Release Pre-release alpha v0.2.2 · h2oai/datatable

Ability to write DataTable into a CSV file: the .to_csv() method. The CSV writer
is multi-threaded and extremely fast.
Added .internal.column(i).data_pointer getter, to allow native code from other
libraries to easily access the data in each column.
Fread can now read hexadecimal floating-point numbers: floats and doubles.
Csv writer will now auto-quote an empty string, and a string containing leading/
trailing whitespace, so that it can be read by fread reliably.
Fread now prints file sizes in "human-readable" form, i.e. KB/MB/GB instead of bytes.
Fread can now understand a variety of "NaN" / "Inf" literals produced by different
systems.
Add option hex to csv writer, which controls whether floats will be written in
decimal (default) or hexadecimal format.
Csv writer now uses the "dragonfly" algorithm for writing doubles, which is faster
than all known alternatives.
It is now allowed to pass a single-row numpy array as an argument to dt(rows=...),
which will be treated the same as if it was a single-column array.
Now datatable's wheel will include libraries libomp and libc++ on the platforms
where they are not widely available.
New fread's argument logger allows the user to supply custom logging mechanism to
fread. When this argument is provided, "verbose" mode is turned on automatically.

datatable will no longer attempt to distinguish between NA and NAN floating-point values.
Constructing DataTable from a 2D numpy array now preserves shape of that array. At the same
time it is no longer true that arr.tolist() == numpy.array(DataTable(arr)).tolist(): the
list will be transposed.
Converting a DataTable into a numpy array now also preserves shape. At the same time it is
no longer true that dt.topython() == dt.tonumpy().tolist(): the list will be transposed.
The internal _datatable module was moved to datatable.lib._datatable.

datatable will now convert huge integers into double inf values instead of raising an exception.

Provide feedback