[jira] [Commented] (ARROW-2172) [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
[
https://issues.apache.org/jira/browse/ARROW-2172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16370558#comment-16370558
]
ASF GitHub Bot commented on ARROW-2172:
---
xhochy closed pull request #1628: ARROW-2172: [C++/Python] Fix converting from
Numpy array with non-natural stride
URL: https://github.com/apache/arrow/pull/1628
This is a PR merged from a forked repository.
As GitHub hides the original diff on merge, it is displayed below for
the sake of provenance:
As this is a foreign pull request (from a fork), the diff is supplied
below (as it won't show otherwise due to GitHub magic):
diff --git a/cpp/src/arrow/python/numpy_to_arrow.cc
b/cpp/src/arrow/python/numpy_to_arrow.cc
index 522bf5174..23418ad92 100644
--- a/cpp/src/arrow/python/numpy_to_arrow.cc
+++ b/cpp/src/arrow/python/numpy_to_arrow.cc
@@ -554,12 +554,22 @@ Status StaticCastBuffer(const Buffer& input, const
int64_t length, MemoryPool* p
return Status::OK();
}
-template
-void CopyStrided(T* input_data, int64_t length, int64_t stride, T2*
output_data) {
+template
+void CopyStridedBytewise(int8_t* input_data, int64_t length, int64_t stride,
+ T* output_data) {
+ // Passing input_data as non-const is a concession to PyObject*
+ for (int64_t i = 0; i < length; ++i) {
+memcpy(output_data + i, input_data, sizeof(T));
+input_data += stride;
+ }
+}
+
+template
+void CopyStridedNatural(T* input_data, int64_t length, int64_t stride, T*
output_data) {
// Passing input_data as non-const is a concession to PyObject*
int64_t j = 0;
for (int64_t i = 0; i < length; ++i) {
-output_data[i] = static_cast(input_data[j]);
+output_data[i] = input_data[j];
j += stride;
}
}
@@ -571,13 +581,19 @@ Status CopyStridedArray(PyArrayObject* arr, const int64_t
length, MemoryPool* po
using T = typename traits::T;
// Strided, must copy into new contiguous memory
- const int64_t stride = PyArray_STRIDES(arr)[0];
- const int64_t stride_elements = stride / sizeof(T);
-
auto new_buffer = std::make_shared(pool);
RETURN_NOT_OK(new_buffer->Resize(sizeof(T) * length));
- CopyStrided(reinterpret_cast(PyArray_DATA(arr)), length, stride_elements,
- reinterpret_cast(new_buffer->mutable_data()));
+
+ const int64_t stride = PyArray_STRIDES(arr)[0];
+ if (stride % sizeof(T) == 0) {
+const int64_t stride_elements = stride / sizeof(T);
+CopyStridedNatural(reinterpret_cast(PyArray_DATA(arr)), length,
stride_elements,
+ reinterpret_cast(new_buffer->mutable_data()));
+ } else {
+CopyStridedBytewise(reinterpret_cast(PyArray_DATA(arr)), length,
stride,
+reinterpret_cast(new_buffer->mutable_data()));
+ }
+
*out = new_buffer;
return Status::OK();
}
diff --git a/python/pyarrow/tests/test_convert_pandas.py
b/python/pyarrow/tests/test_convert_pandas.py
index 95137ffb2..9bd3c77d2 100644
--- a/python/pyarrow/tests/test_convert_pandas.py
+++ b/python/pyarrow/tests/test_convert_pandas.py
@@ -1582,6 +1582,19 @@ def test_convert_empty_table(self):
arr = pa.array([], type=pa.struct([pa.field('a', pa.int64())]))
tm.assert_almost_equal(arr.to_pandas(), np.array([], dtype=object))
+def test_non_natural_stride(self):
+"""
+ARROW-2172: converting from a Numpy array with a stride that's
+not a multiple of itemsize.
