canonical figures

[jovo]

1. fig 1 of ROFLMAO
2. fig 2 of ROFLMAO
3. fig 3 of ROFLMAO
4. fig 4 of ROFLMAO
5. Lhat(RF) - Lhat(X) distribution figure
6. Lhat vs. treesize for each alg (not sure yet what that looks like)
7. Lhat vs. ntrain for each alg (not sure yet what that looks like)

the algorithms we care about include:

1. RF
2. RerF
3. F-RF
4. "james" RerF
5. RR-RF

• rank versions of each.
• i don't think we need to keep running RR-RF, it sucks and takes long.

let's get a version of each these 6 figures next week? this will be the start of the PAMI/JMLR paper....

sound good?

[tyler-tomita]

fig 1 of ROFLMAO

fig 2 of ROFLMAO

Lhat(RF) - Lhat(X) distribution figure

Lhat vs Depth

• Trunk [Sparse parity]- (https://github.com/ttomita/RandomerForest/blob/master/Figures/Sparse_parity_error_vs_depth.pdf)

[jovo]

these are great, including the tree depth. can you do the tree depth plot on the benchmark data, color code by algorithm? not sure if that is the best way to compare tree depth?

maybe x & y axes tree depth (RF) - tree depth (X) and same for Lhat?

[jovo]

and can you please make an issue in weekly-experiments like https://github.com/neurodata/weekly-experiments/issues/300 linking to which issues you will complete this week.

[jovo]

are all the outliers for "Lhat(RF) - Lhat(X) distribution figure" on the black outline box? if so, then we could understand the number of outliers on both sides, which would be nice....

[tyler-tomita]

The outliers are outside the y-axis limits and are not shown. I can try compressing rather than cutting off.

[jovo]

well, i was thinking about our performance improvement being so minor, and then i wondered whether we do substantially better sometimes, and substantially worse fewer times...

On Mon, Nov 28, 2016 at 9:51 AM, Tyler Tomita [email protected] wrote:

The outliers are outside the y-axis limits and are not shown. I can try compressing rather than cutting off.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/ttomita/RandomerForest/issues/99#issuecomment-263290214, or mute the thread https://github.com/notifications/unsubscribe-auth/AACjclqCzRcNV8cTGHinNYDEMu_adhsvks5rCupcgaJpZM4KuxJB .

-- the glass is all full: half water, half air. neurodata.io, jovo calendar https://calendar.google.com/calendar/embed?src=joshuav%40gmail.com&ctz=America/New_York