Age | Commit message (Collapse) | Author | |
---|---|---|---|
2011-08-04 | save command added: save_filename| | Jake Hofman | |
2011-03-21 | fix arm alignment bug from Jani and extra whitespace removal from sudarshan | John Langford | |
2011-01-20 | made 29 bits correct and explicitly disabled greater. | John Langford | |
2010-09-24 | optimizations for speed | John Langford | |
2010-04-20 | deleted debian-specific and tweaked n/skip generation so it made sense. ↵ | John Langford | |
Must debug further. | |||
2010-04-19 | Supports signed differences in the cache format. This uses the ↵ | Shravan M Narayanamurthy | |
ZigZagEncoding & Decoding to convert signed integers to unsigned integers and vice versa essentially using the lsb to represent sign instead of msb. While encoding the signed differences are first converted to unsigned differences & then RLEd into the cache. While decoding the unsigned differences are RL decoded from the cache and converted to signed differences. A new option to sort the features disregarding order of occurrence is provided to allow better compression. Also fixes a bug in compressed option. Currently, reading from stdin doesn't work with compressed option turned on. So we check this and bail out with an error message. This is an effect of using zlib. | |||
2009-12-31 | Applied Shravan's zlib compression patch | John Langford | |
2009-12-11 | Adding support for compression using zlib/gzip | Shravan M Narayanamurthy | |
Abstracted all direct access to opening, closing, resetting, flushing, reading & writing files of io_buf into access through a function of io_buf which have all been declared virtual. Now that these accesses have been abstracted out, adding a new type of io_buf, in this case, comp_io_buf, is pretty easy. Just public inherit io_buf and override the default definitions of these virtual functions to work with compressed files. There is a nuance in this because of the current use of multiple files. The 'files' v_array is conveniently typed to int. So in the comp_io_buf, I hold a v_array of gzFiles whose indices are stored in 'files' v_array. Have introduced a new boolean flag compressed into parse_args. The input & output io_bufs in parser are now just pointers that will point to comp_io_buf if the compressed option is chosen. This is the root cause of most of the changes involving parser::input & parser::output. Because of the design of zlib which allows gzread to read from a normal file we can read from a mix of compressed and uncompressed files(cache) without extra logic. | |||
2009-12-07 | searching for strange heisenbug | John Langford | |
2009-12-05 | debugged a bit | John Langford | |
2009-12-04 | tweaked format for tag and label | John Langford | |
2009-12-01 | fixed performance bug due to unnecessary sorting | John Langford | |
2009-10-18 | Various bugfixes for cluster support | John Langford | |
2009-09-20 | first compiling (but not debugged) cluster version. | John Langford | |
2009-08-31 | Fixed bad input assignment bug. | John Langford | |
2009-08-31 | debugging odd cluster bug | John Langford | |
2009-08-31 | Reworked things so input and output are distinct. | John Langford | |
2009-08-23 | Made binary mode sending work. | John Langford | |
2009-08-22 | Merge branch 'master' into cluster | John Langford | |
Conflicts: main.cc parse_example.cc vw.cc | |||
2009-08-22 | NaN check on input and making it so cache file is valid for smaller cache sizes. | John Langford | |
2009-08-17 | First sending core. | John Langford | |
2009-08-01 | Initial release of version 3.10. I've incorporated some of the | John Langford | |
Gordon's changes. |