Next Previous Up Contents
Next: ascii
Up: Input Formats
Previous: csv ecsv

The Enhanced Character Separated Values format was developed within the Astropy project and is described in Astropy APE6 (DOI). It is composed of a YAML header followed by a CSV-like body, and is intended to be a human-readable and maybe even human-writable format with rich metadata. Most of the useful per-column and per-table metadata is preserved when de/serializing to this format. The version supported by this reader is currently ECSV 1.0.

There are various ways to format the YAML header, but a simple example of an ECSV file looks like this:

# %ECSV 1.0
# ---
# delimiter: ','
# datatype: [
#   { name: index,   datatype: int32   },
#   { name: Species, datatype: string  },
#   { name: Name,    datatype: string  },
#   { name: Legs,    datatype: int32   },
#   { name: Height,  datatype: float64, unit: m },
#   { name: Mammal,  datatype: bool    },
# ]
If you follow this pattern, it's possible to write your own ECSV files by taking an existing CSV file and decorating it with a header that gives column datatypes, and possibly other metadata such as units. This allows you to force the datatype of given columns (the CSV reader guesses datatype based on content, but can get it wrong) and it can also be read much more efficiently than a CSV file and its format can be detected automatically.

The header information can be provided either in the ECSV file itself, or alongside a plain CSV file from a separate source referenced using the header configuration option. In Gaia EDR3 for instance, the ECSV headers are supplied alongside the CSV files available for raw download of all tables in the Gaia source catalogue, so e.g. STILTS can read one of the gaia_source CSV files with full metadata as follows:

   stilts tpipe

The ECSV datatypes that work well with this reader are bool, int8, int16, int32, int64, float32, float64 and string. Array-valued columns are also supported with some restrictions. Following the ECSV 1.0 specification, columns representing arrays of the supported datatypes can be read, as columns with datatype: string and a suitable subtype, e.g. "int32[<dims>]" or "float64[<dims>]". Fixed-length arrays (e.g. subtype: int32[3,10]) and 1-dimensional variable-length arrays (e.g. subtype: float64[null]) are supported; however variable-length arrays with more than one dimension (e.g. subtype: int32[4,null]) cannot be represented, and are read in as string values. Null elements of array-valued cells are not supported; they are read as NaNs for floating point data, and as zero/false for integer/boolean data. ECSV 1.0, required to work with array-valued columns, is supported by Astropy v4.3 and later.

The handler behaviour may be modified by specifying one or more comma-separated name=value configuration options in parentheses after the handler name, e.g. "ecsv(header=,colcheck=FAIL)". The following options are available:

header = <filename-or-url>
Location of a file containing a header to be applied to the start of the input file. By using this you can apply your own ECSV-format metadata to plain CSV files.
Determines the action taken if the columns named in the YAML header differ from the columns named in the first line of the CSV part of the file.

This format can be automatically identified by its content so you do not need to specify the format explicitly when reading ECSV tables, regardless of the filename.

Next Previous Up Contents
Next: ascii
Up: Input Formats
Previous: csv

STILTS - Starlink Tables Infrastructure Library Tool Set
Starlink User Note256
STILTS web page:
Author email:
Mailing list: