3.1.6

https://code.kx.com/pykx/3.1/release-notes/changelog.html

3.1.5

https://code.kx.com/pykx/3.1/release-notes/changelog.html

3.1.4

https://code.kx.com/pykx/3.1/release-notes/changelog.html

3.1.3

https://code.kx.com/pykx/3.1/release-notes/changelog.html

3.1.2

The release notes below cover both the 3.1.2 and 3.1.1 releases of PyKX owing to the feature merge containing the changes for both

PyKX 3.1.1

Release Date

2025-02-14

Fixes and Improvements

• Fixed issue whereby PyKX would prompt for user inputs if a license was not found in a non-interactive session, now correctly falls back to unlicensed mode.
• Fix throwing of errors for .pykx.safeReimport, now throws an error instead of returning a value.

PyKX 3.1.2

Release Date

2025-03-18

Fixes and Improvements

• Fixes an issue when using PYKX_THREADING that could cause a segfault in some scenarios.

3.1.0

📦 PyKX 3.1.0 Released 📦

Full up-to-date release notes are available here.

🎉 Feature 🎉

• Added support for Python 3.13.
• Added support for NumPy 2.0
• Added support for the creation and management of splayed format databases when using the pykx.DB class.
• Addition of .copy() method for all pykx objects allowing users to modify copied objects without interfering with the original object.
• Addition of reshape keyword to the .np() method of kx.List objects. This can provide two benefits:
  a. Conversions of kx.List objects to NumPy by default produce an array of NumPy arrays rather than an N-Dimensional NumPy array. Setting reshape=True when handling N-Dimensional rectangular lists allows the shape to be pre-processed prior to conversion and a more natural N-Dimensional NumPy array to be generated.

>>> kx.q('2 2#4?1f').np(reshape=True)
array([[0.94997503, 0.43908099],
       [0.57590514, 0.59190043]])

b. Provide a performance boost when converting regularly shaped (rectangular) N-Dimensional lists of uniform type when the shape of the resulting numpy array is known prior to conversion

>>> import pykx as kx
>>> lst = kx.q('100000 100 10#100000000?1f')
>>> %timeit lst.np()
9.72 s ± 272 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
>>> %timeit lst.np(reshape=[100000, 100, 10])
883ms ± 19.8 ms per loop (mean ± std. dev. of 7 runs, 1 loops each)

• Added async_response keyword argument when calling AsyncQConnection objects with reuse=False and wait=False to allow keeping the connection alive until an asynchronous response message has been received.

🔧 Fixes & Improvements 🔧

• Application of the str function on empty PyKX objects could return unexpected results

>>> import pykx as kx
>>> str(kx.q('()!()'))
'()!()'
>>> str(kx.q('()'))
'()'

• When a client attempted to retrieve an object which could not be serialized from a PyKX server it resulted in the client process hanging. An appropriate error is now sent to the client.
• Database generation functionality now allows users to pass any data-type which will convert to a pykx.Table as the table parameter
• Attempting to create a partitioned databases/add a partition to a database with a sym_enum keyword but no by_field would result in a KeyError.
• Using math.inf or -math.inf when creating numeric values now creates equivalent PyKX types
• Fixed a bug where QFuture objects returned by AsyncQConnection objects could block each other unnecessarily.

🆕 Beta Features 🆕

Addition of support in Beta form for conversions between PyKX and PyTorch objects.

2.5.5

PyKX 2.5.5

Release Date

2024-11-28

Fixes and Improvements

• PyKX Pandas dependency has been raised to allow <=2.2.3 for Python>3.8
• PyKX Pandas dependency for Python 3.8 has been clamped to <2.0 due to support being dropped for it by Pandas after 2.0.3.

For release notes of all versions see here.

3.0.1

Full up-to-date release notes are available here.

🎉 Feature 🎉

• Addition of the property day to kx.Column objects to allow users to retrieve the day of month of a timestamp.

  >>> import pykx as kx
  >>> tab = kx.Table(data={
  ...     'a': kx.random.random(100, kx.TimestampAtom.inf),
  ...     'b': kx.random.random([100, 3], 10.0)
  ...     })
  >>> tab.exec(kx.Column('a').day)
  pykx.IntVector(pykx.q('7 10 12..'))

🔧 Fixes & Improvements 🔧

• Added warning to kx.q.system.load and context registration when attempting to load path with a space. Can be suppressed by enabling PYKX_SUPPRESS_WARNINGS.

• Changed %%python heading to %%py when calling Python code during jupyter_qfirst mode so as not to conflict with inbuilt Jupyter cell magics.

• Fixed kx.license.check(format='string') to remove newline characters during comparison.

• Configuration file .pykx-config now supports use of boolean toml configuration

  $ cat ~/.pykx-config
  [default]
  PYKX_QDEBUG = true
  $ python
  >>> import pykx as kx
  >>> kx.config.pykx_qdebug
  True

• Reintroduced unsetting/setting of PYKX_SKIP_UNDERQ in PyKXReimport these had been removed in the 3.0.0 release.

