Error Training Vertex AI with Auto ML - Found time series with more than 3000 time steps. Ensure the lengths of all time series do not exceed 3000

Question

Has anyone experienced this error while using Auto ML to train a model in Vertex AI? I don't understand what it means.

The time series I'm using has this format, "2021-06-24 00:00:00 UTC", and the error says:

Found time series with more than 3000 time steps. Ensure the lengths of all time series do not exceed 3000.

Here's a screenshot of the error:

enter image description here

Use Vertex Auto ML to train a model, using a data set from Big Query

• AutoML
• Tabular data set in Big query
• Previsión
• Optimized for RMSE

Answer 1

¡Hola Sebastian!

This error means that one of your timeseries (let's say at least one of the products on your dataset) has more than 3000 data points (hourly/daily entries).

An example - If I have the following dataset:

product, timestamp, count
1, 20230101, 1
1, 20230101, 5
1, 20230101, 2
1, 20230101, 3
1, 20230101, 4
2, 20230101, 6
2, 20230101, 8
2, 20230101, 5
2, 20230101, 1
3, 20230101, 2
3, 20230101, 3
3, 20230101, 4

The timeseries for the product 1 has 5 samples, 2 has 4 samples and 3 has 3 samples.

If you have any individual product/id with more than 3000 samples, you may face the error you are facing.