Hey!
I am trying to integrate our existing custom ...
# experimentationf
fresh-application-65175
01/27/2025, 12:30 PMHey!
I am trying to integrate our existing custom MultiArmBandit and update weights each N time via API https://docs.growthbook.io/api/#tag/experiments/operation/updateExperiment
So, the problem is that new weights are simply ignored and is always 50/50
Is it a known issue?
please see request/response in attached snippet
👀 1
✅ 1
fresh-application-65175
01/27/2025, 12:50 PMFor anyone looking for the same answer
After checking the source code https://github.com/growthbook/growthbook/blob/main/packages/back-end/src/services/experiments.ts
I noticed that the actual weight values are getting from another field
Copy code
if (reweight) {
// apply the latest weights (SDK level)
changes.phases[phase].variationWeights = banditResult.updatedWeights;
// re-randomize to reduce bias (in cases of multiple exposures / failed sticky bucketing)
changes.phases[phase].seed = uuidv4();
}variationWeights Copy code
{
"phases": [
{
"name": "MAB Optimization 2025-01-27T12:46:43Z",
"dateStarted": "2025-01-27T12:46:43Z",
"trafficSplit": [
{
"variationId": "var_404rwz1tm6f1iccp",
"weight": 0.4
},
{
"variationId": "var_404rwz1tm6f1iccq",
"weight": 0.6
}
],
"coverage": 1,
"reason": "Multiarm Bandit Optimization",
"seed": "seed_2025-01-27T12:46:43Z",
"targetingCondition": "{}",
"savedGroupTargeting": [],
"variationWeights": [
0.4,
0.6
]
}
],
"status": "running"
}"}h
helpful-application-7107
01/27/2025, 8:01 PMGlad you were able to resolve it! Sorry there is some tech debt in this part of the code base that can make the API a little confusing. You're correct that you need to set the weights in the latest phase.
🙏 1
23 Views