by

Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
  • May 23 19:20

    rlabbe on master

    Extra parameter in call #337 I… (compare)

  • May 05 00:49

    rlabbe on master

    Improved problem statement #262 (compare)

  • May 05 00:40

    rlabbe on master

    Fixed math for P = FPF^ #274 (compare)

  • May 05 00:32

    rlabbe on master

    fixed description of IMM transi… (compare)

  • May 05 00:21

    rlabbe on master

    typo - missing normalization te… (compare)

  • May 05 00:15

    rlabbe on master

    typo measurement covariance -> … (compare)

  • May 05 00:10

    rlabbe on master

    Print error if required package… (compare)

  • May 04 23:36

    rlabbe on master

    typo Euler ->Taylor #277 (compare)

  • May 04 23:11

    rlabbe on master

    Fixed alignment of measurements… (compare)

  • May 04 22:53

    rlabbe on master

    Use Newton for position equatio… (compare)

  • May 04 22:42

    rlabbe on master

    Fixed terms in Taylor series #3… (compare)

  • May 04 22:35

    rlabbe on master

    Explained row order access of c… (compare)

  • May 04 22:01

    rlabbe on master

    altered velocity covariance pos… (compare)

  • May 04 15:53

    rlabbe on rend

    t2 (compare)

  • May 04 15:49

    rlabbe on rend

    t1 (compare)

  • May 04 01:12

    rlabbe on master

    fixed pylint warnings (compare)

  • May 04 00:30

    rlabbe on master

    Support state addition not usin… (compare)

  • May 04 00:21

    rlabbe on master

    unit test for linear UKF vs EKF… (compare)

  • May 04 00:09

    rlabbe on master

    z reshape bug #160 z cannot be… (compare)

  • May 04 00:00

    rlabbe on master

    fixed get_prediction signature … (compare)

Julia Schwarz
@julenka
Another suggestion is, in chapter 1 function plot_estimate_chart_3(), the residual line is black and is hard to see. I updated my local copy to make it green, though something brighter like magenta could be better. Change line 107 of code/gh_internal.py:
ax.annotate('', xy=[1,159], xytext=[1,164.2], arrowprops=dict(arrowstyle='-', ec='g', lw=1, shrinkA=8, shrinkB=8))
Julia Schwarz
@julenka
Oh, I think I see a downside to using %matplotlib notebook, it seems to prevent generation of other figures in that chapter. Never mind :-)
Roger Labbe
@rlabbe
Yes, I think you need to put %matplotlib notebook into every cell. It seems flakey, sometimes the graph will not regenerate no matter what.
There is a tension that I am trying to balance with the charts - I want more interactivity, but some people use the PDF version of the book. The interactive charting software have not implemented the hooks that allow nbconvert (which converts the notebooks into latex which I then convert to PDF) to extract the graphs. I'm sort of thinking about parsing the notebooks prior to running nbconvert, and replacing the interactive charts with boring old matplotlib, and then executing the cells.
Also, colors are tricky. I'm talking to publishers, and print will be black and white. I keep changing the plot settings to make the plots readable in B&W, but that seems like a bit of a waste of time - I can solve that problem when it comes time.
Roger Labbe
@rlabbe
I'll take a look at the code change you mentioned, and incorporate it if it makes sense in the context of all that. Thanks!
Roger Labbe
@rlabbe
@julenka I found the extra color a bit garish, and I am no designer. I opted to make the line width thicker, and change the grid to dotted lines. I think it is more readable now.
Julia Schwarz
@julenka
@rlabbe dotted lines is a better choice, especially if you're working in black and white. Thanks!
empirical-bayesian
@empirical-bayesian
@rlabbe or anyone listening: Is there a way of getting the log-likelihood of a model on a set of data out from the Kalman filter class? Does the class handle missing data, e.g., by marking missing using Numpy's ma.array? Finally, aside from studying the code, does the implementation use the SVD way of calculation, per R's dlm (see https://hypergeometric.wordpress.com/2015/07/29/comprehensive-and-compact-tutorial-on-petris-dlm-package-in-r/), or square root filter, or some other way? Thank you!
Roger Labbe
@rlabbe

Hi. Those are good additions for the library. Here's the current status:

Missing data is handled by setting z=None. If using batch_filter, you might call it with kf.batch_filter(zs=[1., 2., 3., None, 5.]). That is probably not 'canonical' python behavior, and I will add it to the issues.

I am working on log-likelihood, and metrics like NEES, NIS, etc for the next release of FilterPy.

I do not currently have an SVD filter. It is on the backlog.

The Kalman filter class uses the standard linear Kalman filter equations; this makes it more pedagogical in nature, though I have used it plenty of times in less demanding situations. The only concession I made to real world engineering is in the computation of P - the published (I-KH)P equation is unstable.

A square root filter is implemented by the class SquareRootKalmanFilter, in the filterpy.kalman module. Read the documentation carefully - this is more a reference implementation and i have not used it in production. Brown suggests that square root filters are no longer needed with modern hardware unless P is going to vary by 20 orders of magnitude. His reasoning seems strong, but I do not have empirical evidence to back that up.