+"""
+dtype = np.dtype([('x', np.int32), ('y', np.int16)])
+data = np.array([(42, -1), (-43, 2)], dtype=dtype)
+assert data.strides == (6,)
+arr = pa.array(data['x'], type=pa.int32())
+assert arr.to_pylist() == [42, -43]
+arr = pa.array(data['y'], type=pa.int16())
+assert arr.to_pylist() == [-1, 2]
+
def _fully_loaded_dataframe_example():
from distutils.version import LooseVersion
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
> --
>
> Key: ARROW-2172
> URL: https://issues.apache.org/jira/browse/ARROW-2172
> Project: Apache Arrow
> Issue Type: Bug
> Components: Python
>Affects Versions: 0.8.0
>Reporter: Antoine Pitrou
>Priority: Major
> Labels: pull-request-available
> Fix For: 0.9.0
>
>
> In the example below, the input array has a stride that's not a multiple of
> the itemsize:
> {code:python}
> >>> data = np.array([(42, True), (43, False)],
> ...:
[jira] [Commented] (ARROW-2172) [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
[
https://issues.apache.org/jira/browse/ARROW-2172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16370107#comment-16370107
]
ASF GitHub Bot commented on ARROW-2172:
---
cpcloud commented on a change in pull request #1628: ARROW-2172: [C++/Python]
Fix converting from Numpy array with non-natural stride
URL: https://github.com/apache/arrow/pull/1628#discussion_r169335662
##
File path: cpp/src/arrow/python/numpy_to_arrow.cc
##
@@ -554,12 +554,22 @@ Status StaticCastBuffer(const Buffer& input, const
int64_t length, MemoryPool* p
return Status::OK();
}
-template
-void CopyStrided(T* input_data, int64_t length, int64_t stride, T2*
output_data) {
+template
+void CopyStridedBytewise(int8_t* input_data, int64_t length, int64_t stride,
+ T* output_data) {
+ // Passing input_data as non-const is a concession to PyObject*
+ for (int64_t i = 0; i < length; ++i) {
+memcpy(output_data + i, input_data, sizeof(T));
+input_data += stride;
Review comment:
Fair enough. LGTM
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
> --
>
> Key: ARROW-2172
> URL: https://issues.apache.org/jira/browse/ARROW-2172
> Project: Apache Arrow
> Issue Type: Bug
> Components: Python
>Affects Versions: 0.8.0
>Reporter: Antoine Pitrou
>Priority: Major
> Labels: pull-request-available
>
> In the example below, the input array has a stride that's not a multiple of
> the itemsize:
> {code:python}
> >>> data = np.array([(42, True), (43, False)],
> ...:dtype=[('x', np.int32), ('y', np.bool_)])
> ...:
> ...:
> >>> data['x']
> array([42, 43], dtype=int32)
> >>> pa.array(data['x'], type=pa.int32())
>
> [
> 42,
> 11009
> ]
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
[jira] [Commented] (ARROW-2172) [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
[
https://issues.apache.org/jira/browse/ARROW-2172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16369757#comment-16369757
]
ASF GitHub Bot commented on ARROW-2172:
---
pitrou commented on a change in pull request #1628: ARROW-2172: [C++/Python]
Fix converting from Numpy array with non-natural stride
URL: https://github.com/apache/arrow/pull/1628#discussion_r169225590
##
File path: cpp/src/arrow/python/numpy_to_arrow.cc
##
@@ -554,12 +554,22 @@ Status StaticCastBuffer(const Buffer& input, const
int64_t length, MemoryPool* p
return Status::OK();
}
-template
-void CopyStrided(T* input_data, int64_t length, int64_t stride, T2*
output_data) {
+template
+void CopyStridedBytewise(int8_t* input_data, int64_t length, int64_t stride,
+ T* output_data) {
+ // Passing input_data as non-const is a concession to PyObject*
+ for (int64_t i = 0; i < length; ++i) {
+memcpy(output_data + i, input_data, sizeof(T));
+input_data += stride;
Review comment:
Well, the current code actually works as intended even with such arrays:
```
>>> base = np.arange(8, dtype=np.int8).view(np.int32)
>>> arr = np.lib.stride_tricks.as_strided(base, shape=(5,), strides=(1,))
>>> arr
array([ 50462976, 67305985, 84148994, 100992003, 117835012], dtype=int32)
