types of input and prediction differ

[behrica]

train:

:x	:target
0	0
1	1

predict:

x	y	:target
0.76449511	0.23550489	0.0

0 vs 0.0

logictic regression does gis for example.
Root caused by accepting int as modl input and cast it yo double;

[behrica]

casting happens here:

scicloj.ml.smile/src/scicloj/ml/smile/classification.clj

Line 38 in 8edf398

(let [posterior (double-array n-labels)]

[behrica]

The smile model takes only double-arras double and returns double arrays.
We cast int data to double , but not back to int.
-> input and output are of different type

-> a problem as well for categorical maps. There we "cast" as well to get them working:
https://github.com/techascent/tech.ml.dataset/blob/598e34461386a697d1f34db11c556fedf79225d0/src/tech/v3/dataset/categorical.clj#L176

[behrica]

as well related to this discssusion:
https://clojurians.zulipchat.com/#narrow/stream/236259-tech.2Eml.2Edataset.2Edev/topic/invert-categorical-map.20-.20regression.20tests