file, You can obtain one at https://mozilla.org/MPL/2.0/.
+
+import numpy as np
+
+def diagAB(A, B):
+    """Return diagonal part of AB
+
+    Parameters
+    ==========
+    A: np.ndarray
+        NxM matrix
+    B: np.ndarray
+        MxN matrix
+
+    Returns
+    =======
+    d: np.ndarray
+        Diagonal part of the matrix AB
+    """
+
+    return np.einsum("ij,ji->i", A, B)
+
+
+def logsumexp(x):
+    """Calculate log(sum(exp(x)))
+
+    Parameters
+    ==========
+    x: np.ndarray
+    """
+    xmax = np.max(x)
+    return np.log(np.sum(np.exp(x - xmax))) + xmax
+
+
+def traceAB3(A, B):
+    """Calculates vector of trace of AB[i], where i is the first axis of 3-rank tensor B
+
+    Parameters
+    ==========
+    A: np.ndarray
+        NxM matrix
+    B: np.ndarray
+        dxMxN tensor
+
+    Returns
+    =======
+    traceAB: np.ndarray
+    """
+
+    return np.einsum("ij,kji->k", A, B)
+
+
+def traceAB2(A, B):
+    """Calculates trace of AB
+
+    Parameters
+    ==========
+    A: np.ndarray
+        NxM matrix
+    B: np.ndarray
+        MxN matrix
+
+    Returns
+    =======
+    traceAB: float
+        trace of the matrix AB
+    """
+
+    return np.einsum("ij,ji->", A, B)

{physbo-3.0.0 → physbo-3.1.0}/src/physbo.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: physbo
-Version: 3.0.0
+Version: 3.1.0
 Summary: optimization tool for PHYSics based on Bayesian Optimization
 Author-email: PHYSBO developers <physbo-dev@issp.u-tokyo.ac.jp>
 License-Expression: MPL-2.0
@@ -12,6 +12,8 @@ Requires-Python: >=3.9
 Description-Content-Type: text/markdown
 Requires-Dist: numpy
 Requires-Dist: scipy
+Provides-Extra: odat-se
+Requires-Dist: odat-se; extra == "odat-se"
 
 # optimization tools for PHYsics based on Bayesian Optimization ( PHYSBO )
 
@@ -35,13 +37,6 @@ PHYSBO was developed based on [COMBO](https://github.com/tsudalab/combo) for aca
 - NumPy
 - SciPy
 
-### About Cython
-
-From v3.0.0, PHYSBO no longer uses Cython in order to simplify installation process particularly on Windows.
-This means that the performance of PHYSBO is slightly degraded from older versions.
-If you need more performance, you can install ``physbo-core-cython`` additionally.
-This package offers Cythonized version of some functions of PHYSBO.
-
 ## Install
 
 - From PyPI (recommended)
@@ -67,12 +62,6 @@ python3 -m pip install physbo
         python3 -m pip install ./PHYSBO
         ```
 
