1

I am trying to import time-series data to the Amazon forecast. While importing the dataset group, I am getting an error saying

Too few observations (1055) for the number of items (1055), averaging 1.000 observations per item

My dataset has 1055 entries, csv format:[item_id, timestamp, utilization]

1,2020-01-01 09:01:00,23 
2,2020-01-02 00:00:00,45 
3,2020-01-03 00:00:00,41 
4,2020-01-04 00:00:00,42 
5,2020-01-05 00:00:00,43

I can think of two possibly two reasons for this issue:

  1. Each item needs to be identified uniquely --> Above dataset has item_id which is unique for each item.
  2. 1055 records is too few --> I tried importing 100k records. Still facing the same issue.

Here, each item in the dataset can be identified uniquely. The AWS forecast documentation mentioned nothing about this error.

James Z
  • 12,209
  • 10
  • 24
  • 44
MagicWand
  • 33
  • 1
  • 5

2 Answers2

3

Leaving an answer in case this can help future web surfers.

Issue might be on item_id, as a programmer we think it needs to be unique, in this case is the opposite. Every row is an observation, try to put same item_id for all rows, or in groups (in case it make sense in your data) e.g 5K rows are type1, 5k rows are type2 and so on. Forecast then will have 5k observations per type, if item_id is unique, it only has one observation per type, not enough to make a prediction and showing "Too few observations error".

Hope it helps, this is how I resolved this isue in my case. KR

David Parés
  • 31
  • 3
0

I don't know if it's a bug or so but if the item_id id super unique. Then it won't work.

Make sure you have more then 1000 entries against one item_id.

Mohsin Khan
  • 21
  • 4