Commit 9dae17d8 authored by Yamada Hiroyuki's avatar Yamada Hiroyuki
Browse files

Merge branch 'Feature_Numpy2.0'

parents dc52328f 5c0ca845

.github/workflows/build.yml

+6 −0
Original line number Diff line number Diff line
@@ -21,7 +21,11 @@ jobs: 
      matrix:

        runs-on: [ubuntu-latest, macos-13, macos-14, windows-latest]

        python: ['3.8', '3.9', '3.10', '3.11', '3.12']

        numpy: ["numpy<2.0.0", "numpy>=2.0.0rc2"]

        gym: ['gym', 'gymnasium']

        exclude:

          - python: "3.8"

            numpy: "numpy>=2.0.0rc2"

      fail-fast: false

    runs-on: ${{ matrix.runs-on }}

    steps:
@@ -39,6 +43,8 @@ jobs: 
          path: ${{ steps.pip-cache.outputs.dir }}

          key: ${{ runner.os }}-pip${{ matrix.python }}

          restore-keys: ${{ runner.os }}-pip${{ matrix.python }}

      - name: Install NumPy

        run: pip install '${{ matrix.numpy }}'

      - name: Install cpprb

        run: pip install '.[all]'

      - name: Install Gym(nasium)

Dockerfile

+5 −2
Original line number Diff line number Diff line
@@ -12,8 +12,11 @@ COPY --from=README /work/README.md /work/README.md 
COPY pyproject.toml setup.py LICENSE MANIFEST.in .

COPY cpprb cpprb/

ARG ON_CI

RUN ON_CI=${ON_CI} find /opt/python -name 'cp*' \

    -exec {}/bin/pip wheel . -w /work/wheel --no-deps \; && \

RUN ON_CI=${ON_CI} /opt/python/cp38-cp38/bin/pip wheel . -w /work/wheel --no-deps && \

    ON_CI=${ON_CI} /opt/python/cp39-cp39/bin/pip wheel . -w /work/wheel --no-deps && \

    ON_CI=${ON_CI} /opt/python/cp310-cp310/bin/pip wheel . -w /work/wheel --no-deps && \

    ON_CI=${ON_CI} /opt/python/cp311-cp311/bin/pip wheel . -w /work/wheel --no-deps && \

    ON_CI=${ON_CI} /opt/python/cp312-cp312/bin/pip wheel . -w /work/wheel --no-deps && \

    auditwheel repair /work/wheel/cpprb-*.whl -w /dist

cpprb/PyReplayBuffer.pyx

+37 −25
Original line number Diff line number Diff line
@@ -22,6 +22,9 @@ from .VectorWrapper import (VectorWrapper, 
                            VectorDouble,PointerDouble,VectorFloat)





COPY_ONLY_NECESSARY: bool | None = False if int(np.version.version[0]) < 2 else None





def default_logger(level=INFO):

    """

    Create default logger for cpprb
@@ -43,14 +46,17 @@ def default_logger(level=INFO): 
    return logger



cdef double [::1] Cdouble(array):

    return np.ravel(np.array(array,copy=False,dtype=np.double,ndmin=1,order='C'))

    return np.ravel(np.array(array, copy=COPY_ONLY_NECESSARY,

                             dtype=np.double, ndmin=1, order='C'))



cdef inline const size_t [::1] Csize(array):

    return np.ravel(np.array(array,copy=False,dtype=np.uint64,ndmin=1,order='C'))

    return np.ravel(np.array(array, copy=COPY_ONLY_NECESSARY,

                             dtype=np.uint64, ndmin=1, order='C'))



@cython.embedsignature(True)

cdef inline const float [::1] Cfloat(array):

    return np.ravel(np.array(array,copy=False,dtype=np.single,ndmin=1,order='C'))

    return np.ravel(np.array(array, copy=COPY_ONLY_NECESSARY,

                             dtype=np.single, ndmin=1, order='C'))





def unwrap(d):
@@ -444,7 +450,7 @@ cdef class SharedBuffer: 
                else:

                    raise



            len = int(np.array(shape,copy=False,dtype="int").prod())

            len = int(np.array(shape, copy=COPY_ONLY_NECESSARY, dtype="int").prod())

            self.data = RawArray(ctx,ctype,len,self.backend)

        else:

            self.data = data
@@ -552,7 +558,8 @@ def find_array(dict,key): 
        If `dict` has `key`, returns the values with numpy.ndarray with the minimum

        dimension of 1. Otherwise, returns `None`.

    """

    return None if not key in dict else np.array(dict[key],ndmin=1,copy=False)

    return None if not key in dict else np.array(dict[key], ndmin=1,

                                                 copy=COPY_ONLY_NECESSARY)



@cython.embedsignature(True)

cdef class StepChecker:
@@ -584,7 +591,7 @@ cdef class StepChecker: 
        kwargs: dict

            Added values.

