how to modify the template script on azure auto ml?

Question

Hello to all,

I am working on a machine learning project, I have trained my model on azure auto ml studio, I would like to import it in onnx format, but it is not in the download options, so I want to modify the resulting code directly in azure auto ml, in the script.py I have modified the recording format of the model but I can't find how to execute this script because it tells me that the script.py is not found,

Could you help me please ?

thank you

Lysa

Answer 1

AMROUN Lysa I believe you are looking to create a model wihich is ONNX compatible with AutoML as per your previous thread. In this case in the automl config if you setup enable_onnx_compatible_models to true in the automl config the model should be available for download. Something similar to what is provided as guidance in this notebook.

I would recommend running through this notebook and then change your experiment settings in a similar way to retrieve the best onnx format model.

automl_settings = {
    "experiment_timeout_hours": 0.3,
    "enable_early_stopping": True,
    "iteration_timeout_minutes": 5,
    "max_concurrent_iterations": 4,
    "max_cores_per_iteration": -1,
    # "n_cross_validations": 2,
    "primary_metric": "AUC_weighted",
    "featurization": "auto",
    "verbosity": logging.INFO,
    "enable_code_generation": True,
}

automl_config = AutoMLConfig(
    task="classification",
    debug_log="automl_errors.log",
    compute_target=compute_target,
    experiment_exit_score=0.9984,
    blocked_models=["KNN", "LinearSVM"],
    enable_onnx_compatible_models=True,
    training_data=train_data,
    label_column_name=label,
    validation_data=validation_dataset,
    **automl_settings,
)

If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.

Answer 2

@romungi-MSFT Thanks for your help it worked, and I use this onnnx to do classification of my input text, however it gives me only the first predicted class and I would like to have the first three with their scores in descending order, I do not know if it is possible to have the first three qualifications, for example when I use this onnnx in python via this little code:

import onnxruntime as ort
import numpy as np
import onnxruntime as rt


model_path = r"C:\Users\lysa.amroun\Desktop\modelOnnx\modelCVClassify.onnx"

session = rt.InferenceSession(model_path)

input_name = session.get_inputs()[0].name
clean_texte = "My Input Texte"
data = np.array([[clean_texte]], dtype=np.object)
outputs = session.run(None, {input_name: data})

print(outputs)

I have the following output: ``` ``[array(['LO'], dtype=object), array([[0.05193216, 0.12469713, 0.7025832 , 0.03859689, 0.0821906 ]],dtype=float32)] ```` (LO is my label), I would like to have the first three labels, how can I do it? thank you for your help Lysa