Re: [matplotlib-devel] path simplification can decrease the smoothness of data plots a
Thanks for the note on this. It's nice to know who wrote the original version. I'll add a note about this in the code comments. I'm not seeing a noticable change in this regard between 0.98.5 (which uses a pretty direct refactoring of your code) to the SVN trunk. The trunk does two things rather differently 1) it only ever returns points that exist in the original data, and 2) it clips line segments at the boundary of the plot. The latter is to get around a shortcoming of Agg (and Abode Reader, for that matter) when plotting lines to very high-valued coordinates. But, I'd appreciate you having a comparison look yourself, in case you're seeing some detail that I'm missing. Cheers, Mike Allan Haldane wrote: a writes: Michael Droettboom md...@... writes: Thanks for the pointers. The original simplification code was written by John Hunter (I believe), and I don't know if it was designed by him also or is a replication of something published elsewhere. So I take no credit for and have little knowledge of its original goals. I'm not sure on everything it does, but it seems to do clipping and removes line segments where the change in slope is less than some limit. There are probably better algorithms out there, but this one works surprisingly well and is fast and simple. I think it should be a requirement that it returns points which are a subset of the original points- with the change you've made it does this, right? Oh Hey! I'm the one who originally wrote the path simplification code. I'd have thought it would be gone by now, but I am very happy it turned out to be useful. I made it up in order to plot a very large set of noisy data I had. The goal was to simplify two types of plots at once: Smooth curves, as well as very noisy data where many lines are 'on top' of each other. (eg plot(rand(10)) ). I noticed both could be taken care of by checking for changes in slope. An important goal (for me) was making sure that the min/max span of the points plotted was preserved. (so that eg plot(rand(1000)) spans from the lowest to highest point in the data (ie ~ 0 to 1) for any zoom factor). I'm not sure if this property survived...: If you do plot(rand(1000)) with the latest matplotlib and gradually zoom out on the x axis, you can see the top/bottom tips of the plotted line flickering in height, which is what I was trying to avoid. I forget whether I actually got it as I wanted it though, maybe I gave up. Allan -- Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA -OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise -Strategies to boost innovation and cut costs with open source participation -Receive a $600 discount off the registration fee with the source code: SFAD http://p.sf.net/sfu/XcvMzF8H ___ Matplotlib-devel mailing list Matplotlib-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/matplotlib-devel -- Michael Droettboom Science Software Branch Operations and Engineering Division Space Telescope Science Institute Operated by AURA for NASA -- Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA -OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise -Strategies to boost innovation and cut costs with open source participation -Receive a $600 discount off the registration fee with the source code: SFAD http://p.sf.net/sfu/XcvMzF8H ___ Matplotlib-devel mailing list Matplotlib-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
Re: [matplotlib-devel] path simplification can decrease the smoothness of data plots a
a writes: Michael Droettboom md...@... writes: Thanks for the pointers. The original simplification code was written by John Hunter (I believe), and I don't know if it was designed by him also or is a replication of something published elsewhere. So I take no credit for and have little knowledge of its original goals. I'm not sure on everything it does, but it seems to do clipping and removes line segments where the change in slope is less than some limit. There are probably better algorithms out there, but this one works surprisingly well and is fast and simple. I think it should be a requirement that it returns points which are a subset of the original points- with the change you've made it does this, right? Oh Hey! I'm the one who originally wrote the path simplification code. I'd have thought it would be gone by now, but I am very happy it turned out to be useful. I made it up in order to plot a very large set of noisy data I had. The goal was to simplify two types of plots at once: Smooth curves, as well as very noisy data where many lines are 'on top' of each other. (eg plot(rand(10)) ). I noticed both could be taken care of by checking for changes in slope. An important goal (for me) was making sure that the min/max span of the points plotted was preserved. (so that eg plot(rand(1000)) spans from the lowest to highest point in the data (ie ~ 0 to 1) for any zoom factor). I'm not sure if this property survived...: If you do plot(rand(1000)) with the latest matplotlib and gradually zoom out on the x axis, you can see the top/bottom tips of the plotted line flickering in height, which is what I was trying to avoid. I forget whether I actually got it as I wanted it though, maybe I gave up. Allan -- Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA -OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise -Strategies to boost innovation and cut costs with open source participation -Receive a $600 discount off the registration fee with the source code: SFAD http://p.sf.net/sfu/XcvMzF8H ___ Matplotlib-devel mailing list Matplotlib-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
