Add pcodec

.github/workflows/ci-linux.yaml

@@ -50,7 +50,7 @@ jobs:
         run: |
           conda activate env
           export DISABLE_NUMCODECS_AVX2=""
-          python -m pip install -v -e .[test,test_extras,msgpack,zfpy]
+          python -m pip install -v -e .[test,test_extras,msgpack,zfpy,pcodec]
 
       - name: List installed packages
         shell: "bash -l {0}"

.github/workflows/ci-osx.yaml

@@ -50,7 +50,7 @@ jobs:
         run: |
           conda activate env
           export DISABLE_NUMCODECS_AVX2=""
-          python -m pip install -v -e .[test,test_extras,msgpack,zfpy]
+          python -m pip install -v -e .[test,test_extras,msgpack,zfpy,pcodec]
 
       - name: List installed packages
         shell: "bash -l {0}"

.github/workflows/ci-windows.yaml

@@ -42,7 +42,7 @@ jobs:
         shell: "bash -l {0}"
         run: |
           conda activate env
-          python -m pip install -v -e .[test,test_extras,msgpack,zfpy]
+          python -m pip install -v -e .[test,test_extras,msgpack,zfpy,pcodec]
 
       - name: List installed packages
         shell: "bash -l {0}"

docs/conf.py

@@ -23,7 +23,7 @@ def __getattr__(cls, name):
         return Mock()
 
 
-MOCK_MODULES = ['msgpack']
+MOCK_MODULES = ['msgpack', 'pcodec']
 sys.modules.update((mod_name, Mock()) for mod_name in MOCK_MODULES)
 
 

docs/index.rst

@@ -67,6 +67,7 @@ Contents
     abc
     registry
     blosc
+    pcodec
     lz4
     zfpy
     zstd

docs/pcodec.rst

@@ -0,0 +1,10 @@
+PCodec
+======
+
+.. automodule:: numcodecs.pcodec
+
+.. autoclass:: PCodec
+
+    .. autoattribute:: codec_id
+    .. automethod:: encode
+    .. automethod:: decode