>>> arr.tobytes()
b'\x00\x01\x02\x03\x01\x02\x03\x04\x02\x03\x04\x05\x03\x04\x05\x06\x04\x05\x06\x07'
>>> pa.array(arr, type=pa.int32())
[
50462976,
67305985,
84148994,
100992003,
117835012
]
```
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
> --
>
> Key: ARROW-2172
> URL: https://issues.apache.org/jira/browse/ARROW-2172
> Project: Apache Arrow
> Issue Type: Bug
> Components: Python
>Affects Versions: 0.8.0
>Reporter: Antoine Pitrou
>Priority: Major
> Labels: pull-request-available
>
> In the example below, the input array has a stride that's not a multiple of
> the itemsize:
> {code:python}
> >>> data = np.array([(42, True), (43, False)],
> ...:dtype=[('x', np.int32), ('y', np.bool_)])
> ...:
> ...:
> >>> data['x']
> array([42, 43], dtype=int32)
> >>> pa.array(data['x'], type=pa.int32())
>
> [
> 42,
> 11009
> ]
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
[jira] [Commented] (ARROW-2172) [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
[
https://issues.apache.org/jira/browse/ARROW-2172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16369753#comment-16369753
]
ASF GitHub Bot commented on ARROW-2172:
---
cpcloud commented on a change in pull request #1628: ARROW-2172: [C++/Python]
Fix converting from Numpy array with non-natural stride
URL: https://github.com/apache/arrow/pull/1628#discussion_r169224812
##
File path: cpp/src/arrow/python/numpy_to_arrow.cc
##
@@ -554,12 +554,22 @@ Status StaticCastBuffer(const Buffer& input, const
int64_t length, MemoryPool* p
return Status::OK();
}
-template
-void CopyStrided(T* input_data, int64_t length, int64_t stride, T2*
output_data) {
+template
+void CopyStridedBytewise(int8_t* input_data, int64_t length, int64_t stride,
+ T* output_data) {
+ // Passing input_data as non-const is a concession to PyObject*
+ for (int64_t i = 0; i < length; ++i) {
+memcpy(output_data + i, input_data, sizeof(T));
+input_data += stride;
Review comment:
No, but maybe add a DCHECK?
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
> --
>
> Key: ARROW-2172
> URL: https://issues.apache.org/jira/browse/ARROW-2172
> Project: Apache Arrow
> Issue Type: Bug
> Components: Python
>Affects Versions: 0.8.0
>Reporter: Antoine Pitrou
>Priority: Major
> Labels: pull-request-available
>
> In the example below, the input array has a stride that's not a multiple of
> the itemsize:
> {code:python}
> >>> data = np.array([(42, True), (43, False)],
> ...:dtype=[('x', np.int32), ('y', np.bool_)])
> ...:
> ...:
> >>> data['x']
> array([42, 43], dtype=int32)
> >>> pa.array(data['x'], type=pa.int32())
>
> [
> 42,
> 11009
> ]
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
[jira] [Commented] (ARROW-2172) [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
[
https://issues.apache.org/jira/browse/ARROW-2172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16369747#comment-16369747
]
ASF GitHub Bot commented on ARROW-2172:
---
pitrou commented on a change in pull request #1628: ARROW-2172: [C++/Python]
Fix converting from Numpy array with non-natural stride
URL: https://github.com/apache/arrow/pull/1628#discussion_r169223711
##
File path: cpp/src/arrow/python/numpy_to_arrow.cc
##
@@ -554,12 +554,22 @@ Status StaticCastBuffer(const Buffer& input, const
int64_t length, MemoryPool* p
return Status::OK();
}
-template
-void CopyStrided(T* input_data, int64_t length, int64_t stride, T2*
output_data) {
+template
+void CopyStridedBytewise(int8_t* input_data, int64_t length, int64_t stride,
+ T* output_data) {
+ // Passing input_data as non-const is a concession to PyObject*
+ for (int64_t i = 0; i < length; ++i) {
+memcpy(output_data + i, input_data, sizeof(T));
+input_data += stride;
Review comment:
In normal use, it probably is, but you can create weird arrays using
`stride_tricks`:
https://docs.scipy.org/doc/numpy/reference/generated/numpy.lib.stride_tricks.as_strided.html
That said, I don't think it makes a problem here, does it?