Re: [matplotlib-devel] path simplification can decrease the smoothness of data plots
I've checked this change into SVN so others can test it out. Assuming we don't discover any cases where this is clearly inferior, it should make it into the next major release. Mike Andrew Hawryluk wrote: -Original Message- From: Michael Droettboom [mailto:md...@stsci.edu] Sent: 16 Jan 2009 1:31 PM To: Andrew Hawryluk Cc: matplotlib-devel@lists.sourceforge.net Subject: Re: [matplotlib-devel] path simplification can decrease the smoothness of data plots Michael Droettboom wrote: ... I've attached a patch that will only include points from the original data in the simplified path. I hesitate to commit it to SVN, as these things are very hard to get right -- and just because it appears to work better on this data doesn't mean it doesn't create a regression on something else... ;) That said, it would be nice to confirm that this solution works, because it has the added benefit of being a little simpler computationally. Be sure to blitz your build directory when testing the patch -- distutils won't pick it up as a dependency. I've attached two PDFs -- one with the original (current trunk) behavior, and one with the new behavior. I plotted the unsimplified plot in thick blue behind the simplified plot in green, so you can see how much deviation there is between the original data and the simplified line (you'll want to zoom way in with your PDF viewer to see it.) I've also included a new version of your test script which detects new data values in the simplified path, and also seeds the random number generator so that results are comparable. I also set the solid_joinstyle to round, as it makes the wiggliness less pronounced. (There was another thread on this list recently about making that the default setting). Cheers, Mike -- Michael Droettboom Science Software Branch Operations and Engineering Division Space Telescope Science Institute Operated by AURA for NASA Thanks for looking into this! The new plot is much improved, and the simplified calculations are a pleasant surprise. I was also testing the previous algorithm with solid_joinstyle set to round as it is the default in my matplotlibrc. I am probably not able to build your patch here, unless building matplotlib from source on Windows is easier than I anticipate. May I send you some data off the list for you to test? Regards, Andrew NOVA Chemicals Research Technology Centre Calgary, Canada -- Michael Droettboom Science Software Branch Operations and Engineering Division Space Telescope Science Institute Operated by AURA for NASA -- This SF.net email is sponsored by: SourcForge Community SourceForge wants to tell your story. http://p.sf.net/sfu/sf-spreadtheword ___ Matplotlib-devel mailing list Matplotlib-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
Re: [matplotlib-devel] path simplification can decrease the smoothness of data plots
Michael Droettboom md...@... writes: I've checked this change into SVN so others can test it out. Assuming we don't discover any cases where this is clearly inferior, it should make it into the next major release. Mike Hi, This change looks good- it has the advantage of choosing points that actually lie on the curve, which is better visually, and would seem to be a better solution for publication quality plots. The method for simplifying the paths is quite simple and effective, but a bit crude- there are other algorithms you might look into for simplifying lines: http://en.wikipedia.org/wiki/Ramer-Douglas-Peucker_algorithm This one is fairly simple to implement and has the advantage that you have some control over the errors- the deviation from your simplified path and the actual path. Also, you might consider to make the path simplification tolerance (perdNorm2) an adjustable parameter in the matplotlibrc file: #src/agg_py_path_iterator.h //if the perp vector is less than some number of (squared) //pixels in size, then merge the current vector if (perpdNorm2 (1.0 / 9.0)) kind regards, a -- This SF.net email is sponsored by: SourcForge Community SourceForge wants to tell your story. http://p.sf.net/sfu/sf-spreadtheword ___ Matplotlib-devel mailing list Matplotlib-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
Re: [matplotlib-devel] path simplification can decrease the smoothness of data plots
a wrote: Michael Droettboom md...@... writes: I've checked this change into SVN so others can test it out. Assuming we don't discover any cases where this is clearly inferior, it should make it into the next major release. Mike Hi, This change looks good- it has the advantage of choosing points that actually lie on the curve, which is better visually, and would seem to be a better solution for publication quality plots. The method for simplifying the paths is quite simple and effective, but a bit crude- there are other algorithms you might look into for simplifying lines: http://en.wikipedia.org/wiki/Ramer-Douglas-Peucker_algorithm This one is fairly simple to implement and has the advantage that you have some control over the errors- the deviation from your simplified path and the actual path. Thanks for the pointers. The original simplification