how to set the configs for calling the trained QPTUNA model during staged learning

[MachineGUN001]

hi,

If you need to use the QPTUNA model for scoring during staged learning, my setup file is shown in the attachment. But the resulting file including the csv does not seem to call the trained QPTUNA model. Not sure how to solve this problem, thanks a lot for your help!

stage_test.txt

[halx]

You need to fix the syntax in analogy to the other components:

[[stage.scoring.component]]
[stage.scoring.component.Qptuna]

[[stage.scoring.component.Qptuna.endpoint]]
name = "Qptuna"  # you may want to choose a more descriptive name
weight = 0.7

params.model_file = "../models/best_SVR_20240502D000133.pkl"

# you may have to adjust the transform to the output range of the model
transform.type = "reverse_sigmoid"
transform.high = 0.0
transform.low = -50.0
transform.k = 0.4

[MachineGUN001]

hi, halx,

thank yo so much for kind explanation.

I added the below lines for syntax:

[[stage.scoring.component]]
[stage.scoring.component.Qptuna]

[[stage.component.Qptuna.endpoint]]
name = "Qptuna score" # by using a more specific name
weight = 0.7

params.model_file = "../models/best_SVR_20240502D000133.pkl"


[diversity_filter]

type = "IdenticalMurckoScaffold"
bucket_size = 10
minscore = 0.7

[inception]

smiles_file = ""  # no seed SMILES
memory_size = 50
sample_size = 10

the revised configs file was attached.
stage_test_toml.txt

however, there were no qptuna scores in the output csv.

I tried to use a trained Chemprop model and that works.
the chemprop scores can be found in csv.

if I did some wrong seting for QPtuna model?

many many thanks,

[halx]

You forgot the "scoring" in the section name, see below. This must be all consistent as you eventually describe a tree. Currently we do not have sanity checks in place to detect this but we are working on better input validation for a future release.

[[stage.scoring.component.Qptuna.endpoint]]

[MachineGUN001]

thanks a lot!

that works!