To round out the descriptions, there is also a fading memory and information filter implemented for the linear filters. I have an EKF and UKF, but not with the square root variants.

If you want to compute the log-likelihood yourself you can. This link gives the equation for the computation: http://www.econ.umn.edu/~karib003/help/kalman_example1.htm. Their 'C_t' can be accessed with 'kf.S' in my code after calling update().

Roger Labbe
@rlabbe
Here is some code for likelihood. Haven't really tested it
import numpy as np
from scipy.stats import multivariate_normal
from numpy import dot, log, exp
import scipy.linalg as la
def gaus_pdf(X, M, S):
    DX = (X-M)[0,0]
    E = 0.5*np.dot(DX.T, (S/DX))
    d = M.shape[0]
    E = E + 0.5 * d * log(2*np.pi) + 0.5 * log(la.det(S));
    P = exp(-E)
    return P


def kf_liklihood(x, P, z, H, R):
    IM = np.dot(H, x)
    S = np.dot(H, P).dot(H.T) + R
    print(gaus_pdf(z, IM, S))
    print(multivariate_normal.pdf(z, mean=IM, cov=S))
    return multivariate_normal.pdf(z, mean=IM, cov=S)
Here is code for likelihood that I haven't really tested
from scipy.stats import multivariate_normal
def likelihood(x, P, z, H, R):
    IM = np.dot(H, x)
    S = np.dot(H, P).dot(H.T) + R
    return multivariate_normal.pdf(z, mean=IM, cov=S)
Roger Labbe
@rlabbe

FilterPy 0.0.26 changes:

  • Added likelihood and log-likelihood to the KalmanFilter
    class.

  • Added an MMAE filter bank class.

  • Added function to compute NEES

Soheil
@soheil
hmm
Roger Labbe
@rlabbe
FilterPy 0.0.27 changes:
  • Added function to compute update in the presense of
    correlated process and measurement noise.
    • Added IMM filter.
    • added tests for IMM and MMAE filters
    • Added display of semi-axis for covariance ellipses
    • various bug fixes
skyxtwang
@skyxtwang

When I install the packge filterpy,I got the following error:
C:\WINDOWS\system32>conda install --channel https://conda.anaconda.org/phios filterpy
Fetching package metadata: ......
Solving package specifications: ..........
Error: Unsatisfiable package specifications.
Generating hint:
[ COMPLETE ]|##################################################| 100%

Hint: the following packages conflict with each other:

  • filterpy
  • python 3.5*

Use 'conda info filterpy' etc. to see the dependencies for each package.

Note that the following features are enabled:

