0

What do exactly rad_orig and r_emb represent in umap.UMAP? (docs).

These parameters are available when output_dens flag is set.

Reading the docs:

r_orig: array, shape (n_samples)

Local radii of data points in the original data space (log-transformed)

and

r_emb: array, shape (n_samples)

Local radii of data points in the embedding (log-transformed).

I tried to figure that out with an example, by having a toy array of integers with shape (100,50):

array([[15,  8, 16, ..., 12,  9, 14],
       [ 4,  4,  5, ...,  4, 19, 15],
       [ 2,  4, 16, ...,  4,  7,  8],
       ...,
       [11, 17, 14, ...,  7, 18,  6],
       [ 2, 16, 12, ..., 18, 17, 15],
       [ 3, 11,  9, ..., 11, 14,  8]])

and executing

umap_trans = umap.UMAP(densmap=True,output_dens=True).fit(arr)

then r_orig looks something like

array([7.557382 , 7.6884522, 7.5413175, 7.5586753, 7.526751 , 7.633186 ,
       7.579795 , 7.6138983, 7.4713755, 7.5365367, 7.63102  , 7.627236 ,
       7.5586395, 7.5616612, 7.5946164, 7.626307 , 7.6850867, 7.5265946,
       7.5604353, 7.5958605, 7.5464926, 7.5515323, 7.6224527, 7.5082755,
       7.6015797, 7.5680337, 7.6188903, 7.5625224, 7.6245193, 7.5826597,
       7.6149483, 7.5915165, 7.558839 , 7.613548 , 7.578578 , 7.613815 ,
       7.684106 , 7.5169396, 7.5644665, 7.6615157, 7.6193194, 7.626235 ,
       7.656492 , 7.58103  , 7.5389533, 7.641165 , 7.588751 , 7.554403 ,
       7.647078 , 7.6455092, 7.561126 , 7.5732226, 7.6015496, 7.6265235,
       7.564877 , 7.5956354, 7.6075587, 7.5987916, 7.626135 , 7.539194 ,
       7.5905514, 7.6090746, 7.6593614, 7.6186256, 7.66446  , 7.5629582,
       7.6118226, 7.54342  , 7.5881543, 7.563827 , 7.60424  , 7.6116834,
       7.5791817, 7.5829387, 7.6135163, 7.562068 , 7.7188945, 7.5859914,
       7.6612687, 7.5608892, 7.5465975, 7.5277977, 7.6697884, 7.5451746,
       7.5410295, 7.5975976, 7.588921 , 7.6266494, 7.630443 , 7.621092 ,
       7.5729136, 7.559135 , 7.665758 , 7.585926 , 7.7076025, 7.4915547,
       7.6049953, 7.5991044, 7.637067 , 7.5531616], dtype=float32)

how are these numbers related to the original array? I couldn't find any further mathematical explanation on this.

James Arten
  • 523
  • 5
  • 16

0 Answers0