setup xgboost on sqlflow #662
Conversation
There was a problem hiding this comment.
@sperlingxx Thanks for submitting this PR. Left a few comments.
@Yancey1989 please kindly continue the review.
| FROM iris.test | ||
| PREDICT iris.predict | ||
| WITH | ||
| append_columns = [sepal_length, sepal_width, petal_length, petal_width], |
There was a problem hiding this comment.
Is it possible to avoid append_columns? A training job should memorize the field names later used for a prediction job.
| fmt.Fprintf(&b, "%s %s, ", r.DetailColumn, stype) | ||
| } | ||
| // add encoding column | ||
| if len(r.EncodingColumn) > 0 { |
There was a problem hiding this comment.
What is the usage of the encoding column?
There was a problem hiding this comment.
Encoding column stores leaf indices of this sample in each tree. We transform leaf indices in a string which format like: "index_0,index_1,......,index_n"
| FROM iris.test | ||
| PREDICT iris.predict | ||
| WITH | ||
| append_columns = [sepal_length, sepal_width, petal_length, petal_width], |
There was a problem hiding this comment.
Does prob_column , detail_column is required? Can we add these columns by default, so the WITH statment can be shorter...
There was a problem hiding this comment.
Prob_column, detail_column, leaf_column and append_columns are optional. Only result_column is required for prediction task, which has a default column field: "result".
There was a problem hiding this comment.
Can keep consistent with the TF example, specific result column in PREDICT iris.predict.result?
3 participants
This PR include: