Uploading your demand data to Beam
Since the data includes a date and value, it is said to be a time series because the value is indexed with time. As instructed on the page, the accepted file formats are CSV and JSON with column names in the first row and each column separated by a comma. Click the question mark next to file format documentation for a full explanation of the data formatting requirements.
Beam file format
See the file below for an example of the correct file format for Beam
Below is an example of what the file should look like. The column headers must be date and demand only and the date format YYYY-MM-DD to avoid receiving errors.
The size of data allowed is from 6 months (minimum) to 4 Years (maximum). Beam additionally requires data that starts no earlier than 2017-01-01 and ends no later than the current day.
The file you upload can only contain 1 row per date. You should not have multiple rows with the same date. You must have 1 demand amount per date. If you repeat the same date multiple times, the upload will take the latest instance of a date you uploaded and ignore the previous values.
Demand metric can vary. In other words, any unit (revenue, customer count, etc) that your business uses for demand forecasting will be the best value to use for the demand unit in the data you upload.
Granularity and missing data
It is important that the data is aggregated on a daily level. E.g: On the 29th March 2017, a business recorded a demand of 1000 whereas the next day, the demand was 800.
The Demand column represents a quantity, whether count or sum and cannot be a negative value. The value must be greater than or equal to 0. If a date exists, then the respective demand should not be blank or null. Similarly, if a value for demand exists, then a date ought to be present. Note that whilst it is tempting to replace missing values with 0, it is strongly advised against doing so unless you are sure that no data means no demand.
If a missing row means that the system was down or no record was available for that day, then we suggest leaving the entire row out of the data set for accurate results.