code was written by John Hunter (I believe), and I don't know if it was designed by him also or is a replication of something published elsewhere. So I take no credit for and have little knowledge of its original goals. However, IMHO the primary purpose of the path simplification in matplotlib is to improve interactive performance (and smaller file size is just an convenient side effect of that), I would hesitate to use an algorithm that is any worse than O(n), since it must be recalculated on every pan or zoom since the simplification is related to *pixels* not data units. Even on modern hardware, it is a constant battle keeping the inner drawing loop fast enough. We could, of course, make the choice of algorithm user-configurable, or use something more precise when using a non-interactive backend, but then we would have two separate code paths to keep in sync and bug free --- not a choice I take lightly. The trick with the present algorithm is to keep the error rate at the subpixel level through the correct selection of perpdNorm. It seems to me that the more advanced simplification algorithm is only necessary when you want to simplify more aggressively than the pixel level. But what hasn't been done is a proper study of the error rate along the simplified path of the current approach vs. other possible approaches. Even this latest change was verified by just looking at the results which seemingly are better on the data I looked at. So I'm mostly speaking from my gut rather than evidence here. Also, you might consider to make the path simplification tolerance (perdNorm2) an adjustable parameter in the matplotlibrc file: #src/agg_py_path_iterator.h //if the perp vector is less than some number of (squared) //pixels in size, then merge the current vector if (perpdNorm2 (1.0 / 9.0)) That sounds like a good idea. I'll have a look at doing that. Mike -- Michael Droettboom Science Software Branch Operations and Engineering Division Space Telescope Science Institute Operated by AURA for NASA -- This SF.net email is sponsored by: SourcForge Community SourceForge wants to tell your story. http://p.sf.net/sfu/sf-spreadtheword ___ Matplotlib-devel mailing list Matplotlib-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
Re: [matplotlib-devel] path simplification can decrease the smoothness of data plots
Michael Droettboom md...@... writes: Thanks for the pointers. The original simplification code was written by John Hunter (I believe), and I don't know if it was designed by him also or is a replication of something published elsewhere. So I take no credit for and have little knowledge of its original goals. I'm not sure on everything it does, but it seems to do clipping and removes line segments where the change in slope is less than some limit. There are probably better algorithms out there, but this one works surprisingly well and is fast and simple. I think it should be a requirement that it returns points which are a subset of the original points- with the change you've made it does this, right? However, IMHO the primary purpose of the path simplification in matplotlib is to improve interactive performance (and smaller file size is just an convenient side effect of that), I would hesitate to use an algorithm that is any worse than O(n), since it must be recalculated on every pan or zoom since the simplification is related to *pixels* not data units. Even on modern hardware, it is a constant battle keeping the inner drawing loop fast enough. We could, of course, make the choice of algorithm user-configurable, or use something more precise when using a non-interactive backend, but then we would have two separate code paths to keep in sync and bug free --- not a choice I take lightly. I see your point. I originally encountered a problem when preparing a pdf figure- I had a lot of high resolution data, and with path simplification the resulting pdf looked pretty bad (the lines were jagged). But the advantage was a massive reduction in file size of the pdf. I adjusted perpdNorm2 and got much better results. The trick with the present algorithm is to keep the error rate at the subpixel level through the correct selection of perpdNorm. It seems to me that the more advanced simplification algorithm is only necessary when you want to simplify more aggressively than the pixel level. But what hasn't been done is a proper study of the error rate along the simplified path of the current approach vs. other possible approaches. Even this latest change was verified by just looking at the results which seemingly are better on the data I looked at. So I'm mostly speaking from my gut rather than evidence here. #src/agg_py_path_iterator.h //if the perp vector is less than some number of (squared) //pixels in size, then merge the current vector if (perpdNorm2 (1.0 / 9.0)) That sounds like a good idea. I'll have a look at doing that. Right, perhaps the best thing to do is make the tolerance parameter adjustable, so it can be reduced to speed up drawing in the interactive backends, but it can also be easily bumped up for extra resolution in the non-interactive backends like pdf/ps. Mike a -- This SF.net email is sponsored by: SourcForge Community SourceForge wants to tell your story. http://p.sf.net/sfu/sf-spreadtheword ___ Matplotlib-devel mailing list Matplotlib-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
Re: [matplotlib-devel] path simplification can decrease the smoothness of data plots
