CODEWITHSUNDEEP
amazon-web-servicesperformanceamazon-sagemakerarima
Tobitor
Tobitor

Reputation: 1508

Time Series Forecasting in SageMaker - which instance should I choose?

I am currently trying to forecasting hourly values with a weekly seasonality. In the SARIMAX model that means that I have to set m = 168 (24 hours a day and 7 days a week -> 24*7).

My model looks like this:

model = SARIMAX(df['IdCount'],order=(4,1,3),seasonal_order=(2,2,1,168))
results = model.fit()

And I think because of the relatively big m=168 it takes a very long time to fit the model. When doing this locally, the kernel died.

Therefore, I chose to try it in SageMaker and I am currently using ml.t2.medium instance. In order to perform this task fast and without any issues which instance would you recommend and why? Which is the best one only in regards to performance and which is the best one regarding cost benefit ratio?

Upvotes: 0

Views: 319

Answers (1)

rok
rok

Reputation: 2785

I know nothing about SARIMAX, but what I do with sagemaker is this.. I choose for the notebook the cheapest instance which is ml.t2.medium and then I use the notebook to launch the training job. For that training job I usually choose an instance with GPU. Anyway ml.t2.medium is very basic and I would not recommend it for anything else of what I do (launch other instances). You can take a look at this page for the available instance types, their features and also their price.

Upvotes: 1

Related Questions