I am using scipy.optimize.fmin_l_bfgs_b to solve a gaussian mixture problem. The means of mixture distributions are modeled by regressions whose weights have to be optimized using EM algorithm.
sigma_sp_new, func_val, info_dict = fmin_l_bfgs_b(func_to_minimize, self.sigma_vector[si][pj],
args=(self.w_vectors[si][pj], Y, X, E_step_results[si][pj]),
approx_grad=True, bounds=[(1e-8, 0.5)], factr=1e02, pgtol=1e-05, epsilon=1e-08)
But sometimes I got a warning 'ABNORMAL_TERMINATION_IN_LNSRCH' in the information dictionary:
func_to_minimize value = 1.14462324063e-07
information dictionary: {'task': b'ABNORMAL_TERMINATION_IN_LNSRCH', 'funcalls': 147, 'grad': array([ 1.77635684e-05, 2.87769808e-05, 3.51718654e-05,
6.75015599e-06, -4.97379915e-06, -1.06581410e-06]), 'nit': 0, 'warnflag': 2}
RUNNING THE L-BFGS-B CODE
* * *
Machine precision = 2.220D-16
N = 6 M = 10
This problem is unconstrained.
At X0 0 variables are exactly at the bounds
At iterate 0 f= 1.14462D-07 |proj g|= 3.51719D-05
* * *
Tit = total number of iterations
Tnf = total number of function evaluations
Tnint = total number of segments explored during Cauchy searches
Skip = number of BFGS updates skipped
Nact = number of active bounds at final generalized Cauchy point
Projg = norm of the final projected gradient
F = final function value
* * *
N Tit Tnf Tnint Skip Nact Projg F
6 1 21 1 0 0 3.517D-05 1.145D-07
F = 1.144619474757747E-007
ABNORMAL_TERMINATION_IN_LNSRCH
Line search cannot locate an adequate point after 20 function
and gradient evaluations. Previous x, f and g restored.
Possible causes: 1 error in function or gradient evaluation;
2 rounding error dominate computation.
Cauchy time 0.000E+00 seconds.
Subspace minimization time 0.000E+00 seconds.
Line search time 0.000E+00 seconds.
Total User time 0.000E+00 seconds.
I do not get this warning every time, but sometimes. (Most get 'CONVERGENCE: NORM_OF_PROJECTED_GRADIENT_<=_PGTOL' or 'CONVERGENCE: REL_REDUCTION_OF_F_<=_FACTR*EPSMCH').
I know that it means the minimum can be be reached in this iteration. I googled this problem. Someone said it occurs often because the objective and gradient functions do not match. But here I do not provide gradient function because I am using 'approx_grad'.
What are the possible reasons that I should investigate? What does it mean by "rounding error dominate computation"?
======
I also find that the log-likelihood does not monotonically increase:
########## Convergence !!! ##########
log_likelihood_history: [-28659.725891322563, 220.49993177669558, 291.3513633060345, 267.47745327823907, 265.31567762171181, 265.07311121000367, 265.04217683341682]
It usually start decrease at the second or the third iteration, even through 'ABNORMAL_TERMINATION_IN_LNSRCH' does not occurs. I do not know whether it this problem is related to the previous one.
Scipy calls the original L-BFGS-B implementation. Which is some fortran77 (old but beautiful and superfast code) and our problem is that the descent direction is actually going up. The problem starts on line 2533 (link to the code at the bottom)
gd = ddot(n,g,1,d,1)
if (ifun .eq. 0) then
gdold=gd
if (gd .ge. zero) then
c the directional derivative >=0.
c Line search is impossible.
if (iprint .ge. 0) then
write(0,*)' ascent direction in projection gd = ', gd
endif
info = -4
return
endif
endif
In other words, you are telling it to go down the hill by going up the hill. The code tries something called line search a total of 20 times in the descent direction that you provide and realizes that you are NOT telling it to go downhill, but uphill. All 20 times.
The guy who wrote it (Jorge Nocedal, who by the way is a very smart guy) put 20 because pretty much that's enough. Machine epsilon is 10E-16, I think 20 is actually a little too much. So, my money for most people having this problem is that your gradient does not match your function.
