Daniel suggested that

"As for the code of the covariance, I have been reading the code and it seems
correct to me. The only comments I would have are the following:

Variable n (declared in line 6) is redundant because inside the loop it has
the same value as "i", and outside the loop (lines 22 and 25) has the same
value as "size".
Checking the size in functions "online_covar" and "two_pass_covar" could be
done before the loop. That way, for size=1 the loop wouldn't execute, saving
a few cycles. However the performance impact of this is very small, I don't
think there is any problem as long as these functions are not called
intensively with 1-length vectors (e.g. as base case for a recursive
function).

I would think comparing the result with that obtained by the R function is
indeed sufficient. Maybe another test would be passing a very large matrix
with some wildly different values in it (to facilitate numeric unstability)
but with known covariances. However I suspect the size would have to be VERY
large to cause problems to the R implementation too."

The current way is working on >95% of cases but there are cases that merging on "ObjectNumber" won't work. @shntnu can you elaborate?

Hi!

First of all than you for this wonderful set of tools. This absolutely moves the field miles ahead in ability of new people to get up and going on image analysis.

One things that I found myself needing is a way to replicate your work in the paper. The dataset you provide for training is comprised of 14K cells, rather than the one of 150K from the paper. As well there are many great visualizations you accomplish in the paper that are not shown in some of the vignettes (like heatmaps of pairwise comparisons or heatmaps of plate setup), which can be VERY useful to start with. If you could provide the workflow from the paper with the original dataset, that would be absolutely wonderful!

Thank you!

-Elias

vignettes/feature_distribution_analysis.Rmd

installation in R 3.3.2 using

devtools::install_github("cytomining/cytominergallery", dependencies = TRUE, build_vignettes = TRUE)

fails with the following error log.

Error: processing vignette 'single_cell_analysis.Rmd' failed with diagnostics:
Path does not exist and create = FALSE
In addition: Warning message:
In tools::buildVignettes(dir = ".", tangle = TRUE) :
Files named as vignettes but with no recognized vignette engine:
‘vignettes/feature_distribution_analysis.Rmd’
(Is a VignetteBuilder field missing?)
Execution halted
Error: Command failed (1)

Comment from Shantanu:

Ah that was because we didn't intend for that to be a vignette but
rather a notebook, but somehow overlooked the fact that
install.packages() will try to build it.

vignettes/feature_distribution_analysis.Rmd

vignettes/feature_distribution_analysis.Rmd

Possibly remove outliers first?

vignettes/feature_distribution_analysis.Rmd

vignettes/feature_distribution_analysis.Rmd

Revert #48

The link to the SQLite db dataset file for vignette single_cell_analysis is dead.

Regards,
Ben

robustbase::mc http://svitsrv25.epfl.ch/R-doc/library/robustbase/html/mc.html has a different scale from skewness and so the is_skewed threshold should be changed.

vignettes/feature_distribution_analysis.Rmd

cloned issue from cytomining/cytominer#49

sources: similarity network fusion
http://www.nature.com/nmeth/journal/v11/n3/full/nmeth.2810.html

R implementation
https://github.com/maxconway/SNFtool/blob/master/R/SNF.R

Dependencies in http://r-pkgs.had.co.nz/description.html#dependencies

Install using

devtools::install_github("cytomining/cytominergallery", dependencies = TRUE, build_vignettes = TRUE)

fails with

Quitting from lines 47-63 (single_cell_analysis.Rmd)
Error: processing vignette 'single_cell_analysis.Rmd' failed with diagnostics:
Path does not exist and create = FALSE
Execution halted
Error: Command failed (1)