Thanks for looking into this! The new plot is much improved, and the simplified calculations are a pleasant surprise. I was also testing the previous algorithm with solid_joinstyle set to round as it is the default in my matplotlibrc. I am probably not able to build your patch here, unless building matplotlib from source on Windows is easier than I anticipate. May I send you some data off the list for you to test? No problem. I'd also want testing from others -- there aren't a lot of examples in matplotlib itself where simplification even kicks in. Mike -- Michael Droettboom Science Software Branch Operations and Engineering Division Space Telescope Science Institute Operated by AURA for NASA -- This SF.net email is sponsored by: SourcForge Community SourceForge wants to tell your story. http://p.sf.net/sfu/sf-spreadtheword ___ Matplotlib-devel mailing list Matplotlib-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
Re: [matplotlib-devel] path simplification can decrease the smoothness of data plots
-Original Message- From: Michael Droettboom [mailto:md...@stsci.edu] Sent: 16 Jan 2009 1:31 PM To: Andrew Hawryluk Cc: matplotlib-devel@lists.sourceforge.net Subject: Re: [matplotlib-devel] path simplification can decrease the smoothness of data plots Michael Droettboom wrote: ... I've attached a patch that will only include points from the original data in the simplified path. I hesitate to commit it to SVN, as these things are very hard to get right -- and just because it appears to work better on this data doesn't mean it doesn't create a regression on something else... ;) That said, it would be nice to confirm that this solution works, because it has the added benefit of being a little simpler computationally. Be sure to blitz your build directory when testing the patch -- distutils won't pick it up as a dependency. I've attached two PDFs -- one with the original (current trunk) behavior, and one with the new behavior. I plotted the unsimplified plot in thick blue behind the simplified plot in green, so you can see how much deviation there is between the original data and the simplified line (you'll want to zoom way in with your PDF viewer to see it.) I've also included a new version of your test script which detects new data values in the simplified path, and also seeds the random number generator so that results are comparable. I also set the solid_joinstyle to round, as it makes the wiggliness less pronounced. (There was another thread on this list recently about making that the default setting). Cheers, Mike -- Michael Droettboom Science Software Branch Operations and Engineering Division Space Telescope Science Institute Operated by AURA for NASA Thanks for looking into this! The new plot is much improved, and the simplified calculations are a pleasant surprise. I was also testing the previous algorithm with solid_joinstyle set to round as it is the default in my matplotlibrc. I am probably not able to build your patch here, unless building matplotlib from source on Windows is easier than I anticipate. May I send you some data off the list for you to test? Regards, Andrew NOVA Chemicals Research Technology Centre Calgary, Canada -- This SF.net email is sponsored by: SourcForge Community SourceForge wants to tell your story. http://p.sf.net/sfu/sf-spreadtheword ___ Matplotlib-devel mailing list Matplotlib-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
[matplotlib-devel] path simplification can decrease the smoothness of data plots
I'm really excited about the new path simplification option for vector output formats. I tried it the first time yesterday and reduced a PDF from 231 kB to 47 kB. Thanks very much for providing this feature! However, I have noticed that the simplified paths often look more jagged than the original, at least for my data. I can recreate the effect with the following: [start] import numpy as np import matplotlib.pyplot as plt x = np.arange(-3,3,0.001) y = np.exp(-x**2) + np.random.normal(scale=0.001,size=x.size) plt.plot(x,y) plt.savefig('test.png') plt.savefig('test.pdf') [end] A sample output is attached, and close inspection shows that the PNG is a smooth curve with a small amount of noise while the PDF version has very noticeable changes in direction from one line segment to the next. test.png test.pdf The simplification algorithm (agg_py_path_iterator.h) does the following: If line2 is nearly parallel to line1, add the parallel component to the length of line1, leaving it direction unchanged which results in a new data point, not contained in the original data. Line1 will continue to be lengthened until it has deviated from the data curve enough that the next true data point is considered non-parallel. The cycle then continues. The result is a line that wanders around the data curve, and only the first point is guaranteed to have existed in the original data set. Instead, could the simplification algorithm do: If line2 is nearly parallel to line1, combine them by removing the common point, leaving a single line where both end points existed in the original data Thanks again, Andrew Hawryluk attachment: test.png test.pdf Description: test.pdf -- This SF.net email is sponsored by: SourcForge Community SourceForge wants to tell your story. http://p.sf.net/sfu/sf-spreadtheword___ Matplotlib-devel mailing list Matplotlib-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/matplotlib-devel