Changelogs » Arkouda

Arkouda

0.0.9

------------------------
* more optimized groupby/aggregate using segmented scans
* fixed in1d bug
* added in1d unit test
* added groupby unit test
* refactored a bunch of code
* doubled speed of binary operators
* much faster sort added
* multi-level groupby added
* many other changes lost track of them all

0.0.8

------------------------
* added initial ak.argsort functionality
* added initial ak.is_sorted functionality
* added unit test to tests directory for ak.argsort
* optimized argsort
* added ak.where
* opt some a[pdarray] gather-indexing
* added initial ak.GroupBy and ak.aggregate functionality
* added optimized per-locale version of groupby and aggregate

0.0.7

------------------------
* fixed multi-file/multi-threaded bug in HDF5 I/O
* improved HDF5 I/O error handling + glob filenames
* added initial versions of ak.array() and pdarray.to_ndarray()
* refactored histogram and some other procedures using helper nested procedures
* added more operators for bool
* added min and max reductions
* optimized ak.int1d()

0.0.6

------------------------
* added toys/ subdir with some prototype sorting code
* made some changes to make naming consistent in Python and Chapel code
* added initial HDF5 I/O
* added a[slice] = value indexing
* added a[slice] = pdarray indexing
* added initial ak.argmin/ak.argmax functionality
* added some different optimizations and optional return of the counts from ak.unique
* added more error handling to GenSymIO (HDF5 I/O)

0.0.5

------------------------
* finally got a BlockDist to work properly -- thanks Brad ;-)
* changed ak.startup() to ak.connect() also added ak.disconnect()
* added initial version of value_counts
* added a[pdarray]=value indexing
* changed to bool_ = type(True) in arkouda.py
* added a[pdarray]=pdarray indexing
* compiled with chapel 1.19 and cleaned up compilation warnings
* substituted domain.contains() for domain.member()
* fixed up parallel scans on bool arrays with int cast through a copy to int array (blah!)
* created github repo
* added backward compat code for 1.18
* added some prototype python arkouda/numpy check/test code

0.0.4

------------------------
* added threshold to pdarray.__iter__() to limit comms with arkouda_server
* refactored how dist domain mapped arrays for symbol table entries
* added some more tests to ak_test.py
* simplified ones_like and zeros_like in arkouda.py
* added bitwise {&,|,>>,<<,^,~} operators to int64
* added unary-
* added initial version of histogram
* added bounds checking on pdarray indexing "a[iv]" where iv is an int64 pdarray
* moved all operator stuff into OperatorMgs.chpl
* factored the code put config const/param into ServerConfig.chpl module
* added initital version of in1d
* added initial version of randint
* cleaned up ValueError vs. TypeError in arkouda.py
* indexing operations moved to IndexingMsg.chpl
* added initial version of unique

0.0.3

------------------------
* changed over to distributed domain mapped arrays for arkouda server
* added stdout.flush() after verbose messages to overcome buffering on big runs
* added timer to report processing time around main loop
* a little optimization of arange and linspace

0.0.2

------------------------
* fixed bug in linspace in MsgProcessing.chpl
* changed to using DType in all GenSymOps.chpl module from dynamic casts
* simplified messaging in hpda.py
* refactored binary op overloading in pdarray class
* overloaded pdarray __str__ and __repr__, server messages and support also
* added server error prop to raise exception in python
* refactored errors in MsgProcessing.chpl
* changed name from hpda to arkouda (Greek for Bear)
* fixed "/ and /= aka __truediv__" to return float64 and added "// and //= aka __floordiv__"
* fixed ones and zeros to behave like numpy
* added slices to __getitem__ and it does negative strides now!
* added bool dtype
* added relops
* server gives out names instead of python
* added sum, prod, any, and all reductions
* added elemental functions log, exp, and abs
* add numpy scans like cumsum and cumprod
* add pdarray indexing b[] = a[iv[]],
iv[] can be either int64 which does gather or bool which does a compress-out-false-indicies