Now, it could also be that "2. rounding errors dominate computation". By this, he means that your function is a very flat surface in which increases are of the order of machine epsilon (in which case you could perhaps rescale the function),
Now, I was thiking that maybe there should be a third option, when your function is too weird. Oscillations? I could see something like $\sin({\frac{1}{x}})$ causing this kind of problem. But I'm not a smart guy, so don't assume that there's a third case.
So I think the OP's solution should be that your function is too flat. Or look at the fortran code.
https://github.com/scipy/scipy/blob/master/scipy/optimize/lbfgsb/lbfgsb.f
Here's line search for those who want to see it. https://en.wikipedia.org/wiki/Line_search
Note. This is 7 months too late. I put it here for future's sake.
As pointed out in the answer by Wilmer E. Henao, the problem is probably in the gradient. Since you are using approx_grad=True, the gradient is calculated numerically. In this case, reducing the value of epsilon, which is the step size used for numerically calculating the gradient, can help.
I also got the error "ABNORMAL_TERMINATION_IN_LNSRCH" using the L-BFGS-B optimizer.
While my gradient function pointed in the right direction, I rescaled the actual gradient of the function by its L2-norm. Removing that or adding another appropriate type of rescaling worked. Before, I guess that the gradient was so large that it went out of bounds immediately.
The problem from OP was unbounded if I read correctly, so this will certainly not help in this problem setting. However, googling the error "ABNORMAL_TERMINATION_IN_LNSRCH" yields this page as one of the first results, so it might help others...
I had a similar problem recently. I sometimes encounter the ABNORMAL_TERMINATION_IN_LNSRCH message after using fmin_l_bfgs_b function of scipy. I try to give additional explanations of the reason why I get this. I am looking for complementary details or corrections if I am wrong.
In my case, I provide the gradient function, so approx_grad=False. My cost function and the gradient are consistent. I double-checked it and the optimization actually works most of the time. When I get ABNORMAL_TERMINATION_IN_LNSRCH, the solution is not optimal, not even close (even this is a subjective point of view). I can overcome this issue by modifying the maxls argument. Increasing maxls helps to solve this issue to finally get the optimal solution. However, I noted that sometimes a smaller maxls, than the one that produces ABNORMAL_TERMINATION_IN_LNSRCH, results in a converging solution. A dataframe summarizes the results. I was surprised to observe this. I expected that reducing maxls would not improve the result. For this reason, I tried to read the paper describing the line search algorithm but I had trouble to understand it.
The line "search algorithm generates a sequence of
nested intervals {Ik} and a sequence of iterates αk ∈ Ik ∩ [αmin ; αmax] according to the [...] procedure". If I understand well, I would say that the maxls argument specifies the length of this sequence. At the end of the maxls iterations (or less if the algorithm terminates in fewer iterations), the line search stops. A final trial point is generated within the final interval Imaxls. I would say the the formula does not guarantee to get an αmaxls that respects the two update conditions, the minimum decrease and the curvature, especially when the interval is still wide. My guess is that in my case, after 11 iterations the generated interval I11 is such that a trial point α11 respects both conditions. But, even though I12 is smaller and still containing acceptable points, α12 is not. Finally after 24 iterations, the interval is very small and the generated αk respects the update conditions.
Is my understanding / explanation accurate?
If so, I would then be surprised that when maxls=12, since the generated α11 is acceptable but not α12, why α11 is not chosen in this case instead of α12?
Pragmatically, I would recommend to try a few higher maxls when getting ABNORMAL_TERMINATION_IN_LNSRCH.
Related
I am trying to minimize a function defined as follows:
utility(decision) = decision * (risk - cost)
where variables take the following form:
decision = binary array
risk = array of floats
cost = constant
I know the solution will take the form of:
decision = 1 if (risk >= threshold)
decision = 0 otherwise
Therefore, in order to minimize this function I can assume that I transform the function utility to depend only on this threshold. My direct translation to scipy is the following:
def utility(threshold,risk,cost):
selection_list = [float(risk[i]) >= threshold for i in range(len(risk))]
v = np.array(risk.astype(float)) - cost
total_utility = np.dot(v, selection_list)
return -1.0*total_utility
result = minimize(fun=utility, x0=0.2, args=(r,c),bounds=[(0,1)], options={"disp":True} )
This gives me the following result:
fun: array([-17750.44298655]) hess_inv: <1x1 LbfgsInvHessProduct with
dtype=float64>
jac: array([0.])
message: b'CONVERGENCE: NORM_OF_PROJECTED_GRADIENT_<=_PGTOL'
nfev: 2
nit: 0 status: 0 success: True
x: array([0.2])
However, I know the result is wrong because in this case it must be equal to cost. On top of that, no matter what x0 I use, it always returns it as the result. Looking at the results I observe that jacobian=0 and does not compute 1 iteration correctly.
