search

dfm / Star-Field

Other
5 stars 1 forks source link

Format of the *_ground_truth.txt files #22

Closed dfm closed 11 years ago

dfm commented 11 years ago

@eggplantbren can you describe the format of these files? I can see how you load it in your demo but a few words explaining why the vast majority of the data in the file is not being looked at would be helpful! Thanks.

eggplantbren commented 11 years ago

It's the same format as the results files (sample.txt) etc. One model is written out on one line. It begins with the number of stars and a few other parameters, then goes into the catalog. Vector of x positions, vector of y positions, vector of fluxes. The catalog is padded with zero-flux stars so that multiple catalogs can be written in the same file and then the matrix stays rectangular.

dfm commented 11 years ago

I guess the thing that's confusing is that the single row in the file 100_ground_truth.txt has 10611 numbers. I expect that the first 11 of these are the hyperparameters from what you do here and then it looks like 11:211 are the x-coordinates, 211:411 are the y-coordinates and 411:611 are the fluxes. That leaves 10000 entries. What are these?

dfm commented 11 years ago

Also, in the 1000_ground_truth.txt file how do I know what to set the maxNstars value to?

eggplantbren commented 11 years ago

The 10,000 entries are the mock noise-free image computed from the catalog. In 1000_ground_truth.txt maxNumStars is 2000 I believe (yet for posterior samples using that data it's 1200, that was an error by me).

Apologies for the lack of clarity here. I didn't intend for someone else to have to look at it in this much detail (lame excuse I know).

dfm commented 11 years ago

Haha! Cool. That makes sense... don't know why I didn't notice the whole 100² thing.

Thanks!