How to handle duplicate timestamp?

rakha

I want to forecast the value of "main", but there is a notification that the date is duplicate. In fact, each row is a unique row that has an identifier. How to solve this problem?

Operating system used: MacOS

rakha

training_cat-activity-log.txt

preparation_cat-activity-log.txt

Yasmine_T

Hi!

I hope that you are doing well.

This error means that you have several timeseries in your dataset with the same time stamps.

In order for training to work, you need to make sure that all columns that identify your various timeseries are marked as identifiers (see: https://doc.dataiku.com/dss/latest/machine-learning/time-series-forecasting/introduction.html#time-series-identifier-columns ). If this is not done, you will have time series where a given time stamps will have multiple target values hence the error you encountered above: If your time series does have multiple target values per time stamp, the default handling is to fail.

Note that you can also take the mean/mode of the values or drop time stamps with several target values as well.

Best,

Yasmine

tgb417

@rakha

Sounds like your raw data has not yet been setup for time series forecasting. Typically a time series forecasting dataset has a column with a date and time column. This column will have a complete set of equally spaced values. All of the days in a period, or all of the hours, or all of the weeks..,, This column will have no repeats or missing values or repeated (duplicate ) values. And then you will have another column with the value that you’d like to forecast. You may also have other set of columns features.

If you have multiple values that are the same in the date column you will have to decide what to do about the duplicates. Do the values you are trying to forecast need to be added together. For example, do you have a set of individual sales orders and do you want to forecast daily sales overtime. In a case like this in your raw data you might have multiple sales on a single day. And on holidays you might have no sales. In a case like this you might choose to add the dollar amount from each order to come up with a total daily sales. And fill in missing days with a 0 sales or average sales.

You may find this knowledge base article helpful for more details on working out this kind of issue. https://knowledge.dataiku.com/latest/ml-analytics/time-series/preparation/tutorial-index.html#tutorial-time-series-preparation