        """

        return np.reshape(np.array(kwargs[self.check_str], copy=False),

        return np.reshape(np.asarray(kwargs[self.check_str]),

                          self.check_shape,order='A').shape[0]



@cython.embedsignature(True)
@@ -614,7 +621,7 @@ cdef class NstepBuffer: 
        self.default_dtype = default_dtype or np.single



        if next_of is not None: # next_of is not support yet.

            for name in np.array(next_of,copy=False,ndmin=1):

            for name in np.array(next_of, copy=COPY_ONLY_NECESSARY, ndmin=1):

                self.env_dict[f"next_{name}"] = self.env_dict[name]



        self.Nstep_size = Nstep["size"]
@@ -784,7 +791,7 @@ cdef class NstepBuffer: 


    cdef _extract(self,kwargs,name):

        _dict = self.env_dict[name]

        return np.reshape(np.array(kwargs[name],copy=False,ndmin=2,

        return np.reshape(np.array(kwargs[name], copy=COPY_ONLY_NECESSARY, ndmin=2,

                                   dtype=_dict.get("dtype",self.default_dtype)),

                          _dict["add_shape"])
@@ -963,13 +970,13 @@ cdef class ReplayBuffer: 
        self.episode_len = 0



        self.compress_any = stack_compress

        self.stack_compress = np.array(stack_compress,ndmin=1,copy=False)

        self.stack_compress = np.array(stack_compress, ndmin=1, copy=COPY_ONLY_NECESSARY)



        self.default_dtype = default_dtype or np.single



        self.has_next_of = next_of

        self.next_of = np.array(next_of,

                                ndmin=1,copy=False) if self.has_next_of else None

        self.next_of = np.array(next_of, ndmin=1,

                                copy=COPY_ONLY_NECESSARY) if self.has_next_of else None

        self.next_ = {}

        self.cache = {} if (self.has_next_of or self.compress_any) else None
@@ -1001,7 +1008,7 @@ cdef class ReplayBuffer: 
            for name in self.stack_compress:

                self.cache_size = max(self.cache_size,

                                      np.array(self.env_dict[name]["shape"],

                                               ndmin=1,copy=False)[-1] -1)

                                               ndmin=1, copy=COPY_ONLY_NECESSARY)[-1] -1)



        if self.has_next_of:

            self.cache_size += 1
@@ -1164,13 +1171,14 @@ cdef class ReplayBuffer: 
                self.add_cache_i(key, index)



        for name, b in self.buffer.items():

            b[add_idx] = np.reshape(np.array(kwargs[name],copy=False,ndmin=2),

            b[add_idx] = np.reshape(np.array(kwargs[name],

                                             copy=COPY_ONLY_NECESSARY, ndmin=2),

                                    self.env_dict[name]["add_shape"])



        if self.has_next_of:

            for name in self.next_of:

                self.next_[name][...]=np.reshape(np.array(kwargs[f"next_{name}"],

                                                          copy=False,

                                                          copy=COPY_ONLY_NECESSARY,

                                                          ndmin=2),

                                                 self.env_dict[name]["add_shape"])[-1]
@@ -1324,7 +1332,7 @@ cdef class ReplayBuffer: 
        cdef cache_idx

        cdef bool use_cache



        idx = np.array(idx,copy=False,ndmin=1)

        idx = np.array(idx,copy=COPY_ONLY_NECESSARY, ndmin=1)

        for name, b in self.buffer.items():

            sample[name] = b[idx]
@@ -1582,7 +1590,7 @@ cdef class PrioritizedReplayBuffer(ReplayBuffer): 
        self.check_for_update = check_for_update

        if self.check_for_update:

            self.unchange_since_sample = np.ones(np.array(size,

                                                          copy=False,

                                                          copy=COPY_ONLY_NECESSARY,

                                                          dtype='int'),

                                                 dtype='bool')
@@ -1659,7 +1667,7 @@ cdef class PrioritizedReplayBuffer(ReplayBuffer): 
        """

        cdef size_t N = self.size_check.step_size(kwargs)

        if priorities is not None:

            priorities = np.ravel(np.array(priorities,copy=False,

            priorities = np.ravel(np.array(priorities, copy=COPY_ONLY_NECESSARY,

                                           ndmin=1, dtype=np.single))

            if N != priorities.shape[0]:

                raise ValueError("`priorities` shape is incompatible")
@@ -1683,7 +1691,8 @@ cdef class PrioritizedReplayBuffer(ReplayBuffer): 
        cdef const float [:] ps



        if priorities is not None:

            ps = np.ravel(np.array(priorities,copy=False,ndmin=1,dtype=np.single))

            ps = np.ravel(np.array(priorities, copy=COPY_ONLY_NECESSARY,

                                   ndmin=1, dtype=np.single))

            self.per.set_priorities(index,&ps[0],N,self.get_buffer_size())

