new task: read

[jangorecki]

Reading data benchmark is on the roadmap.
It should cover:

• reading csv
  most portable tabular data format, to cover transferring data between different solutions
• reading a binary formats
  most solution-specific formats, to cover transferring data within the same solution
• data of numeric fields only (integer and floats)
• data of 50% categorical fields (integer, floats and categorical)
• character fields
• date, time and datetime fields

ideas for testing particular features (maybe advanced questions?)

• top N rows
• particular rows
• particular columns
• ?

feedback welcome

[jangorecki]

I collected some feedback about this task from our internal discussion.

Initially I will focus only on reading csv, not a binary formats.

For real world data NYT will be good first case, we should probably find one more popular dataset, to have two real world data.

For simulated data:

• shape: long, wide, long and wide (fixed rows*cols?)
• types separately (3 columns of each type): int, double, char, factor, date, datetime
• types mixed (one columns of each type)
• cardinality (count of unq values)

[MichaelChirico]

relevant issue Rdatatable/data.table#2634