Looking more thoroughly into the function. I plot it and observe that it is not convex on the limits of the bounds but we can clearly see the minimum at 0.1. However, no matter how much I adjust the bounds to be in the convex part only, the result is still the same.
What could I do to minimize this function?
The error message tells you that the gradient was at some point too small and thus numerically the same as zero. This is likely due to the thresholding that you do when you calculate your selection_list. There you say float(risk[i]) >= threshold, which has derivative 0 almost everywhere. Hence, almost every starting value will give you the warning you receive.
A solution could be to apply some smoothing to the thresholding operation. So instead of float(risk[i]) >= threshold, you would use a continuous function:
def g(x):
return 1./(1+np.exp(-x))
With this function, you can express the thresholding operation as
g((risk[i] - threshold)/a), which a parameter a. The larger a, the closer is this modified error function to what you are doing so far. At something like a=20 or so, you would probably have pretty much the same that you have at the moment. You would therefore derive a sequence of solutions, where you start with a=1 and then take that solution as a starting value for the same problem with a=2, take that solution as a starting value for the problem with a=4, and so on. At some point, you will notice that changing a does no longer change the solution and you're done.
I use the scipy.optimize.minimize ( https://docs.scipy.org/doc/scipy/reference/tutorial/optimize.html ) function with method='L-BFGS-B.
An example of what it returns is here above:
fun: 32.372210618549758
hess_inv: <6x6 LbfgsInvHessProduct with dtype=float64>
jac: array([ -2.14583906e-04, 4.09272616e-04, -2.55795385e-05,
3.76587650e-05, 1.49213975e-04, -8.38440428e-05])
message: 'CONVERGENCE: REL_REDUCTION_OF_F_<=_FACTR*EPSMCH'
nfev: 420
nit: 51
status: 0
success: True
x: array([ 0.75739412, -0.0927572 , 0.11986434, 1.19911266, 0.27866406,
-0.03825225])
The x value correctly contains the fitted parameters. How do I compute the errors associated to those parameters?
TL;DR: You can actually place an upper bound on how precisely the minimization routine has found the optimal values of your parameters. See the snippet at the end of this answer that shows how to do it directly, without resorting to calling additional minimization routines.
The documentation for this method says
The iteration stops when (f^k - f^{k+1})/max{|f^k|,|f^{k+1}|,1} <= ftol.
Roughly speaking, the minimization stops when the value of the function f that you're minimizing is minimized to within ftol of the optimum. (This is a relative error if f is greater than 1, and absolute otherwise; for simplicity I'll assume it's an absolute error.) In more standard language, you'll probably think of your function f as a chi-squared value. So this roughly suggests that you would expect
Of course, just the fact that you're applying a minimization routine like this assumes that your function is well behaved, in the sense that it's reasonably smooth and the optimum being found is well approximated near the optimum by a quadratic function of the parameters xi:
where Δxi is the difference between the found value of parameter xi and its optimal value, and Hij is the Hessian matrix. A little (surprisingly nontrivial) linear algebra gets you to a pretty standard result for an estimate of the uncertainty in any quantity X that's a function of your parameters xi:
which lets us write
That's the most useful formula in general, but for the specific question here, we just have X = xi, so this simplifies to
Finally, to be totally explicit, let's say you've stored the optimization result in a variable called res. The inverse Hessian is available as res.hess_inv, which is a function that takes a vector and returns the product of the inverse Hessian with that vector. So, for example, we can display the optimized parameters along with the uncertainty estimates with a snippet like this:
ftol = 2.220446049250313e-09
tmp_i = np.zeros(len(res.x))
for i in range(len(res.x)):
tmp_i[i] = 1.0
hess_inv_i = res.hess_inv(tmp_i)[i]
uncertainty_i = np.sqrt(max(1, abs(res.fun)) * ftol * hess_inv_i)
tmp_i[i] = 0.0
print('x^{0} = {1:12.4e} ± {2:.1e}'.format(i, res.x[i], uncertainty_i))
Note that I've incorporated the max behavior from the documentation, assuming that f^k and f^{k+1} are basically just the same as the final output value, res.fun, which really ought to be a good approximation. Also, for small problems, you can just use np.diag(res.hess_inv.todense()) to get the full inverse and extract the diagonal all at once. But for large numbers of variables, I've found that to be a much slower option. Finally, I've added the default value of ftol, but if you change it in an argument to minimize, you would obviously need to change it here.