-- To install ``physbo-core-cython`` ::
-
-  ```bash
-  python3 -m pip install physbo-core-cython
-  ```
-
 ## Uninstall
 
 ```bash

{physbo-3.0.0 → physbo-3.1.0}/src/physbo.egg-info/SOURCES.txt

@@ -30,8 +30,6 @@ src/physbo/gp/core/_prior.py
 src/physbo/gp/core/learning.py
 src/physbo/gp/cov/__init__.py
 src/physbo/gp/cov/_gauss.py
-src/physbo/gp/cov/_src/__init__.py
-src/physbo/gp/cov/_src/pure.py
 src/physbo/gp/inf/__init__.py
 src/physbo/gp/inf/exact.py
 src/physbo/gp/lik/__init__.py
@@ -41,12 +39,11 @@ src/physbo/gp/mean/_const.py
 src/physbo/gp/mean/_zero.py
 src/physbo/misc/__init__.py
 src/physbo/misc/_centering.py
+src/physbo/misc/_cholesky.py
 src/physbo/misc/_gauss_elim.py
+src/physbo/misc/_matrix.py
 src/physbo/misc/_set_config.py
-src/physbo/misc/_use_cython.py
 src/physbo/misc/permutation_importance.py
-src/physbo/misc/_src/__init__.py
-src/physbo/misc/_src/pure.py
 src/physbo/opt/__init__.py
 src/physbo/opt/_adam.py
 src/physbo/search/__init__.py

physbo-3.1.0/src/physbo.egg-info/requires.txt

@@ -0,0 +1,5 @@
+numpy
+scipy
+
+[odat-se]
+odat-se

physbo-3.0.0/src/physbo/gp/cov/_src/__init__.py

@@ -1,15 +0,0 @@
-# SPDX-License-Identifier: MPL-2.0
-# Copyright (C) 2020- The University of Tokyo
-#
-# This Source Code Form is subject to the terms of the Mozilla Public
-# License, v. 2.0. If a copy of the MPL was not distributed with this
-# file, You can obtain one at https://mozilla.org/MPL/2.0/.
-
-from ....misc import use_cython
-
-if use_cython():
-    from physbo_core_cython.gp.cov._src.enhance_gauss import grad_width64, grad_width32
-else:
-    from .pure import grad_width64, grad_width32
-
-__all__ = ["grad_width64", "grad_width32"]

physbo-3.0.0/src/physbo/gp/cov/_src/pure.py

@@ -1,72 +0,0 @@
-# SPDX-License-Identifier: MPL-2.0
-# Copyright (C) 2020- The University of Tokyo
-#
-# This Source Code Form is subject to the terms of the Mozilla Public
-# License, v. 2.0. If a copy of the MPL was not distributed with this
-# file, You can obtain one at https://mozilla.org/MPL/2.0/.
-
-import numpy as np
-
-DTYPE64 = np.float64
-DTYPE32 = np.float32
-
-
-def grad_width64(X, width, G):
-    """
-    Gradiant along width direction (64bit).
-
-    Parameters
-    ----------
-    X: numpy.ndarray[numpy.float64_t, ndim = 2]
-
-    width: numpy.ndarray[numpy.float64_t, ndim = 1]
-        The grid width
-    G: numpy.ndarray[numpy.float64_t, ndim = 2]
-        The gram matrix
-    Returns
-    -------
-    numpy.ndarray
-    """
-    N = X.shape[0]
-    D = X.shape[1]
-
-    gradG = np.zeros([D, N, N], dtype=DTYPE64)
-
-    for d in range(D):
-        for i in range(N):
-            for j in range(i + 1, N):
-                gradG[d, i, j] = (X[i, d] - X[j, d]) / width[d]
-                gradG[d, i, j] = gradG[d, i, j] ** 2 * G[i, j]
-                gradG[d, j, i] = gradG[d, i, j]
-    return gradG
-
-
-def grad_width32(X, width, G):
-    """
-
-    Gradiant along width direction (32bit).
-
-    Parameters
-    ----------
-    X: numpy.ndarray[numpy.float32_t, ndim = 2]
-
-    width: numpy.ndarray[numpy.float32_t, ndim = 1]
-        The grid width
-    G: numpy.ndarray[numpy.float32_t, ndim = 2]
-        The gram matrix
-    Returns
-    -------
-    numpy.ndarray
-    """
-    N = X.shape[0]
-    D = X.shape[1]
-
-    gradG = np.zeros([D, N, N], dtype=DTYPE32)
-
-    for d in range(D):
-        for i in range(N):
-            for j in range(i + 1, N):
-                gradG[d, i, j] = (X[i, d] - X[j, d]) / width[d]
-                gradG[d, i, j] = gradG[d, i, j] ** 2 * G[i, j]
-                gradG[d, j, i] = gradG[d, i, j]
-    return gradG

physbo-3.0.0/src/physbo/misc/_src/__init__.py

@@ -1,27 +0,0 @@
-# SPDX-License-Identifier: MPL-2.0
-# Copyright (C) 2020- The University of Tokyo
-#
-# This Source Code Form is subject to the terms of the Mozilla Public
-# License, v. 2.0. If a copy of the MPL was not distributed with this
-# file, You can obtain one at https://mozilla.org/MPL/2.0/.
-
-from .. import use_cython
-
-if use_cython():
-    from physbo_core_cython.misc._src import (
-        cholupdate64 as cholupdate,
-        diagAB_64 as diagAB,
-        logsumexp64 as logsumexp64,
-        traceAB2_64 as traceAB2,
-        traceAB3_64 as traceAB3,
-    )
-else:
-    from .pure import (
-        cholupdate64 as cholupdate,
-        diagAB_64 as diagAB,
-        logsumexp64 as logsumexp64,
-        traceAB2_64 as traceAB2,
-        traceAB3_64 as traceAB3,
-    )
-
-__all__ = ["cholupdate", "diagAB", "logsumexp64", "traceAB2", "traceAB3"]

physbo-3.0.0/src/physbo/misc/_src/pure.py

@@ -1,159 +0,0 @@
-# SPDX-License-Identifier: MPL-2.0
-# Copyright (C) 2020- The University of Tokyo
