Add minimal CQM implementation #839
Conversation
|
Still need to fix the windows bug obviously. This also will need a pretty serious second pass when we add |
arcondello
force-pushed
the
feature/constrained
branch
from
3b0a614 to
d15798e
Compare
Codecov Report
@@ Coverage Diff @@
## main #839 +/- ##
==========================================
+ Coverage 88.26% 88.52% +0.25%
==========================================
Files 70 71 +1
Lines 5862 6046 +184
==========================================
+ Hits 5174 5352 +178
- Misses 688 694 +6
Continue to review full report at Codecov.
|
| header_data = json.dumps(data, sort_keys=True).encode('ascii') | ||
| header_data += b'\n' | ||
| header_data += b' '*(64 - (len(prefix) + 6 + len(header_data)) % 64) | ||
| header_len = np.dtype('<u4').type(len(header_data)).tobytes() |
There was a problem hiding this comment.
I did just a quick check, but it looks .tobytes() in this setup actually always uses native byteorder. I believe your intention was to use little endian.
There was a problem hiding this comment.
Yup, excellent catch
In [6]: np.frombuffer(np.dtype('<u4').type(256).tobytes(), '<u4')
Out[6]: array([256], dtype=uint32)
In [7]: np.frombuffer(np.dtype('>u4').type(256).tobytes(), '<u4')
Out[7]: array([256], dtype=uint32)
In [8]: np.frombuffer(np.dtype('>u4').type(256).tobytes(), '>u4')
Out[8]: array([65536], dtype=uint32)
In [9]: np.frombuffer(np.dtype('<u4').type(256).tobytes(), '>u4')
Out[9]: array([65536], dtype=uint32)
There was a problem hiding this comment.
I see we have the same bug in BQM and DQM.
There was a problem hiding this comment.
Indeed. Investigating
There was a problem hiding this comment.
Have a fix I'll make in a followup PR.
New Features ------------ * Add `add_bqm` method to C++ BinaryQuadraticModel #821, #823 * Add `Structured.valid_bqm_graph` method for verifying input problem structure #832 * Reimplement `BinaryQuadraticModel` to use new C++ code #828 * `BinaryQuadraticModel` can now be manipulated symbolically #834 * `load` function can now load all model types #841, #843 * `DiscreteQuadraticModel` now has an `.offset` attribute #838 * Add `ConstrainedQuadraticModel` class #839 Fix --- * Fix type promotions in binary quadratic models with object biases #836
Allows the construction of binary constrained quadratic models.