One approach to this common problem is to use scipy.optimize.leastsq after using minimize with 'L-BFGS-B' starting from the solution found with 'L-BFGS-B'. That is, leastsq will (normally) include and estimate of the 1-sigma errors as well as the solution.
Of course, that approach makes several assumption, including that leastsq can be used and may be appropriate for solving the problem. From a practical view, this requires the objective function return an array of residual values with at least as many elements as variables, not a cost function.
You may find lmfit (https://lmfit.github.io/lmfit-py/) useful here: It supports both 'L-BFGS-B' and 'leastsq' and gives a uniform wrapper around these and other minimization methods, so that you can use the same objective function for both methods (and specify how to convert the residual array into the cost function). In addition, parameter bounds can be used for both methods. This makes it very easy to first do a fit with 'L-BFGS-B' and then with 'leastsq', using the values from 'L-BFGS-B' as starting values.
Lmfit also provides methods to more explicitly explore confidence limits on parameter values in more detail, in case you suspect the simple but fast approach used by leastsq might be insufficient.
It really depends what you mean by "errors". There is no general answer to your question, because it depends on what you're fitting and what assumptions you're making.
The easiest case is one of the most common: when the function you are minimizing is a negative log-likelihood. In that case the inverse of the hessian matrix returned by the fit (hess_inv) is the covariance matrix describing the Gaussian approximation to the maximum likelihood.The parameter errors are the square root of the diagonal elements of the covariance matrix.
Beware that if you are fitting a different kind of function or are making different assumptions, then that doesn't apply.
I have implemented an algorithm that uses two other algorithms for calculating the shortest path in a graph: Dijkstra and Bellman-Ford. Based on the time complexity of the these algorithms, I can calculate the running time of my implementation, which is easy giving the code.
Now, I want to experimentally verify my calculation. Specifically, I want to plot the running time as a function of the size of the input (I am following the method described here). The problem is that I have two parameters - number of edges and number of vertices.
I have tried to fix one parameter and change the other, but this approach results in two plots - one for varying number of edges and the other for varying number of vertices.
This leads me to my question - how can I determine the order of growth based on two plots? In general, how can one experimentally determine the running time complexity of an algorithm that has more than one parameter?
It's very difficult in general.
The usual way you would experimentally gauge the running time in the single variable case is, insert a counter that increments when your data structure does a fundamental (putatively O(1)) operation, then take data for many different input sizes, and plot it on a log-log plot. That is, log T vs. log N. If the running time is of the form n^k you should see a straight line of slope k, or something approaching this. If the running time is like T(n) = n^{k log n} or something, then you should see a parabola. And if T is exponential in n you should still see exponential growth.
You can only hope to get information about the highest order term when you do this -- the low order terms get filtered out, in the sense of having less and less impact as n gets larger.
In the two variable case, you could try to do a similar approach -- essentially, take 3 dimensional data, do a log-log-log plot, and try to fit a plane to that.
However this will only really work if there's really only one leading term that dominates in most regimes.
Suppose my actual function is T(n, m) = n^4 + n^3 * m^3 + m^4.
When m = O(1), then T(n) = O(n^4).
When n = O(1), then T(n) = O(m^4).
When n = m, then T(n) = O(n^6).
In each of these regimes, "slices" along the plane of possible n,m values, a different one of the terms is the dominant term.
