Closed mikejiang closed 7 years ago
mikejiang
commented
11 years ago normalizing 2 channels for 12 samples:
chnls <- c("<Violet A 610/20-A>","<Blue F 525/50-A>")
system.time(warpSetNCDF(x = fs ,stains = chnls)))
user system elapsed
81.605 1.668 83.321 The enhanced warpSetNCDF based on #14 and #13 takes
system.time(warpSetNCDF(x = fs[,chnls] ,stains = chnls)))
user system elapsed
50.599 2.516 53.649As sample size grows, we should see more significant gains.
raphg
commented
11 years ago Well done Mike!
On Tue, Jun 25, 2013 at 4:17 PM, Mike Jiang notifications@github.comwrote:
normalizing 2 channels for 12 samples:
chnls <- c("<Violet A 610/20-A>","<Blue F 525/50-A>")system.time(warpSetNCDF(x = fs ,stains = chnls))) user system elapsed 81.605 1.668 83.321
The enhanced warpSetNCDF based on #14https://github.com/RGLab/ncdfFlow/issues/14and
13 https://github.com/RGLab/ncdfFlow/issues/13 takes
system.time(warpSetNCDF(x = fs[,chnls] ,stains = chnls)))user system elapsed 50.599 2.516 53.649
As sample size grows, we should see more significant gains.
— Reply to this email directly or view it on GitHubhttps://github.com/RGLab/flowStats/issues/8#issuecomment-20015347 .
mikejiang
commented
11 years ago Since
R has no single precision data type
Thus mode setting to single is unnecessary in ncdfFlow::[[<-.
By removing it, we are able to squeeze out another 5s
user system elapsed
45.527 1.508 47.898 Soon the time will be negative if you keep going Mike.
On Wed, Jun 26, 2013 at 2:09 PM, Mike Jiang notifications@github.comwrote:
Since
R has no single precision data type
Thus mode setting to single is unnecessary in ncdfFlow::[[<-.
By removinghttps://github.com/RGLab/ncdfFlow/commit/94432f1cdb7b886398a9c6951d6c3f180fddc10cit, we are able to squeeze out another 5s
user system elapsed 45.527 1.508 47.898
— Reply to this email directly or view it on GitHubhttps://github.com/RGLab/flowStats/issues/8#issuecomment-20080229 .
mikejiang
commented
11 years ago Unfortunately I can't go any further, given that the current major bottleneck is at fda::landmarkreg call :
$by.total
total.time total.pct self.time self.pct
"system.time" 50.22 100.00 0.00 0.00
"<Anonymous>" 49.88 99.32 0.92 1.83
"eval" 37.16 73.99 0.04 0.08
"capture.output" 34.14 67.98 0.00 0.00
"evalVis" 34.14 67.98 0.00 0.00
"withVisible" 34.14 67.98 0.00 0.00
"landmarkreg" 34.12 67.94 0.00 0.00
"smooth.morph" 32.60 64.91 0.04 0.08
There are lots of processes within this routine involve cdf reading:
Most of them are channel-wise operations. With the enhancement from ncdfFlow#13, we can potentially greatly speed normalization by passing a subsetted
ncdfFlowSetobject .Also we want to change the
ncdfFlow::writeSliceC API to only write normalized channels.