Price vs. Time Diagrams

Subscribe to Price vs. Time Diagrams 5 post(s), 2 voice(s)

Avatar Eike 10 post(s)

I’ve written a little program that creates price vs. time diagrams, with a moving median price. It can read JBidwatcher’s “auctions.xml” files, and the “.csv” file that JBidwatcher creates.

The program is a bit difficult to install, because it is written in Python, and depends on two big scientific libraries: Numpy and Matplotlib. Nevertheless I hope it is useful at least to some of you.

The diagrams look like this:

Installation and usage instructions are here:

There is an entry in the Python Package Index, with minimal instructions:

There is also a development site, for reporting bugs, requesting new features, and sharing code.

Small ideas should be entered as bugs, with the “Importance” set to “Wishlist”; big ideas should be entered as so called “Blueprints”. Off course I would also love to have collaborators who help with coding.

I would also like to hear what other users of JBidwatcher think about such a program. Suggestions for features and the user interface are welcome.

Post has been completely rewritten on 2011-02-20.
Updated on 2011-03-27.

Avatar Eike 10 post(s)

Post can be deleted. I previously didn’t know that posts can be edited.

Avatar Quo 10 post(s)

excellent idea, looks very promising on first try!. I had been collecting some data and used gnuplot to plot it, but I haven’t been into parsing that xml stuff and did that manually…

What I’d love to see would be:

  • vertical markers at the start of each week (depending on the category, ending prices have a tendency to show some kind of pattern, let’s make that visible)
  • for the same reason: graph scaled to real time (right now it looks, as if all auctions within e a day are displayed at the same x-coordinate).
  • optional: align data from a certain period of time on their corresponding weekday (fold it all to cover one week, to stress intra-day / intra-week patterns), with a scale Mon-Sun

I like the idea to group auctions by tab, this allows to filter out new versus used, or items with additional gear versus “bare” items, or those vendors you suspect to be pushing their own auctions rather quickly.
To make that even more useful, an optional handling of the comment as comma-separated taglist would be great (to have some items show up among several tags).

So far my experiments had a very straight median. I’d love to see it moving according to the in-week fluctuation, and to show the tendency of rising/falling prices. I haven’t found out how to achieve that yet, though.

Thanks for sharing this!

Avatar Eike 10 post(s)

Hello Quo!

Great that I really have a user!

To make the median change faster (more wiggly) use the `-w` option. The default is `-w 30`, the minimum is `-w 3` which computes the median over three days.

Thank you for the suggestions, I’ll incorporate them into the TODO list.

  • Your suggestion, that the time axis should have a higher resolution than just days, seems to be an easy task. I need to convert the datetime objects to a number with a different function, I just stumbled on it: matplotlib.dates.date2num(d).
  • The idea, to interpret the comment as a comma separated list of tags, seems pretty smart to me. And not much work.
  • The weekday markers and the week diagram are more work.
  • I would like to integrate information about the seller and high bidder into the diagram, but I can’t envision a nice way to do it.

If you can program and want to implement some of the ideas yourself, let me know. I’ll create a project in Launchpad then, and give you commit rights.


Avatar Eike 10 post(s)

Empty post. See pointers to the websites in the first post.