So there's no way to determine the function just from taking some points with fixed m, and some points with fixed n. If you did that, you wouldn't get the right answer for n = m -- you wouldn't be able to discover "middle" leading terms like that.
I would recommend that the best way to predict asymptotic growth when you have lots of variables / complicated data structures, is with a pencil and piece of paper, and do traditional algorithmic analysis. Or possibly, a hybrid approach. Try to break the question of efficiency into different parts -- if you can split the question up into a sum or product of a few different functions, maybe some of them you can determine in the abstract, and some you can estimate experimentally.
Luckily two input parameters is still easy to visualize in a 3D scatter plot (3rd dimension is the measured running time), and you can check if it looks like a plane (in log-log-log scale) or if it is curved. Naturally random variations in measurements plays a role here as well.
In Matlab I typically calculate a least-squares solution to two-variable function like this (just concatenates different powers and combinations of x and y horizontally, .* is an element-wise product):
x = log(parameter_x);
y = log(parameter_y);
% Find a least-squares fit
p = [x.^2, x.*y, y.^2, x, y, ones(length(x),1)] \ log(time)
Then this can be used to estimate running times for larger problem instances, ideally those would be confirmed experimentally to know that the fitted model works.
This approach works also for higher dimensions but gets tedious to generate, maybe there is a more general way to achieve that and this is just a work-around for my lack of knowledge.
I was going to write my own explanation but it wouldn't be any better than this.
My problem is explained in the following image
http://i.stack.imgur.com/n6mZt.png
I have a finite (but rather large) amount of such pieces that need to be stacked in a way so that the REMAINING area is the smallest possible. The pieces are locked in the horizontal axis (time) and have fixed height. They can only be stacked.
The remaining area is defined by the maximum point of the stack that depends on which pieces have been selected. The best combination in the example image would be the [1 1 0]. (The trivial [0 0 0] case will not be allowed by other constraints)
My only variables are binaries (Yes or No) for each piece. The objective is a little more complicated than what I am describing, but my greatest problem right now is how to formulate the expression
Max{Stacked_Pieces} - Stacked_Pieces_Profile
in the objective function. The result of this expression is a vector of course (timeseries) but it will be further reduced to a number through other manipulations.
Essentially my problem is how to write
Max{A} - A, where A = 1xN vector
In a way compatible with a linear (or even quadratic) objective. Or am I dealing with a non-linear problem?
EDIT: The problem is like a Knapsack problem the main difference being that there is no knapsack to fill up. i.e. the size of the knapsack varies according to the selected pieces and is always equal to the top of the stacked profile
Thanks everybody!
From what I understand you can basically try to solve it as a normal knapsack problem in multiple iterations, finding the minimal.
Now, finding the height of the knapsack is a problem, which means you need multiple iterations. Because you need to solve the knapsack problem to see if a certain height will work, you need multiple iterations.
Note that you do know an upper and a lower bound for the height. I'm not sure if rotation is applicable, but you can fill in the gaps here:
Min = max(max height of smallest piece, total size / width)
Max = sum(height of all pieces).
Basically solving it means finding the smallest height [Min <= x <= Max] that fits all pieces. The easiest way to do that is by using a 'for' loop, but you can do it better:
Try min, max, half
if half fits -> max = half; iterate (goto 1)
if half doesn't fit -> min = half; iterate (goto 1)
As for solving the knapsack problem, for each iteration, I'd check if all pieces can still be fitted. Use bit-masks and AND/OR/XOR operations if you can to speed things up.
Basically you can do it like this:
Grab bit 'x'. Fill with next block
Check if this leads to a possible solution
Find next bit that can be filled
Note that you might want to use intrinsics in C++ to speed this up. Modern CPU's are quite good with this.
As for code: I've made some code that solves the bedlam cube in the past; I'm pretty sure that if you google for that, you'll find some fast solvers.
Good luck!
I've got a GPS track produced by gpxlogger(1) (supplied as a client for gpsd). GPS receiver updates its coordinates every 1 second, gpxlogger's logic is very simple, it writes down location (lat, lon, ele) and a timestamp (time) received from GPS every n seconds (n = 3 in my case).
After writing down a several hours worth of track, gpxlogger saves several megabyte long GPX file that includes several thousands of points. Afterwards, I try to plot this track on a map and use it with OpenLayers. It works, but several thousands of points make using the map a sloppy and slow experience.
