Load a Registered Model in Azure ML Studio in an Interactive Notebook

Question

I'm using Azure Machine Learning Studio and I have an sklearn mlflow model stored in my default datastore (blob storage) which I have then registered as a model asset. How can I load this model inside an interactive notebook to perform some quick model inferencing and testing before deploying this as a batch endpoint.

I have seen a post linked here that suggests downloading the model artefacts locally but I shouldn't need to do this. I should be able to load the model directly from the datastore or the registered asset without the need to duplicate the model in multiple places. I have tried the following without success.

Reading from Registered Model Asset

import mlflow
from azure.ai.ml import MLClient
from azure.ai.ml.entities import Model

ml_client = MLClient(DefaultAzureCredential(), "<subscription_id>", "<resource_group>", "<workspace_id>")

model = ml_client.models.get("<model_name>", version="1")
loaded_model = mlflow.sklearn.load_model(model.id)

>>> OSError: No such file or directory: ...

Reading from Datastore

import mlflow

model_path = "<datastore_uri_to_model_folder>"
loaded_model = mlflow.sklearn.load_model(model_path)

>>> DeserializationError: Cannot deserialize content-type: text/html

Answer 1

According to this documentation any one of the paths should be given for loading the model.

• /Users/me/path/to/local/model

• relative/path/to/local/model

• s3://my_bucket/path/to/model

• runs:/<mlflow_run_id>/run-relative/path/to/model

• models:/<model_name>/<model_version>

• models:/<model_name>/<stage>

But you are giving model id and datastore path which is not supported.

So, try this code.

loaded_model = mlflow.sklearn.load_model("models:/local-mlflow-example/1")
loaded_model.predict(sample_data["data"])

Output:

Here the path should be models:/<model_name>/<model_version>

model.id or model.path or datastore path is supported when using azure ml jobs context.

So, to use model.id or model.path you submit the command job like below.

from azure.ai.ml import command
from azure.ai.ml.entities import Model
from azure.ai.ml.constants import AssetTypes
from azure.ai.ml import Input, Output

inputs = {
    "input_data": Input(
        type=AssetTypes.URI_FILE, path="./mlflow-model/input_example.json"
    ),
    "input_model": Input(type=AssetTypes.MLFLOW_MODEL, path=model.path),
}

outputs = {
    "output_folder": Output(
        type=AssetTypes.URI_FOLDER,
        path=f"azureml://subscriptions/{subscription_id}/resourcegroups/{resource_group}/workspaces/{workspace}/datastores/workspaceblobstore/paths/predictions",
    )
}

job = command(
    code="./src",  # local path where the code is stored
    command="python load_score.py --input_model ${{inputs.input_model}} --input_data ${{inputs.input_data}} --output_folder ${{outputs.output_folder}}",
    inputs=inputs,
    outputs=outputs,
    environment="AzureML-sklearn-1.0-ubuntu20.04-py38-cpu:1",
    compute="cpu-cluster",
)

# submit the command
returned_job = ml_client.jobs.create_or_update(job)
# get a URL for the status of the job
returned_job.studio_url

load_score.py script which loads the model and prints the result.

import argparse
import pandas as pd
import mlflow.sklearn
import pandas as pd
import json
import os

parser = argparse.ArgumentParser()
parser.add_argument("--input_data", type=str)
parser.add_argument("--input_model", type=str)
parser.add_argument("--output_folder", type=str)
args = parser.parse_args()

with open(args.input_data) as f:
    sample_data = json.load(f)

f.close()

print(sample_data)

sk_model = mlflow.sklearn.load_model(args.input_model)
predictions = sk_model.predict(sample_data["data"])

# Writing to stdout
print(predictions)

with open(os.path.join(args.output_folder, "predictions.txt"), "x") as output:
    # Writing data to a file
    output.write(str(predictions))
output.close()

Job output:

Refer this notebook for more information.

If you face no such file error that is because of the dependencies required is not present like MLmodel or model.pkl etc is deleted in storage account or moved to other folders.