        else:

            self.per.set_priorities(index,N,self.get_buffer_size())
@@ -2098,7 +2107,8 @@ cdef class MPReplayBuffer: 
        self._lock_explorer()



        for name, b in self.buffer.items():

            b[add_idx] = np.reshape(np.array(kwargs[name],copy=False,ndmin=2),

            b[add_idx] = np.reshape(np.array(kwargs[name], ndmin=2,

                                             copy=COPY_ONLY_NECESSARY),

                                    self.env_dict[name]["add_shape"])



        self._unlock_explorer()
@@ -2132,7 +2142,7 @@ cdef class MPReplayBuffer: 
    def _encode_sample(self,idx):

        cdef sample = {}



        idx = np.array(idx,copy=False,ndmin=1)

        idx = np.array(idx, copy=COPY_ONLY_NECESSARY, ndmin=1)



        for name, b in self.buffer.items():

            sample[name] = b[idx]
@@ -2454,8 +2464,8 @@ cdef class MPPrioritizedReplayBuffer(MPReplayBuffer): 
        cdef const float [:] ps



        if priorities is not None:

            priorities = np.ravel(np.array(priorities,copy=False,

                                           ndmin=1,dtype=np.single))

            priorities = np.ravel(np.array(priorities, ndmin=1, dtype=np.single,

                                           copy=COPY_ONLY_NECESSARY))

            if N != priorities.shape[0]:

                raise ValueError("`priorities` shape is incompatible")
@@ -2470,7 +2480,8 @@ cdef class MPPrioritizedReplayBuffer(MPReplayBuffer): 
        self._lock_explorer_per()



        if priorities is not None:

            ps = np.ravel(np.array(priorities,copy=False,ndmin=1,dtype=np.single))

            ps = np.ravel(np.array(priorities, copy=COPY_ONLY_NECESSARY,

                                   ndmin=1, dtype=np.single))

            self.per.ptr().set_priorities(index,&ps[0],N,self.get_buffer_size())

        else:

            self.per.ptr().set_priorities(index,N,self.get_buffer_size())
@@ -2485,7 +2496,8 @@ cdef class MPPrioritizedReplayBuffer(MPReplayBuffer): 
        self._unlock_explorer_per()



        for name, b in self.buffer.items():

            b[add_idx] = np.reshape(np.array(kwargs[name],copy=False,ndmin=2),

            b[add_idx] = np.reshape(np.array(kwargs[name], ndmin=2,

                                             copy=COPY_ONLY_NECESSARY),

                                    self.env_dict[name]["add_shape"])



        self._unlock_explorer()

pyproject.toml

+27 −2
Original line number Diff line number Diff line
@@ -18,21 +18,45 @@ dependencies = [ 
DEBUG_CPPRB = "1"



[tool.hatch.envs.test.scripts]

install = "pip install -e ."

run = "python -m unittest discover test"

run-cov = "coverage run -m unittest discover test"



[[tool.hatch.envs.test.matrix]]

python = ["3.8", "3.9", "3.10", "3.11", "3.12"]

python = ["3.9", "3.10", "3.11", "3.12"]

gym = ["gym", "gymnasium"]

numpy = ["numpy1", "numpy2"]



[tool.hatch.envs.test.overrides]

matrix.python.dependencies = [

    { value = "ray", if = ["3.8", "3.9", "3.10", "3.11"] },

    { value = "ray", if = ["3.9", "3.10", "3.11"] },

]

matrix.gym.dependencies = [

    { value = "gym", if = ["gym"] },

    { value = "gymnasium", if = ["gymnasium"] },

]

matrix.numpy.dependencies = [

    { value = "numpy<2.0.0", if = ["numpy1"] },

    { value = "numpy>=2.0.0rc2", if = ["numpy2"] },

]



[tool.hatch.envs.test-38]

template = "test"

python = "3.8"

extra-dependencies = [

    "numpy<2.0.0",

    "ray",

]



[[tool.hatch.envs.test-38.matrix]]

gym = ["gym", "gymnasium"]



[tool.hatch.envs.test-38.overrides]

matrix.gym.dependencies = [

    { value = "gym", if = ["gym"] },

    { value = "gymnasium", if = ["gymnasium"] },

]





[tool.hatch.envs.cov]

dependencies = [
@@ -77,6 +101,7 @@ extend-exclude = [ 
]



[tool.ruff.lint]

select = ["NPY201"]

ignore = [

  "PT009",  # We still use standard unittest

  "N999" ,  # This breaks API compatibility

run-test.sh

+6 −0
Original line number Diff line number Diff line
@@ -3,7 +3,13 @@ 
set -eux



hatch run cov:erase



hatch run test:install

hatch run test-38:install



hatch run test:run-cov

hatch run test-38:run-cov



hatch run cov:combine

hatch run cov:report