I understand that having several thousands of points of suboptimal. There are myriads of points that can be deleted without losing almost anything: when there are several points making up roughly the straight line and we're moving with the same constant speed between them, we can just leave the first and the last point and throw away anything else.
I thought of using gpsbabel for such track simplification / optimization job, but, alas, it's simplification filter works only with routes, i.e. analyzing only geometrical shape of path, without timestamps (i.e. not checking that the speed was roughly constant).
Is there some ready-made utility / library / algorithm available to optimize tracks? Or may be I'm missing some clever option with gpsbabel?
Yes, as mentioned before, the Douglas-Peucker algorithm is a straightforward way to simplify 2D connected paths. But as you have pointed out, you will need to extend it to the 3D case to properly simplify a GPS track with an inherent time dimension associated with every point. I have done so for a web application of my own using a PHP implementation of Douglas-Peucker.
It's easy to extend the algorithm to the 3D case with a little understanding of how the algorithm works. Say you have input path consisting of 26 points labeled A to Z. The simplest version of this path has two points, A and Z, so we start there. Imagine a line segment between A and Z. Now scan through all remaining points B through Y to find the point furthest away from the line segment AZ. Say that point furthest away is J. Then, you scan the points between B and I to find the furthest point from line segment AJ and scan points K through Y to find the point furthest from segment JZ, and so on, until the remaining points all lie within some desired distance threshold.
This will require some simple vector operations. Logically, it's the same process in 3D as in 2D. If you find a Douglas-Peucker algorithm implemented in your language, it might have some 2D vector math implemented, and you'll need to extend those to use 3 dimensions.
You can find a 3D C++ implementation here: 3D Douglas-Peucker in C++
Your x and y coordinates will probably be in degrees of latitude/longitude, and the z (time) coordinate might be in seconds since the unix epoch. You can resolve this discrepancy by deciding on an appropriate spatial-temporal relationship; let's say you want to view one day of activity over a map area of 1 square mile. Imagining this relationship as a cube of 1 mile by 1 mile by 1 day, you must prescale the time variable. Conversion from degrees to surface distance is non-trivial, but for this case we simplify and say one degree is 60 miles; then one mile is .0167 degrees. One day is 86400 seconds; then to make the units equivalent, our prescale factor for your timestamp is .0167/86400, or about 1/5,000,000.
If, say, you want to view the GPS activity within the same 1 square mile map area over 2 days instead, time resolution becomes half as important, so scale it down twice further, to 1/10,000,000. Have fun.
Have a look at Ramer-Douglas-Peucker algorithm for smoothening complex polygons, also Douglas-Peucker line simplification algorithm can help you reduce your points.
OpenSource GeoKarambola java library (no Android dependencies but can be used in Android) that includes a GpxPathManipulator class that does both route & track simplification/reduction (3D/elevation aware).
If the points have timestamp information that will not be discarded.
https://sourceforge.net/projects/geokarambola/
This is the algorith in action, interactively
https://lh3.googleusercontent.com/-hvHFyZfcY58/Vsye7nVrmiI/AAAAAAAAHdg/2-NFVfofbd4ShZcvtyCDpi2vXoYkZVFlQ/w360-h640-no/movie360x640_05_82_05.gif
This algorithm is based on reducing the number of points by eliminating those that have the greatest XTD (cross track distance) error until a tolerated error is satisfied or the maximum number of points is reached (both parameters of the function), wichever comes first.
An alternative algorithm, for on-the-run stream like track simplification (I call it "streamplification") is:
you keep a small buffer of the points the GPS sensor gives you, each time a GPS point is added to the buffer (elevation included) you calculate the max XTD (cross track distance) of all the points in the buffer to the line segment that unites the first point with the (newly added) last point of the buffer. If the point with the greatest XTD violates your max tolerated XTD error (25m has given me great results) then you cut the buffer at that point, register it as a selected point to be appended to the streamplified track, trim the trailing part of the buffer up to that cut point, and keep going. At the end of the track the last point of the buffer is also added/flushed to the solution.
