sg-s/srinivas.gs_mtools

A collection of small functions and MATLAB tools, some that I have written.

srinivas.gs_mtools's Issues

Bug in PrettyFig

with EqualiseY=1, PrettyFig tries to equalise all the axes, even when they have very different types of data. For example, if two plots have line series, and two other plots have colormaps, PrettyFig should only equalise the line series.

TrialPlot should accept arbitrary colours

readPref is not robust

the called func readPref.m is not robust. It should not expect a prefs file to exist in the same location as the main function; further if it can't find a prefs file it should create an empty or a default set of parameters.

FitModel2Data should read lower and upper bounds from the model file

and automatically apply those bounds while optimizing

improvement to manualCluster

instead of plotting crosses corresponding to each point, why not plot the original waveforms?

this is a much richer display. like figure 7 in the original t-sne paper

labelFigure position is inconsistent for different sizes plots

because it calculates it position as a fraction of the plot size (should it be a constant?)

add the following algorithms to spiketimes2f

1. a Gaussian kernel
2. CDF-based firing rates. compute the CDF by spike counting, then differnetiate that based on the required timestep.
3. ISI-based firing rates.

Manipulate does not recognise given `p` as a "saved state"

We should be able to go back to the original case.

getDependencies does not look inside @class folders

when they are not on the path (and they typically aren't)

Write sem

to compute the standard error of the mean for a matrix

pFooter doesn't see classes that are inside @folders

labelFigure labels tend to be hidden by plots, or other obscured

problem cropped up in R2017a

Manipulate does not read bounds correctly from models

install should try to use git if possible

FitFitler2Data returns all NaN K for some vectors

strange.

Improvement to Manipulate

Manipulate should save and load mulitple states, corresponding to different sets of parameters.

To do this, p is now a structure array, and each element corresponds to a new state. The Name of the state is stored in p.Name

sweetspot is broken

And when writing it, we need to keep this in mind:

https://scottsievert.github.io/blog/2015/04/23/image-sqrt/

FitModel2Data should work with 2 arguments only

and intelligently figure out the parameter structure using getModelParameters

makePDF fails when there is a space in your path to the script to be made

spin off FitModel2Data to its own repo

Improvements to Manipulate.m

1. show function name being manipulated in the title bar
2. link the two windows so that closing one closes the other (as in Kontroller)
3. warning when model evaluation takes too long
4. package XJW's model neuron as an example
5. branch off in its own repo

Internally cache all functions

Now that we have cache.m running, we can internally cache all functions, so that they will refuse to do any computations till absolutely required.

Bug in PrettyFig

X, Y labels (and ticks) on log scale are too sparse.

mctsne breaks when path it is in contains a space

spin off FitFilter2Data to its own repoistory

it's now grown up, and has the following methods:
-least squares

• reverse correlation
• transfer function estimation
• parametric fits

update to fitmodel2data

it should be able to handle fitting to multiple data sets at once... simple extend data to be a structure.

raster2 cosmetics

leave small gap between subsequent trials.

lineWrite doesn't work on windows

FitModel2Data should be internally cached

and it should use the cache to find a starting point

prettyFig gives an error when run on an empty plot

or more specifically, on a plot with only NaN values

the expected behaviour is that it should simply ignore it

improvements to mean2

when mean2 is called with a matrix, and there are NaNs scattered in the matrix, mean2 should intelligently average so that NaNs are ignored.

switch prettyFig to the new options handling parser

switch to inputParser for all functions

https://www.mathworks.com/help/matlab/ref/inputparser-class.html

Improve rsquare.m

it captures the coefficeint of determination, but we also want to know is vector A and B are exactly alike. when we fit lines to get the r^2, also get the slope and report it.

labelFigure breaks when using plotyy

it thinks that the individual axes in a plotyy figure are two different axes

Manipulate should support simultaneous equation solving

the idea is we use the parameter structure p to pass the coefficients in the equations, and the function file that is being manipulated contains the problem specification. So no need for a stimulus at all.

improvement to `cache.m`

cache should also store a last-accessed time. so that we can prune the cached data intelligently

FitModel2Data should need 2 core arguments, and all other arguments should be specified in name-value pairs

this will allow greater control on what it does, and allow easy manipulation of internal parameters.

splinehist crashes when given long vectors

it has trouble fitting splines to very long vectors. there are no plans to fix this bug.
don't call it with arguments that are longer than 1e4 elements

make this improvement to makePDF:

instead of using the normal tex output from matlab, modify it so that:

\hspace*{-1in}
\includegraphics [width=\paperwidth]{paper_01.pdf}

this makes the figures bigger

DesignFig needs some more work

1. the code generation is a bit off. add a clf; a hold on, and output the axes handles to something usable.
2. add a delete plot funciton.

FitModel2Data should work even with no input intial parameters

it can get a good guess using getModelParameters

manualCluster is very slow because it redraws everything.

labelFigure defaults are sub-optimal

x_offset should be 0

add this to the repo

https://districtdatalabs.silvrback.com/time-maps-visualizing-discrete-events-across-many-timescales

Improvements to DesignFig

it should generate code like this:

fig_handle=figure('Units','pixels','outerposition',[100 143 1241 657],'Color','w','PaperUnits','points','PaperSize',[1241 657],'Toolbar','none','Menubar','none');

which involves some painful escaping and evaling

maybe look into this?

http://www.mathworks.com/help/matlab/ref/matfile.html

to speed up cache

lineFind is pretty useless

it should also search for substrings within each line -- that's what the real use case is

Improvement to Manipulate.m

should support custom upper and lower bounds of parameters defined by user. The way to do this is to make the parameter strucutre p an array, and the second element specifies the minimum, and the third specifies the maximum

FitModel2Data shouldn't spam so many plots

There is an intellegent way to plot: using hashes to make sure that if the same bit of data is being re-run, there is no need to make a new plot.