3

An example of an ideal dataset would look like:

  • There are some devices, say computers
  • Each computer has parameters (number of cpus, amount of ram, hdd space, etc.)
  • Also we know some time-series data (energy consumption log, load log, etc.) about each of the devices
  • And we need to predict whether it will fail soon or not.

This is just an example, the domain of the dataset does not matter, it could be hospital records about patients, some other devices, unicorns, whatever. The main requirement is that each instance has some number of useful "static" features and some number of useful "dynamic" features.

Has anyone seen a dataset(s) like this?

Improve this question

2 Answers 2

Reset to default
2

What you are looking for is a time series with descriptive metadata. Metadata, in this sense, would be the non-time series values that are never or seldom changing.

For example, the metadata for a power plant includes:

  • units of measurement

  • max production capacity

  • latitude and longitude

  • installation date

  • manufacturer

The time series data for the same power plant then only includes:

  • timestamp (always include UTC AND local time!)

  • quantity of production

The distinction is actually based on how to store the data, meaning that the time series is always being updated, where the metadata is much more static. In relational databases, you usually store metadata and timeseries data for the same entity in separate tables, and then JOIN to combine them with a unique ID (primary key). This decreases the space, while the metadata table may have hundreds of columns, the timeseries table will only have a handful.

To find a time series:

Time Series Data Library

or

Green Button energy data

enter image description here

Improve this answer
1
  • Yes, "metadata" is indeed a good word to describe what I am looking for. The problem with the datasets like the one you brought as an example is that samples do not have any labels, so there is nothing to classify.
    – Kuz
    Nov 4, 2015 at 16:53
0

Err, a readily available data set that seems to meet your parameter requirements would be the access logs of any web based application, e.g., httpd access logs. Specifically -

  1. parameters include size of the page requested, type of device requesting, status of request, etc.

  2. time-series data include when the request was made, how many requests per time unit, etc.

Improve this answer
1
  • 1
    What could be a class (label) in such dataset?
    – Kuz
    Nov 4, 2015 at 16:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.