This algorithm is lightweight enough that it runs on an AndroidWear smartwatch and gives optimal output regardless of if you move slow or fast, or stand idle at the same place for a long time. The ONLY thing that maters is the SHAPE of your track. You can go for many minutes/kilometers and, as long as you are moving in a straight line (a corridor within +/- tolerated XTD error deviations) the streamplify algorithm will only output 2 points: those of the exit form last curve and entry on next curve.
I ran in to a similar issue. The rate at which the gps unit takes points is much larger that needed. Many of the points are not geographically far away from each other. The approach that I took is to calculate the distance between the points using the haversine formula. If the distance was not larger than my threshold (0.1 miles in my case) I threw away the point. This quickly gets the number of points down to a manageable size.
I don't know what language you are looking for. Here is a C# project that I was working on. At the bottom you will find the haversine code.
http://blog.bobcravens.com/2010/09/gps-using-the-netduino/
Hope this gets you going.
Bob
This is probably NP-hard. Suppose you have points A, B, C, D, E.
Let's try a simple deterministic algorithm. Suppose you calculate the distance from point B to line A-C and it's smaller than your threshold (1 meter). So you delete B. Then you try the same for C to line A-D, but it's bigger and D for C-E, which is also bigger.
But it turns out that the optimal solution is A, B, E, because point C and D are close to the line B-E, yet on opposite sides.
If you delete 1 point, you cannot be sure that it should be a point that you should keep, unless you try every single possible solution (which can be n^n in size, so on n=80 that's more than the minimum number of atoms in the known universe).
Next step: try a brute force or branch and bound algorithm. Doesn't scale, doesn't work for real-world size. You can safely skip this step :)
Next step: First do a determinstic algorithm and improve upon that with a metaheuristic algorithm (tabu search, simulated annealing, genetic algorithms). In java there are a couple of open source implementations, such as Drools Planner.
All in all, you 'll probably have a workable solution (although not optimal) with the first simple deterministic algorithm, because you only have 1 constraint.
A far cousin of this problem is probably the Traveling Salesman Problem variant in which the salesman cannot visit all cities but has to select a few.
You want to throw away uninteresting points. So you need a function that computes how interesting a point is, then you can compute how interesting all the points are and throw away the N least interesting points, where you choose N to slim the data set sufficiently. It sounds like your definition of interesting corresponds to high acceleration (deviation from straight-line motion), which is easy to compute.
Try this, it's free and opensource online Service:
https://opengeo.tech/maps/gpx-simplify-optimizer/
I guess you need to keep points where you change direction. If you split your track into the set of intervals of constant direction, you can leave only boundary points of these intervals.
And, as Raedwald pointed out, you'll want to leave points where your acceleration is not zero.
Not sure how well this will work, but how about taking your list of points, working out the distance between them and therefore the total distance of the route and then deciding on a resolution distance and then just linear interpolating the position based on each step of x meters. ie for each fix you have a "distance from start" measure and you just interpolate where n*x is for your entire route. (you could decide how many points you want and divide the total distance by this to get your resolution distance). On top of this you could add a windowing function taking maybe the current point +/- z points and applying a weighting like exp(-k* dist^2/accuracy^2) to get the weighted average of a set of points where dist is the distance from the raw interpolated point and accuracy is the supposed accuracy of the gps position.
One really simple method is to repeatedly remove the point that creates the largest angle (in the range of 0° to 180° where 180° means it's on a straight line between its neighbors) between its neighbors until you have few enough points. That will start off removing all points that are perfectly in line with their neighbors and will go from there.
You can do that in Ο(n log(n)) by making a list of each index and its angle, sorting that list in descending order of angle, keeping how many you need from the front of the list, sorting that shorter list in descending order of index, and removing the indexes from the list of points.
def simplify_points(points, how_many_points_to_remove)
angle_map = Array.new
(2..points.length - 1).each { |next_index|
removal_list.add([next_index - 1, angle_between(points[next_index - 2], points[next_index - 1], points[next_index])])
}
removal_list = removal_list.sort_by { |index, angle| angle }.reverse
removal_list = removal_list.first(how_many_points_to_remove)
removal_list = removal_list.sort_by { |index, angle| index }.reverse
removal_list.each { |index| points.delete_at(index) }
return points
end