diff --git a/docs/_source/customise.rst b/docs/_source/customise.rst index 14738fcb6751495b5dcb10b83bcd2d420b79d71b..45971f5fa5ccdc934c3ea448c2f2d0ebbd8eb070 100644 --- a/docs/_source/customise.rst +++ b/docs/_source/customise.rst @@ -286,6 +286,35 @@ Custom Data Handler method is modified (see previous bullet). * (optionally) Overwrite the base class :py:`self.get_coordinates()` method to return coordinates as dictionary with keys *lon* and *lat*. +* (recommended) It is highly recommended to stay with our naming convention of data axis for an better integration into + MLAir. In principle, the naming can be set from outside and changed to any naming, but this must be done consequently. + We recommend therefore to use the following convention: + +============== =========== ============================================================================================= +dimension name default meaning +============== =========== ============================================================================================= +time_dim "datetime" temporal axis, every new entry along this axis is assumed to be a new sample +iter_dim "Stations" iter axis that is a collection of different data "sources", this could be a collection of + stations or also a temporal split for a single year. +window_dim "window" this axis is the "2nd" dimension of each samples. This could be 2D data if only a single + timestamp is provided for each sample. In the default data handler, we use this dimension to + provide lagged data in each sample. +target_dim "variables" the target dim indicates where to find the target vector. Furthermore, this also contains + information on the "3rd" dimension of each sample. On input site, this can be used to provide + different channels (e.g. color channels). On target site, this indicates the dimension along + the target variable can be found. In the default data handler this dimension is used to flag + different variables. +filter_dim "filter" This is a custom dimension, which is used by some specific data handlers implemented by the + authors. It could be described as the "4th" dimension. It is always possible to define a + custom 4th dimension, but it could happen, that some parts of the code will not work very + well with this additional dimension. But have a try and otherwise report a bug. +============== =========== ============================================================================================= + + +.. note:: + + The following code snippets are not up to date currently. Please read the requirements above for in-depth + information what MLAir is expecting from a custom data handler! .. code-block:: python