Hyperparameter Limit Constraint

[anabia-alam]

Hi,

I am working on this notebook, to use detectron2's Retinanet on our custom dataset, instead of SKU110k dataset. Our dataset has 140 labels, so while running the training job this error is coming up

'hyperParameters' failed to satisfy constraint: Map value must satisfy constraint: [Member must have length less than or equal to 2500, Member must have length greater than or equal to 0, Member must satisfy regular expression pattern: .*]

Can you suggest a workaround for this issue? TIA

[paoloirrera]

Hi,

Could you please share the definition of the hyper-parameters that you would like to tune?

[anabia-alam]

These are the hyper parameters,

od_algorithm = "retinanet" # choose one in ("faster_rcnn", "retinanet") training_job_hp = {

Dataset

"classes": json.dumps(classes),
"dataset-name": json.dumps("sku110k"),
"label-name": json.dumps(classes2),
# Algo specs
"model-type": json.dumps(od_algorithm),
"backbone": json.dumps("R_101_FPN"),
# Data loader
"num-iter": 900,
"log-period": 500,
"batch-size": 16,
"num-workers": 8,
# Optimization
"lr": 0.005,
"lr-schedule": 3,
# Faster-RCNN specific
"num-rpn": 517,
"bbox-head-pos-fraction": 0.2,
"bbox-rpn-pos-fraction": 0.4,
# Prediction specific
"nms-thr": 0.2,
"pred-thr": 0.1,
"det-per-img": 300,
# Evaluation
"evaluation-type": "fast",

}

Classes here is the list of 140 labels on which we're trying to perform object detection. Here's a snippet of the classes list

[paoloirrera]

Could you also tell me the results of the following? len(json.dumps(classes))

[anabia-alam]

[paoloirrera]

OK, that is the problem. As the error message indicates, each hyper-parameter must not exceed 2500 chars. This is a SageMaker condition, and, hence, beyond the scope of this repository. In your case, it is 3508. I would suggest to rename your classes so that json.dumps(classes) does not exceed 2500.