For the APB data, it looks like the boxplot shows some APB area data, but the heatmap doesn't show it. I should take a closer look at this.

There are some bivariate distributions where it's useful to segment the halves, and find out where this occurs. There are a few options to consider here.

1. Show all extrema
2. Show all minima
3. Show all minima bounded by maxima
4. Show points of high concavity (it might be that the bivariate distribution isn't segmented by a minimum, and show instead be segmented by a weaker change.)

It almost looks like the left end of the distribution has been moved to the right part of the graph. Weird.

Maybe this is the right behavior. If you reference a particular dataset, then run a new analysis, perhaps you should give it a new id. You'll need to handle any problems this might cause, however. If you hardcode a reference to a plate, then change its ID, that may break some visualizations.

of APBs / (# of FITC spots + # of TxRed spots)

of Coloc spots / (# of FITC spots + # of TxRed spots)

Area of Coloc spots / Area of FITC-TxRed spot x 100

Normalized APB spots

Normalized Coloc spots

Normalized Coloc area

There's an alternative to the heatmap I'll call the bar map. The layout of the plot is the same for each, but, instead of each cell having a color, each cell has a bar with a representative width. Apparently this is an option in Excel.

For the clustering, I've filled some nans with zeros. I think this is a reasonable decision, but I'd like to avoid being mislead, so it might be good to explicitly call out which cells were handled this way. It might also be good to have an option not to fill the nans with zeros.

in set cell phase notebook there is a execfile on a file called display_utils.py but that file is not present.

This isn't high priority, but it might be nice for the bandwidth values to match across my UI and my analysis with Scipy. Right now, it looks like the bandwidth number I provide to the Scipy.kde function is normalized between 0 and 1, while that's not the case for the seaborn plot.

I don't necessary need to reimplement anything, but I would like there to be a clear mapping between the two.

Right now, there are multiple cell types represented in each pane of the factor plot. It would be good to be able to show only those conditions you'd like.

For now, I think I just need to write a python script making use of sh.py.

The script needs to:

1. SSH into remote machine
2. Stop IPython server
3. Download latest code (and get rid of any local changes)
4. Restart IPython server

In the well-plate plot, there's an option to show values. Since I'm limiting by decimal places, rather than significant digits, numbers much smaller than zero just show up as 0.0.

Figures need titles when they're pasted elsewhere outside notebook.

You should be able to select a reference parameter and see how other parameters correlate.

This could be done as a bar chart, an array of bar charts, or a heatmap.

It might also be good to show things on the cell level, rather than the aggregate level.

Shouldn't the frequencies always be below one?
Haroldo brought this up because some distributions were hard to compare, since they're magnitudes were so different. This may just be a separate problem which could require rescaling.

Right now, there's a section to show show a plot for each parameter with conditions overlaid on one another. There should be another view with a plot for each condition, with parameters overlaid on one another.

There should be an option while showing distributions to have a 'kernel' or a 'histogram'.

This is probably especially import for the case where there is a normal distribution, plus a large number of zeros in a dataset.

Ideally, the experiment could run, the post analysis and visualizations would be prepared, and then anyone who wanted to could just get a text message when everything was done.

It should be possible to make things like new normalized columns very easily without code changes.

Right now, I manually remove extraneous info that's found in many column names.
Many columns begin with 'Cell:' or end with (plate name).

I suspect this extraneous information, and its format, don't change that often, but it might be nice to have a more general solution. I think it's just an interesting and fun problem to work on.

May need to limit how many things can be selected.
Could also set some kind of timeout to kill processes that take too long.

This won't be generated from the raw data, but will instead be a product of the well-plate data (which generally only has 4 wells per condition.)

Possible solution: add special case in distribution plot to use well plate summary instead of raw cell data for plotting.

Create a good list, excluding things that shouldn't be plotted, and also order the list once in a good way. Everything is haphazard right now.

Right now, the urls are a bit random, and the login process looks suspicious, and takes longer than it should.

It should be possible to rename columns and parameters and such. This is especially important for plotting. It might even be a decent intermediate measure to add some inputs to control axis labels, titles, and legend labels.