Not possible: https://doc.dataiku.com/dss/latest/schemas/recipes.html#python-r-pyspark-sparkr
It would be marvelous for productivity throughout flow development to be able to pass a schema method to the input/output dataframes which constructs the schema when doing a schema propagation.
def get_ouput_schemas():
result = {}
schema1 = ds_input1.get_schema()
schema2 = ds_input2.get_schema()
schema_1, schema_2 = do_something(schema1, schema2)
result['output_name_1'] = schema_1
result['output_name_2'] = schema_2
dataiku.get_schemas(get_output_schemas)