docs/release.rst

@@ -14,6 +14,8 @@ Unreleased
 Enhancements
 ~~~~~~~~~~~~
 
+* Add PCodec
+  By :user:`Ryan Abernathey <rabernat>`.
 * Use PyData theme for docs
   By :user:`John Kirkham <jakirkham>`, :issue:`485`.
 

fixture/pcodec/codec.00/config.json

@@ -0,0 +1,8 @@
+{
+    "delta_encoding_order": null,
+    "equal_pages_up_to": 262144,
+    "float_mult_spec": "enabled",
+    "id": "pcodec",
+    "int_mult_spec": "enabled",
+    "level": 8
+}

fixture/pcodec/codec.01/config.json

@@ -0,0 +1,8 @@
+{
+    "delta_encoding_order": null,
+    "equal_pages_up_to": 262144,
+    "float_mult_spec": "enabled",
+    "id": "pcodec",
+    "int_mult_spec": "enabled",
+    "level": 1
+}

fixture/pcodec/codec.02/config.json

@@ -0,0 +1,8 @@
+{
+    "delta_encoding_order": null,
+    "equal_pages_up_to": 262144,
+    "float_mult_spec": "enabled",
+    "id": "pcodec",
+    "int_mult_spec": "enabled",
+    "level": 5
+}

fixture/pcodec/codec.03/config.json

@@ -0,0 +1,8 @@
+{
+    "delta_encoding_order": null,
+    "equal_pages_up_to": 262144,
+    "float_mult_spec": "enabled",
+    "id": "pcodec",
+    "int_mult_spec": "enabled",
+    "level": 9
+}

fixture/pcodec/codec.04/config.json

@@ -0,0 +1,8 @@
+{
+    "delta_encoding_order": null,
+    "equal_pages_up_to": 262144,
+    "float_mult_spec": "disabled",
+    "id": "pcodec",
+    "int_mult_spec": "disabled",
+    "level": 8
+}

fixture/pcodec/codec.05/config.json

@@ -0,0 +1,8 @@
+{
+    "delta_encoding_order": null,
+    "equal_pages_up_to": 300,
+    "float_mult_spec": "enabled",
+    "id": "pcodec",
+    "int_mult_spec": "enabled",
+    "level": 8
+}

numcodecs/__init__.py

@@ -115,3 +115,6 @@
 
 from numcodecs.fletcher32 import Fletcher32
 register_codec(Fletcher32)
+
+from numcodecs.pcodec import PCodec
+register_codec(PCodec)

numcodecs/pcodec.py

@@ -0,0 +1,89 @@
+from typing import Optional, Literal
+
+import numcodecs
+import numcodecs.abc
+from numcodecs.compat import ensure_contiguous_ndarray
+
+try:
+    from pcodec import standalone, ChunkConfig, PagingSpec
+except ImportError:  # pragma: no cover
+    standalone = None
+
+
+DEFAULT_MAX_PAGE_N = 262144
+
+
+class PCodec(numcodecs.abc.Codec):
+    """
+    PCodec (or pco, pronounced "pico") losslessly compresses and decompresses
+    numerical sequences with high compression ratio and fast speed.
+
+    See `PCodec Repo <https://github.com/mwlon/pcodec>`_ for more information.
+
+    PCodec supports only the following numerical dtypes: uint32, unit64, int32,
+    int64, float32, and float64.
+
+    Parameters
+    ----------
+    level : int
+        A compression level from 0-12, where 12 take the longest and compresses
+        the most.
+    delta_encoding_order : init or None
+        Either a delta encoding level from 0-7 or None. If set to None, pcodec
+        will try to infer the optimal delta encoding order.
+    int_mult_spec : {'enabled', 'disabled'}
+        If enabled, pcodec will consider using int mult mode, which can
+        substantially improve compression ratio but decrease speed in some cases
+        for integer types.
+    float_mult_spec : {'enabled', 'disabled'}
+        If enabled, pcodec will consider using float mult mode, which can
+        substantially improve compression ratio but decrease speed in some cases
+        for float types.
+    equal_pages_up_to : int
+        Divide the chunk into equal pages of up to this many numbers.
+    """
+
+    codec_id = "pcodec"
+
+    def __init__(
+        self,
+        level: int = 8,
+        delta_encoding_order: Optional[int] = None,
+        int_mult_spec: Literal["enabled", "disabled"] = "enabled",
+        float_mult_spec: Literal["enabled", "disabled"] = "enabled",
+        equal_pages_up_to: int = 262144
+    ):
+        if standalone is None:  # pragma: no cover
+            raise ImportError(
+                "pcodec must be installed to use the PCodec codec."
+            )
+
+        # note that we use `level` instead of `compression_level` to
+        # match other codecs
+        self.level = level
+        self.delta_encoding_order = delta_encoding_order
+        self.int_mult_spec = int_mult_spec
+        self.float_mult_spec = float_mult_spec
+        self.equal_pages_up_to = equal_pages_up_to
+
+    def encode(self, buf):
+        buf = ensure_contiguous_ndarray(buf)
+
+        paging_spec = PagingSpec.equal_pages_up_to(self.equal_pages_up_to)
+
+        config = ChunkConfig(
+            compression_level=self.level,
+            delta_encoding_order=self.delta_encoding_order,
+            int_mult_spec=self.int_mult_spec,
+            float_mult_spec=self.float_mult_spec,
+            paging_spec=paging_spec,
+        )
+        return standalone.simple_compress(buf, config)
+
+    def decode(self, buf, out=None):
+        if out is not None:
+            out = ensure_contiguous_ndarray(out)
+            standalone.simple_decompress_into(buf, out)
+            return out
+        else:
+            return standalone.simple_decompress(buf)

numcodecs/tests/common.py

@@ -217,6 +217,17 @@ def check_encode_decode_array(arr, codec):
     assert_array_items_equal(arr, dec)
 
 
+def check_encode_decode_array_to_bytes(arr, codec):
+
+    enc = codec.encode(arr)
+    dec = codec.decode(enc)
+    assert_array_items_equal(arr, dec)
+
+    out = np.empty_like(arr)
+    codec.decode(enc, out=out)
+    assert_array_items_equal(arr, out)
+
+
 def check_config(codec):
     config = codec.get_config()
     # round-trip through JSON to check serialization

numcodecs/tests/test_pcodec.py

@@ -0,0 +1,78 @@
+import pytest
+import numpy as np
+
+from numcodecs.pcodec import PCodec
+
+try:
+    # initializing codec triggers ImportError
+    PCodec()
+except ImportError:  # pragma: no cover
+    pytest.skip(
+        "pcodec not available", allow_module_level=True
+    )
+
+from numcodecs.tests.common import (
+    check_encode_decode_array_to_bytes,
+    check_config,
+    check_repr,
+    check_backwards_compatibility,
+    check_err_decode_object_buffer,
+    check_err_encode_object_buffer,
+)
+
+
+codecs = [
+    PCodec(),
+    PCodec(level=1),
+    PCodec(level=5),
+    PCodec(level=9),
+    PCodec(float_mult_spec="disabled", int_mult_spec="disabled"),
+    PCodec(equal_pages_up_to=300),
+]
+
+
+# mix of dtypes: integer, float
+# mix of shapes: 1D, 2D
+# mix of orders: C, F
+arrays = [
+    np.arange(1000, dtype="u4"),
+    np.arange(1000, dtype="u8"),
+    np.arange(1000, dtype="i4"),
+    np.arange(1000, dtype="i8"),
+    np.linspace(1000, 1001, 1000, dtype="f4"),
+    np.linspace(1000, 1001, 1000, dtype="f8"),
+    np.random.normal(loc=1000, scale=1, size=(100, 10)),
+    np.asfortranarray(np.random.normal(loc=1000, scale=1, size=(100, 10))),
+    np.random.randint(0, 2**60, size=1000, dtype="u8"),
+    np.random.randint(-(2**63), -(2**63) + 20, size=1000, dtype="i8"),
+]
+
+
+@pytest.mark.parametrize("arr", arrays)
+@pytest.mark.parametrize("codec", codecs)
+def test_encode_decode(arr, codec):
+    check_encode_decode_array_to_bytes(arr, codec)
+
+
+def test_config():
+    codec = PCodec(level=3)
+    check_config(codec)
+
+
+def test_repr():
+    check_repr(
+        "PCodec(delta_encoding_order=None, equal_pages_up_to=262144, float_mult_spec='enabled', "
+        "int_mult_spec='enabled', level=3)"
+    )
+
+
+def test_backwards_compatibility():
+    check_backwards_compatibility(PCodec.codec_id, arrays, codecs)
+
+
+def test_err_decode_object_buffer():
+    check_err_decode_object_buffer(PCodec())
+
+
+def test_err_encode_object_buffer():
+    check_err_encode_object_buffer(PCodec())

pyproject.toml

@@ -66,6 +66,9 @@ msgpack = [
 zfpy = [
     "zfpy>=1.0.0",
 ]
+pcodec = [
+    "pcodec>=0.1.0",
+]
 
 [tool.setuptools]
 license-files = ["LICENSE.txt"]