search

tyler-tomita / RandomerForest

Discriminant Projection Forest results, datasets, etc.
44 stars 21 forks source link

fig 5: analysis of performance vs. RF #48

Closed jovo closed 9 years ago

jovo commented 9 years ago

i wanna try to understand when we are doing better. for that reason, i want a fig (doesn't need to be pretty now), 1 row, 5 columns. y-axis: Lhat(R'erF(s+d+r)) - Lhat(RF) x-axis: 1) D 2) n 3) n/D 4) Lhat(naive bayes classifier), in matlab 'classify' with the type set to 'diaglinear' does it, and so does the 'naive bayes' method.

tyler-tomita commented 9 years ago

for naive bayes Lhat, should I do loocv?

jovo commented 9 years ago

i guess so.

On Tue, May 12, 2015 at 5:40 PM, ttomita notifications@github.com wrote:

for naive bayes Lhat, should I do loocv?

— Reply to this email directly or view it on GitHub https://github.com/ttomita/DPForest/issues/48#issuecomment-101429531.

the glass is all full: half water, half air. openconnecto.me, jovo.me, office hours https://www.google.com/calendar/embed?src=e2ktu4lrgul8anp8hclrcminp8%40group.calendar.google.com&ctz=America/New_York

tyler-tomita commented 9 years ago

I saved each panel to a different file (Fig5panel#). The relationship between LhatRF_delta and n and d is very weak. There also seems to be a slight negative correlation between LhatRF_delta and LhatNB

jovo commented 9 years ago

can you plot the first 3 panels on a log scale for the x-axis so we can see things better? we should also think about how to quantify sparsity of the discriminant boundary, which might matter...

On Fri, May 15, 2015 at 12:44 PM, ttomita notifications@github.com wrote:

Reopened #48 https://github.com/ttomita/DPForest/issues/48.

— Reply to this email directly or view it on GitHub https://github.com/ttomita/DPForest/issues/48#event-306008351.

the glass is all full: half water, half air. openconnecto.me, jovo.me, office hours https://www.google.com/calendar/embed?src=e2ktu4lrgul8anp8hclrcminp8%40group.calendar.google.com&ctz=America/New_York

tyler-tomita commented 9 years ago

It's log scale now. What is meant by sparsity of the discriminant boundary? Is that like a measure of how much each dimension contributes to the discriminant boundary?

jovo commented 9 years ago

interesting. can you add a regression line?

sparsity of the discriminant boundary means something like "how well can we do only using k ambient dimensions?" not sure what is the right way to quantify this. i'm just trying to figure out when RerF beats RF...

On Fri, May 15, 2015 at 1:50 PM, ttomita notifications@github.com wrote:

It's log scale now. What is meant by sparsity of the discriminant boundary? Is that like a measure of how much each dimension contributes to the discriminant boundary?

— Reply to this email directly or view it on GitHub https://github.com/ttomita/DPForest/issues/48#issuecomment-102470409.

the glass is all full: half water, half air. openconnecto.me, jovo.me, office hours https://www.google.com/calendar/embed?src=e2ktu4lrgul8anp8hclrcminp8%40group.calendar.google.com&ctz=America/New_York