• Added type checking for the cast flag when calling kx.toq() or creating a kx.K variable such as kx.FloatVector() or kx.DatetimeAtom().

• Removed the need to enable PYKX_BETA_FEATURES to use pykx_threading.

• Fixed a memory leak when calling pickle.loads on a PyKX object which previously had been called with pickle.dumps.

• Removal of column type from the return of dtypes method for kx.Table objects, previously this had raised a deprecation warning

  >>> tab = kx.q('([] a:1 2 3j;b:4 5 6i)')
  >>> tab.dtypes
  pykx.Table(pykx.q('
  columns datatypes
  ---------------------
  a       "kx.LongAtom"
  b       "kx.IntAtom" 
  '))

3.0.0

Full up-to-date release notes are available here.

🎉 Major Features/Changes 🎉

• Addition of functionality to allow for development of end-to-end streaming workflows consisting of data-ingestion, persistence and query. This functionality is outlined in-depth here.
• Update to the PyKX Query API to support a significantly more Python first approach to querying kdb+ in-memory and on-disk databases.

>>> table = kx.Table(data={
...     'sym': kx.random.random(100, ['AAPL', 'GOOG', 'MSFT']),
...     'date': kx.random.random(100, kx.q('2022.01.01') + [0,1,2]),
...     'price': kx.random.random(100, 1000.0),
...     'size': kx.random.random(100, 100) 
... })
>>> table.select(columns=kx.Column('price').max(), where=kx.Column('size') > 5)
>>> table.update(column=kx.Column('price').wavg(kx.Column('size')).rename('vwap'), by=kx.Column('sym'))
>>> table.delete(column=kx.Column('sym'))
>>> table.update(column=(kx.Column('price') * kx.Column('size')).rename('total'))

Beta features available in the 2.* versions of PyKX have now been migrated to full support.
- The full list of these features are as follows:
- Database Creation and Management
- Compression and Encryption Module
- Remote Function Execution
- Streamlit Integration
- Multi-threaded use of PyKX

❓ What else? ❓

• Extension to our integration with Jupyter Notebooks by adding a q-first mode of operation which allows users working between the two languages to more easily automate workflows depending on both
• Addition of a new utility function kx.util.detect_bad_columns to validate if the columns of a table object conform to the naming conventions supported by kdb+ and highlighting if the table contains duplicate column names raising a warning indicating potential issues and returning True if the table contains invalid columns.
• When generating IPC connections with reconnection_attempts users can now configure the initial delay between first and second attempts and the function which updates the delay on successive attempts using the reconnection_delay and reconnection_function keywords. See here for a worked example.
• Two new options added on first initialisation of PyKX to allow users to:
  • Use the path to their already downloaded kc.lic/k4.lic licenses without going through the “Do you want to install a license” workflow
  • Allow users to persist for future use that they wish to use the IPC only unlicensed mode of PyKX, this will persist a file ~/.pykx-config which sets configuration denoting unlicensed mode is to be used.
• Addition of function kx.util.install_q to allow users who do not have access to a q executable at the time of installing PyKX. See here for instructions regarding its use.
  Addition of function kx.util.start_q_subprocess to allow a q process to be started on a specified port with supplied initialisation arguments

🔧 Fixes & Improvements 🔧

As with any release PyKX 3.0 provides a significant number of bug fixes and improvements, the following are a subset:

• Addition of support for help when interacting with q keywords and operators via PyKX
• Previously loading pykx.q during q startup using QINIT or QHOME/q.q resulted in a segfault or a corruption.
• The function kx.util.debug_environment now returns the applied configuration values at startup instead of customised values
• Operations on kx.GroupbyTable objects which have been indexed previously would raise an error indicating invalid key access
• Attempts to load a database using the kx.DB module previously would raise an nyi error if the path to the database contained a space

2.5.2

PyKX 2.5.2 has been released 🎉 Full release notes for client consumption can be found here.

Highlights:

Converting PyKX generic lists using the keyword parameter raw=True would previously return incorrect results, the values received being the memory address of the individual elements of the list, this has now been resolved:

>>> a = kx.q('(1; 3.4f; `asad; "asd")')
>>> a.np(raw=True)
array([1, 3.4, b'asad', b'asd'], dtype=object)

• Fix to issue where use of kx.SymbolAtom with getitem method on kx.Table objects would return a table rather then vector/list. The return now mirrors the expected return which matches str type inputs

>>> import pykx as kx
>>> tab = kx.Table(data={'x': [1, 2, 3], 'y': ['a', 'b', 'c']})
>>> tab['x']
pykx.LongVector(pykx.q('1 2 3'))
>>> tab[kx.SymbolAtom('x')]
pykx.LongVector(pykx.q('1 2 3'))

Fix to issue where loading PyKX on Windows from 2.5.0 could result in a users working directory being changed to site-packages/pykx

The full list including more fixes and improvements is available here.