Ensuring that parameters in configuration files are coherent
For now, there are a lot of tests to check if a parameter is given with the right type: string, list or int for instance.
The recent check_input feature allows to validate the input data before the chain is launched.
However, there ir no verification that the provided parameters are coherent and well filled. For instance, we can provide parameters for autocontext and sample augmentation while these two modes can not be used together.
For the custom_features and probably other modes, a check with fake data could be done before the chain starts, to avoid simple errors like a division by zero or an incorrect call to get methods.