-#
-# This Source Code Form is subject to the terms of the Mozilla Public
-# License, v. 2.0. If a copy of the MPL was not distributed with this
-# file, You can obtain one at https://mozilla.org/MPL/2.0/.
-
-# Pure Python version of merged Cython functions
-# Converted from traceAB.pyx, cholupdate.pyx, diagAB.pyx, and logsumexp.pyx
-
-import math
-import numpy as np
-
-DTYPE = np.float64
-
-# ==============================================================================
-# Functions from cholupdate.pyx
-# ==============================================================================
-
-
-def hypot(x, y):
-    """Pure Python version of hypot function"""
-    x = abs(x)
-    y = abs(y)
-    t = min(x, y)
-    x = max(x, y)
-    if x == 0:
-        return 0
-    t = t / x
-    return x * math.sqrt(1 + t * t)
-
-
-def cholupdate64(L, x):
-    """Pure Python version of Cholesky update"""
-    N = x.shape[0]
-    x2 = x.copy()
-
-    for k in range(N):
-        r = hypot(L[k, k], x2[k])
-        c = r / L[k, k]
-        s = x2[k] / L[k, k]
-        L[k, k] = r
-
-        for i in range(k + 1, N):
-            L[k, i] = (L[k, i] + s * x2[i]) / c
-            x2[i] = c * x2[i] - s * L[k, i]
-
-
-# ==============================================================================
-# Functions from diagAB.pyx
-# ==============================================================================
-
-
-def diagAB_64(A, B):
-    """Return diagonal part of AB
-
-    Parameters
-    ==========
-    A: np.ndarray
-        NxM matrix
-    B: np.ndarray
-        MxN matrix
-
-    Returns
-    =======
-    d: np.ndarray
-        Diagonal part of the matrix AB
-    """
-    N = A.shape[0]
-    M = A.shape[1]
-
-    diagAB = np.zeros(N, dtype=DTYPE)
-
-    for i in range(N):
-        for j in range(M):
-            diagAB[i] += A[i, j] * B[j, i]
-
-    return diagAB
-
-
-# ==============================================================================
-# Functions from logsumexp.pyx
-# ==============================================================================
-
-
-def logsumexp64(x):
-    """Calculate log(sum(exp(x)))
-
-    Parameters
-    ==========
-    x: np.ndarray
-    """
-    N = x.shape[0]
-    tmp = 0.0
-
-    xmax = np.max(x)
-
-    for i in range(N):
-        tmp += math.exp(x[i] - xmax)
-
-    return math.log(tmp) + xmax
-
-
-# ==============================================================================
-# Functions from traceAB.pyx
-# ==============================================================================
-
-
-def traceAB3_64(A, B):
-    """Calculates vector of trace of AB[i], where i is the first axis of 3-rank tensor B
-
-    Parameters
-    ==========
-    A: np.ndarray
-        NxM matrix
-    B: np.ndarray
-        dxMxN tensor
-
-    Returns
-    =======
-    traceAB: np.ndarray
-    """
-    N = A.shape[0]
-    M = A.shape[1]
-    D = B.shape[0]
-
-    traceAB = np.zeros(D, dtype=DTYPE)
-
-    for d in range(D):
-        for i in range(N):
-            for j in range(M):
-                traceAB[d] += A[i, j] * B[d, j, i]
-    return traceAB
-
-
-def traceAB2_64(A, B):
-    """Calculates trace of AB
-
-    Parameters
-    ==========
-    A: np.ndarray
-        NxM matrix
-    B: np.ndarray
-        MxN matrix
-
-    Returns
-    =======
-    traceAB: float
-        trace of the matrix AB
-    """
-    N = A.shape[0]
-    M = A.shape[1]
-
-    traceAB = 0.0
-
-    for i in range(N):
-        for j in range(M):
-            traceAB += A[i, j] * B[j, i]
-    return traceAB

physbo-3.0.0/src/physbo/misc/_use_cython.py

@@ -1,33 +0,0 @@
-# SPDX-License-Identifier: MPL-2.0
-# Copyright (C) 2020- The University of Tokyo
-#
-# This Source Code Form is subject to the terms of the Mozilla Public
-# License, v. 2.0. If a copy of the MPL was not distributed with this
-# file, You can obtain one at https://mozilla.org/MPL/2.0/.
-
-import os
-from importlib.util import find_spec
-
-if find_spec("physbo_core_cython"):
-    CYTHON_AVAILABLE = True
-else:
-    CYTHON_AVAILABLE = False
-
-PHYSBO_USE_CYTHON = os.environ.get("PHYSBO_USE_CYTHON", "auto")
-if PHYSBO_USE_CYTHON == "0":
-    USING_CYTHON = False
-elif PHYSBO_USE_CYTHON == "1":
-    if not CYTHON_AVAILABLE:
-        raise ImportError(
-            "PHYSBO_USE_CYTHON is set to 1, but physbo_core_cython is not installed"
-        )
-    USING_CYTHON = True
-else:  # PHYSBO_USE_CYTHON == "auto"
-    USING_CYTHON = CYTHON_AVAILABLE
-
-if USING_CYTHON:
-    print("Cythonized version of physbo is used")
-
-
-def use_cython():
-    return USING_CYTHON

physbo-3.0.0/src/physbo.egg-info/requires.txt

@@ -1,2 +0,0 @@
-numpy
-scipy