Continuously update Azure Automated ML timeseries model

user_1932379 0
Viewed 87

I have recently deployed a simple Azure automated ML model and I have an endpoint that can give me predictions. Now, what I'd like to do is push updates to the AI model, so that it is always up to date.

For example, I want to predict an event that happens in 10 minutes. After 10 minutes have gone by, and I learn the real value, I'd like to push that value at the end of the ML model data array without having to re-train/deploy everything. Is that possible in Automated ML?

Edit: Alternatively, it would also be fine if I could run the predictor/retraining offline, in my own local server. The azure does allow downloading the trained model, but I don't really know how to use the pkl file, or whats in it.

time-series machine-learning azure
1 Answers

Checkout Set up AutoML to train a time-series forecasting model with Python which uses past time-series values are "pivoted" to become additional dimensions for the regressor together with other predictors. And you can specify separate train and validation sets directly in the AutoMLConfig.

An Azure model is one or more serialized Python objects, packaged as a Python pickle file (.pkl extension). The contents of the pickle file depend on the machine learning library or technique used to train the model.

To un-pickle the data you can:

import pickle


with open('serialized.pkl', 'rb') as f:
    data = pickle.load(f)

Pickle serializes a single object at a time, and reads back a single object - the pickled data is recorded in sequence on the file.

If you simply do pickle.load you should be reading the first object serialized into the file. After un-serializing the first object, the file-pointer is at the beginning of the next object - if you simply call pickle.load again, it will read that next object - do that until the end of the file.

objects = []
with (open("myfile", "rb")) as openfile:
    while True:
        try:
            objects.append(pickle.load(openfile))
        except EOFError:
            break

See, Deploy models trained with Azure Machine Learning on your local machines for learning how to use your local computer as a target for training or deploying models created in Azure Machine Learning.

Refer: How to unpack pkl file?, How to read pickle file?

If you'd want to use the ML Studio's Web API to obtain predictions ,see Download a trained ML Model from Azure ML studio to deploy on a standalone computer

user_159694170
Related