Version 0.3.0¶

Version 0.3.0
Release date:	2018-03-19
Next release:	Version 0.3.1
Previous release:	Version 0.2.2
Wheels
MacOS	python-3.5

General¶

Method Frame.tonumpy() now has argument stype which will force conversion into a numpy array of the specific stype.
Enums stype and ltype that encapsulate the type-system of the datatable module.
It is now possible to fread from a bytes object.
Allow columns to be renamed by setting the names property on the frame.
Internal “MemoryMapManager” will make datatable more robust when opening a frame with many columns on Linux systems. In particular, error 12 “not enough memory” should become much more rare now.
Number of threads used by fread can now be controlled via parameter nthreads.
It is now possible to supply string argument to dt.DataTable constructor, which in turn will try to interpret that argument via fread().
fread() can now read compressed .xz files.
fread() now automatically skips Ctrl+Z / NUL characters at the end of the file.
It is now possible to create a datatable from string numpy array.
Added parameters skip_blank_lines, strip_white, quotechar and dec to fread().
Single-column files with blank lines can now be read successfully.
Fread now recognizes \r\r\n as a valid line ending.
Added parameters url and cmd to fread(), as well as ability to detect URLs automatically. The url parameter downloads file from HTTP/HTTPS/FTP server into a temporary location and reads it from there. The cmd parameter executesthe provided shell command and then reads the data from the stdout.
It is now possible to pass file objects to fread() (or any objects exposing method read()).
File path given to fread() can now transparently select files within .zip archives. This doesn’t work with archives-within-archives.
GenericReader now supports auto-detecting and reading UTF-16 files.
GenericReader now attempts to detect whether the input file is an HTML, and if so raises an exception with the appropriate error message.
Datatable can now use either llvm-4.0 or llvm-5.0 depending on what the user has.
fread() now allows sep="", causing the file to be read line-by-line.
range arguments can now be passed to a DataTable constructor.
Datatable will now fall back to eager execution if it cannot detect LLVM runtime.
Added simple Excel file reader.
It is now possible to select columns from DataTable by type: df[int] selects all integer columns from df.
Allow creating DataTable from list, while forcing a specific stype(s).
Added ability to delete rows from a DataTable: del df[rows, :]
DataTable can now accept pandas/numpy frames with columns of float16 dtype (which will be automatically converted to float32).
isna() function now works on strings too.
Function save() is now a method of Frame class.
Warnings now have custom display hook.
Added global option nthreads which control the number of Omp threads used by datatable for parallel execution. Example: dt.options.nthreads = 1.
Add method .scalar() to quickly convert a 1x1 Frame into a python scalar.
New methods .min1(), .max1(), .mean1(), .sum1(), .sd1(), .countna1() that are similar to .min(), .max(), etc. but return a scalar instead of a Frame (however they only work with a 1-column Frames).
Implemented method .nunique() to compute the number of unique values in each column.
Added stats functions .mode() and .nmodal().
When writing “round” doubles/floats to CSV, they’ll now always have trailing zero. For example, [0.0, 1.0, 1e23] now produces "0.0,1.0,1.0e+23" instead of "0,1,1e+23".
df.stypes now returns a tuple of stype elements (previously it was returning a list of strings). Likewise, df.types was renamed into df.ltypes and now it returns a tuple of ltype elements instead of strings.
Parameter colnames= in DataTable constructor was renamed to names=. The old parameter may still be used, but it will result in a warning.
DataTable can no longer have duplicate column names. If such names are given, they will be mangled to make them unique, and a warning will be issued.
Special characters (in the ASCII range \x00 - \x1F) are no longer permitted in the column names. If encountered, they will be replaced with a dot ..
Fread now ignores trailing whitespace on each line, even if ' ' separator is used.
Fread on an empty file now produces an empty DataTable, instead of an exception.
Fread’s parameter skip_lines was replaced with skip_to_line, so that it’s more in sync with the similar argument skip_to_string.
When saving datatable containing obj64 columns, they will no longer be saved, and user warning will be shown (previously saving this column would eventually lead to a segfault).
DataTable class was renamed into Frame.
“eager” evaluation engine is now the default.
Parameter inplace of method dt.Frame.rbind() was removed: instead you can now rbind frames to an empty frame: dt.Frame().rbind(df1, df2).
datatable will no longer cause the C locale settings to change upon importing.
reading a csv file with invalid UTF-8 characters in column names will no longer throw an exception.
creating a DataTable from pandas.Series with explicit colnames will no longer ignore those column names.
fread(fill=True) will correctly fill missing fields with NAs.
fread(columns=set(...)) will correctly handle the case when the input contains multiple columns with the same names.
fread will no longer crash if the input dataset contains invalid utf8/win1252 data in the column headers. #594 #628
fixed bug in exception handling, which occasionally caused empty exception messages.
fixed bug in fread where string fields starting with “NaN” caused an assertion error.
Fixed bug when saving a DataTable with unicode column names into .nff format on systems where default encoding is not unicode-aware.
More robust newline handling in fread. #634 #641 #647
Quoted fields are now correctly unquoted in fread().
Fixed a bug in fread which occurred if the number of rows in the CSV file was estimated too low. #664
Fixed fread bug where an invalid DataTable was constructed if parameter max_nrows was used and there were any string columns. #671
Fixed a rare bug in fread which produced error message “Jump X did not finish reading where jump X+1 started”. #682
Prevented memory leak when using PyObject columns in conjunction with numpy.
View frames can now be properly saved.
Fixed crash when sorting view frame by a string column.
Deleting 0 columns is no longer an error.
Rows filter now works properly when applied to a view table and using “eager” evaluation engine.
Computed columns expression can now be combined with rows expression, or applied to a view Frame.

Contributors¶

This release was created with the help of 5 people who contributed code and documentation, and 6 more people who submitted bug reports and feature requests.

Code & documentation contributors:

Issues contributors: