search

KlugerLab / FIt-SNE

Fast Fourier Transform-accelerated Interpolation-based t-SNE (FIt-SNE)
Other
593 stars 108 forks source link

Changing the defaults in the Python wrapper #93

Closed dkobak closed 4 years ago

dkobak commented 4 years ago
  1. Learning rate is set to max(200, N/early_exag_coeff) by default.
  2. Iteration number is set to 750 by default (250+500).
  3. Initialization is set to PCA (via ARPACK) by default.
  4. N-body algorithm is set to FFT for N>=8000 and to BH for N<8000 by default. UPDATE: I TOOK THIS OUT!
  5. Late exaggeration start is set to the early exagg end by default (if late exagg coeff is provided).

I updated the example notebook too.

This fixes issues #88 #89 #90.

UPDATE: also implements multithreaded Barnes-Hut!

dkobak commented 4 years ago

@pavlin-policar How is it looking with the next openTSNE release, by the way? George and me have an upcoming comment where we mention that openTSNE 0.3.0 implements spectral initialization :-) Should make sure that it's actually out before the comment is published. It's not accepted yet, so there is time, but I wanted to check with you.

pavlin-policar commented 4 years ago

Hey, I've had to shift gears this the past week or so because I've got deadlines for some coursework to catch. I should be back on this in a week or so. I have a couple things to update in the readme, and I've rerun benchmarks, so minor things like that need to be updated. Switching over to annoy actually brought openTSNE on par with FIt-SNE performance-wise, so that's nice :)

dkobak commented 4 years ago

Sounds good. Thanks for the heads up.