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
> --
>
> Key: ARROW-2172
> URL: https://issues.apache.org/jira/browse/ARROW-2172
> Project: Apache Arrow
> Issue Type: Bug
> Components: Python
>Affects Versions: 0.8.0
>Reporter: Antoine Pitrou
>Priority: Major
> Labels: pull-request-available
>
> In the example below, the input array has a stride that's not a multiple of
> the itemsize:
> {code:python}
> >>> data = np.array([(42, True), (43, False)],
> ...:dtype=[('x', np.int32), ('y', np.bool_)])
> ...:
> ...:
> >>> data['x']
> array([42, 43], dtype=int32)
> >>> pa.array(data['x'], type=pa.int32())
>
> [
> 42,
> 11009
> ]
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
[jira] [Commented] (ARROW-2172) [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
[
https://issues.apache.org/jira/browse/ARROW-2172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16369554#comment-16369554
]
ASF GitHub Bot commented on ARROW-2172:
---
cpcloud commented on a change in pull request #1628: ARROW-2172: [C++/Python]
Fix converting from Numpy array with non-natural stride
URL: https://github.com/apache/arrow/pull/1628#discussion_r169177655
##
File path: cpp/src/arrow/python/numpy_to_arrow.cc
##
@@ -554,12 +554,22 @@ Status StaticCastBuffer(const Buffer& input, const
int64_t length, MemoryPool* p
return Status::OK();
}
-template
-void CopyStrided(T* input_data, int64_t length, int64_t stride, T2*
output_data) {
+template
+void CopyStridedBytewise(int8_t* input_data, int64_t length, int64_t stride,
+ T* output_data) {
+ // Passing input_data as non-const is a concession to PyObject*
+ for (int64_t i = 0; i < length; ++i) {
+memcpy(output_data + i, input_data, sizeof(T));
+input_data += stride;
Review comment:
Is `stride >= sizeof(T)` guaranteed?
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
> --
>
> Key: ARROW-2172
> URL: https://issues.apache.org/jira/browse/ARROW-2172
> Project: Apache Arrow
> Issue Type: Bug
> Components: Python
>Affects Versions: 0.8.0
>Reporter: Antoine Pitrou
>Priority: Major
> Labels: pull-request-available
>
> In the example below, the input array has a stride that's not a multiple of
> the itemsize:
> {code:python}
> >>> data = np.array([(42, True), (43, False)],
> ...:dtype=[('x', np.int32), ('y', np.bool_)])
> ...:
> ...:
> >>> data['x']
> array([42, 43], dtype=int32)
> >>> pa.array(data['x'], type=pa.int32())
>
> [
> 42,
> 11009
> ]
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
[jira] [Commented] (ARROW-2172) [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
[
https://issues.apache.org/jira/browse/ARROW-2172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16369218#comment-16369218
]
ASF GitHub Bot commented on ARROW-2172:
---
pitrou commented on issue #1628: ARROW-2172: [C++/Python] Fix converting from
Numpy array with non-natural stride
URL: https://github.com/apache/arrow/pull/1628#issuecomment-366720200
AppVeyor build at https://ci.appveyor.com/project/pitrou/arrow/build/1.0.96
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
> --
>
> Key: ARROW-2172
> URL: https://issues.apache.org/jira/browse/ARROW-2172
> Project: Apache Arrow
> Issue Type: Bug
> Components: Python
>Affects Versions: 0.8.0
>Reporter: Antoine Pitrou
>Priority: Major
> Labels: pull-request-available
>
> In the example below, the input array has a stride that's not a multiple of
> the itemsize:
> {code:python}
> >>> data = np.array([(42, True), (43, False)],
> ...:dtype=[('x', np.int32), ('y', np.bool_)])
> ...:
> ...:
> >>> data['x']
> array([42, 43], dtype=int32)
> >>> pa.array(data['x'], type=pa.int32())
>
> [
> 42,
> 11009
> ]
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
[jira] [Commented] (ARROW-2172) [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
[
https://issues.apache.org/jira/browse/ARROW-2172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16369209#comment-16369209
]
ASF GitHub Bot commented on ARROW-2172:
---
pitrou opened a new pull request #1628: ARROW-2172: [C++/Python] Fix converting
from Numpy array with non-natural stride
URL: https://github.com/apache/arrow/pull/1628
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> [Python] Incorrect conversion from Numpy array when stride % itemsize != 0
> --
>
> Key: ARROW-2172
> URL: https://issues.apache.org/jira/browse/ARROW-2172
> Project: Apache Arrow
> Issue Type: Bug
> Components: Python
>Affects Versions: 0.8.0
>Reporter: Antoine Pitrou
>Priority: Major
> Labels: pull-request-available
>
> In the example below, the input array has a stride that's not a multiple of
> the itemsize:
> {code:python}
> >>> data = np.array([(42, True), (43, False)],
> ...:dtype=[('x', np.int32), ('y', np.bool_)])
> ...:
> ...:
> >>> data['x']
> array([42, 43], dtype=int32)
> >>> pa.array(data['x'], type=pa.int32())
>
> [
> 42,
> 11009
> ]
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
