by

Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
  • Nov 30 2019 19:07

    ttomita on master

    smerf experiments (compare)

  • Jun 23 2019 20:40

    ttomita on master

    add code for plotting manuscrip… (compare)

  • Jun 11 2019 17:37

    ttomita on master

    add striped image classificatio… (compare)

  • Jun 04 2019 05:15

    ttomita on master

    update scripts for running uci … (compare)

  • Jun 04 2019 00:00

    ttomita on master

    script for running matlab ccf o… (compare)

  • Jun 03 2019 03:20

    ttomita on master

    fix typo in save file name (compare)

  • Jun 03 2019 03:12

    ttomita on master

    fix typo in save file name (compare)

  • May 30 2019 15:21

    ttomita on master

    load necessary libraries within… (compare)

  • May 30 2019 03:47

    ttomita on master

    code to run uci real data exper… (compare)

  • May 30 2019 03:39

    ttomita on master

    code to run uci real data exper… (compare)

  • May 30 2019 03:30

    ttomita on master

    code to run uci real data exper… (compare)

  • Jul 03 2018 20:29

    ttomita on master

    ccf classification results on s… (compare)

  • Jul 02 2018 22:22

    ttomita on master

    ccf experiments on simulated da… (compare)

  • Jul 02 2018 22:02

    ttomita on master

    ccf experiments on simulated da… (compare)

  • Jul 02 2018 21:45

    ttomita on master

    ccf experiments on simulated da… (compare)

  • Jun 19 2018 03:04

    ttomita on master

    clean up scripts for running 20… (compare)

  • Jun 19 2018 03:03

    ttomita on master

    clean up scripts for running 20… (compare)

  • Jun 19 2018 03:01

    ttomita on master

    clean up scripts for running 20… (compare)

  • Jun 18 2018 16:01

    ttomita on master

    more ccf benchmark results (compare)

  • Jun 18 2018 05:36

    ttomita on master

    more ccf benchmark results Merge branch 'master' of https:… (compare)

joshua vogelstein
@jovo
let’s chat today if you are around?
ttomita
@ttomita
I had to take care of some stuff at the dmv. I'm free anytime today or any other day.
joshua vogelstein
@jovo
@ttomita dude?
we couldn’t find you, we are in carey’s whitehead office
ttomita
@ttomita
Invariance_Parity_RerF.png
here is Fig3 plot of RerF for Trunk
I mean Parity
Invariance_Trunk_RerF.png
This is Trunk
Invariance_Trunk_FRC2.png
ttomita
@ttomita
This is for Forest-RC
Fig4_Real_Data_Panel_A_RerFr.png
Fig4_Complexity.png
Top is comparing complexity (number of important variables) RerF(r) vs RF
RerF(r) tends to be about twice as complex as RF
ttomita
@ttomita
Bottom is Lhat for RerF(r). Sorry I didn't plot it with the others, but the Lhat is basically identical to RF
Training time is way longer than it was for the other RerF variants. I wonder if io20 and 21 were in heavier use than for the previous runs. I'm not sure how reliable it is to compare on a shared machine and when they aren't all done at the same time...
joshua vogelstein
@jovo
you there?
ttomita
@ttomita
Yep
I'm working on the intro
joshua vogelstein
@jovo
@ttomita we can chat via skype/G+ today at 4pm if that works?
ttomita
@ttomita
Works for me!
joshua vogelstein
@jovo
great!
joshua vogelstein
@jovo
ready when you are...
ttomita
@ttomita
I just walked home from school to realize I didn't have my house key. I'm walking back to Clark now. 5 min?
ttomita
@ttomita
Performance_profile.png
abaloneooberror_vs_ntrees.png
irisooberror_vs_ntrees.png
connect-4ooberror_vs_ntrees.png
ttomita
@ttomita
minibooneooberror_vs_ntrees.png
ringnormooberror_vs_ntrees.png
ecoliooberror_vs_ntrees.png
The first plot is the performance profile. The y-axis is the empirical distribution of the performance ratio as defined in the paper (except here it is defined using misclassification rate rather than time)
joshua vogelstein
@jovo
nice.
ttomita
@ttomita
The empirical distribution of RerF approaches 1 slightly faster than RF, which indicates that it is slightly better
joshua vogelstein
@jovo
which benchmarks is this? all? those with d < 100?
ttomita
@ttomita
d<100
joshua vogelstein
@jovo
also, can you tell me the code for the naming conventions?
ttomita
@ttomita
rf = RF, rerf = plain RerF, rerfd = RerF with original mean difference vector, rerfdn = RerF with node mean difference vector
The ones with 'rot' at the end are the rotation variants
ttomita
@ttomita
oh also rerfdr = RerF with original mean difference vector and pass to ranks
joshua vogelstein
@jovo
interesting
rot doesn’t seem to help much?
what do you think?
is it worth looking at ooberror for each dataset for all the different lags (both rot & not-rot), to see if we can see any patterns?
the titles could include n & d.
ttomita
@ttomita
rot seems to hurt
joshua vogelstein
@jovo
but i don’t see why
ttomita
@ttomita
yeah i can add the n and d
joshua vogelstein
@jovo
we need to look at ooberror vs. ntrees for both rot & not