Open
Description
The doc of COLUMN
clause is not quite clear. For example:
SELECT a, b, c ... COLUMN a, b
. Are the input columnsa, b, c
ora, b
?SELECT a, b ... COLUMN b, a
. Which is the first feature column,a
orb
?- Some column clauses would combine 2 or more selected columns, for example
CROSS(col1, col2)
.SELECT a, b, c COLUMN CROSS(a, c), CROSS(a, b)
. Do the input columns excludea, b, c
inSELECT
clause?SELECT a, b, c, d COLUMN CROSS(b, d)
. Are the input columnsa, CROSS(b, d), c
ora, c, CROSS(b, d)
?
COLUMN
clauses cannot refer to the old feature name when usingAS
. For example,SELECT a, b as b_alias ... COLUMN b
would raise error.- It seems that
COLUMN ... FOR ...
is only used in a few TensorFlow models (for example,DNNLinearCombinedClassifier
), and it is not valid in XGBoost models. Is it necessary to addCOLUMN ... FOR ...
for only these few models? Or, can we find a better way to handle this, instead of adding extraFOR
grammar?