  • vc14
    I want to know wether the filterpy can be installed with python3.5 and how can solute this problems.Thanks.
Roger Labbe
@rlabbe
Sorry, I jsut saw this. I don't know who phios is, it is some random person that put filterpy on conda. filterpy works in python 3.5 - I do all development on the latest releases of Python 3. Try pip install filterpy instead; that should work. Recently I tried to get filterpy working with conda; conda install filterpy might work for windows, but I have reports it doesn't work for mac and linux yet. I need to put more effort on this.
Mateusz Sadowski
@msadowski
hi @rlabbe! Just started reading your book and it looks great so far! A quick question: would you like me to let you know about any typos I find and if so is this chat room the right place?
Just before I forget here is the first one I spotted: But sitting down and trying to read many of these books is a dismal and trying experience if you do not have the necessary background. [chapter 00]. I think you meant tiring
Just noticed that I marked the wrong trying, I meant the next one dismal and trying experience . Let me know if I should continue doing that or not :). Thanks for the great book!
Roger Labbe
@rlabbe
Hi @msadowski , sure, typo reports are great! You can do it here or type up a github issue, whichever is easier.
Mateusz Sadowski
@msadowski
would it make your life easier if I corrected them as I go and then make commits chapter by chapter and then make pull requests for the changes? I would be quite happy to do that since I'm planning to read the book 'cover to cover'
Roger Labbe
@rlabbe
yes, that is how most people have done it. Once or twice we have had issues where about every line is for some reason marked as a change (I suspect line endings differences between windows and linux), and endup with something like 50,000 lines changes. Of course i had to reject those because there was no reasonable way for me to code/word review before accepting. Just keep the repo up to date and there should be no problem. my pace of change has really slowed down so that shouldn't be an issue
Mateusz Sadowski
@msadowski
Hi rlabbe, Just to let you know: it seems that filterpy is missing again from your binder builld (http://app.mybinder.org/2637687816/notebooks/03-Gaussians.ipynb)
Roger Labbe
@rlabbe
sorry. just saw this. I seem to have it fixed now
Adam Milner
@carmiac
Hi Roger! Thanks for this book, so far my read through it has been very helpful. There is one concept that I am having problems understanding related to a problem I am working on. How does the UKF deal with control inputs to the system? The system I am looking at has a fairly simple A matrix (diagonal, mostly unity), with a full B matrix and a non-linear measurement function. Does the UKF deal with this well?
Kishan
@kishb87
Does anyone know of smoothing techniques that can be used for a EKF? It looks like all the smoothing techniques in FilterPy are linear problems. Am I right?
Suresh Sharma
@sursha
Hi @rlabbe! Just getting started with your book, thanks for your work!
alimohebbi
@alimohebbi
hi
i wonna run a code about robot motion using ekf
code are in the book
but get this erros
File "/usr/local/lib/python2.7/dist-packages/scipy/_lib/_util.py", line 231, in _asarray_validated
raise ValueError('object arrays are not supported')
ValueError: object arrays are not supported
what should i do?
Matthias
@maluethi
@rlabbe: I am reading through your book online and so far it has been amazing, thanks for publishing all this!
Although I did not read all chapters yet, it came to my understanding that outliers (measurements containing no information) are pretty bad for a 'standard' Kalman filter. By just searching trough the pdf version I didn't find any mention of outliers. So I was wondering if you could point me to any resources that introduce outlier handling within kalman filtering? Of course I googled but I cannot really judge the quality for a beginner of all the different results I get back...
Roger Labbe
@rlabbe
@maluethi , sorry, I haven't logged in here for a long time.
I intend to add a section on just this topic in a few day. In the meantime, I suggest looking up 'mahalanobis distance', which is a measure of how far a measurement is from the KF's prior. You can use this to 'gate' your data - discard data that is "too far away". Theory says throw away anything > 3 std, but in practice you may find 4,5, even 6 std to be a better gating distance
If you throw the data away, you just don't call update for that time period. You will thus call predict twice in a row, and your estimate will gain uncertainty because you did 2 predictions in a row
that's the general idea. The search term "kalman filter gating" is also a fruitful search.
Roger Labbe
@rlabbe
@maluethi the book is updated with a section on outliers. see chapter 8.
Matthias
@maluethi
@rlabbe Thanks a lot for letting me know. I will certainly look into the new chapter!
noemecd
@noemecd
I wish to congratulate you for your excellent book on Kalman and Bayesian filters.
It is clear, didactic and well-documented.
I had to build my very first Kalman filter in a quite complex configuration (7 state variables, strong non-linearities and very low signal to noise ratio).
Your book has brought me tremendous help in doing that - although, as a Scilab user, I have found the recourse to Python more troublesome than helpful -.
It is rare to find such a thorough, simple, user-oriented while scientifically sound presentation of the Kalman filter.
I think you are a born pedagogue.
Lots of thanks.
Miguel Oyarzun
@Miguel-O-Matic
@rlabbe I just started working through the Jupyter notebooks. Your intuitive approach to the subject matter is very refreshing. I do have a question/observation about the material in Chapter 3... I find the discussion of the product vs sum of "Gaussians" a bit confusing. It seems that you are discussing the sum of Gaussian random "variables" and the product of Gaussian probability "distributions". The sum of two independent Gaussian random variables is also Gaussian-distributed. The product of two Gaussian random variables is not, in general, Gaussian-distributed.
Miguel Oyarzun
@Miguel-O-Matic

@rlabbe I just started working through the Jupyter notebooks. Your intuitive approach to the subject matter is very refreshing. I do have a question/observation about the material in Chapter 3... I find the discussion of the product vs sum of "Gaussians" a bit confusing. It seems that you are discussing the sum of Gaussian random "variables" and the product of Gaussian probability "distributions". The sum of two independent Gaussian random variables is also Gaussian-distributed. The product of two Gaussian random variables is not, in general, Gaussian-distributed.

Having now made it through Chapter 4... I think the source of the confusion is that, in both cases, we are really talking about operations on Gaussian "distributions" rather than random "variables" . The mathematical operation involved in the "prediction" step is really a convolution, rather than a sum, of Gaussian "distributions", which can be shown to be a Gaussian "distribution" with mean and variance as described in Chapter 3. At least, that's what I think after reading Chapter 4... Looking forward to further enlightenment in the upcoming chapters... :-)

han-so1omon
@han-so1omon
Thank you for writing this book. I hope it is valuable to students, as I wish that I had it when I was a student.
Douglas Daly
@douglas-daly_gitlab
Wow - what an amazing book! You offer a very clear and conceptual approach to these topics without glossing over the foundations. And of course, presenting as Jupyter notebooks is a huge help and removes the mysteries of tuning parameters and such.
gabrielegranello
@gabrielegranello
Thank you very much for providing the library and the book, both have been of immense value for me to understand how Kalman filters work.
dangle1
@dangle1
Is there anyone here interested in forming a study group for this book? I'm working my way through it and understand it for the most part, but think studying together would make learning better.
Rachel Cohen Yeshurun
@rachelyeshurun
Wow, just amazing notebooks, hope one day to work through this. In the meantime, this notebook series is my inspiration for my own work.. Question for you @rlabbe or anyone who knows: I haven't been able to find out how to show all output on startup. I see all your notebooks when rendered on Binder too, show up with all the plots rendered, and no cell output number. It looks so clean, how do you do that??? My cells just show the code. Markdown is rendered, but code is not run, even though I ran all and saved before commit.