In linear or logistic regression if we find a hypothesis function which fits the training set perfectly then it should be a good thing because in that case we have used 100 % of the information given to predict new information.
While it is called to be overfitting and said to be bad thing.
By making the hypothesis function simpler we may be actually increasing the noise instead of decreasing it.
Why is it so?
Overfitting occurs when you try "too hard" to make the examples in the training set fit the classification rule.
It is considered bad thing for 2 reasons main reasons:
The data might have noise. Trying too hard to classify 100% of the examples correctly, will make the noise count, and give you a bad rule while ignoring this noise - would usually be much better.
Remember that the classified training set is just a sample of the real data. This solution is usually more complex than what you would have got if you tolerated a few wrongly classified samples. According to Occam's Razor, you should prefer the simpler solution, so ignoring some of the samples, will be better,
Example:
According to Occam's razor, you should tolerate the misclassified sample, and assume it is noise or insignificant, and adopt the simple solution (green line) in this data set:
Because you actually didn't "learn" anything from your training set, you've just fitted to your data.
Imagine, you have a one-dimensional regression
x_1 -> y_1
...
x_n -> y_1
The function, defined this way
y_n, if x = x_n
f(x)=
0, otherwise
will give you perfect fit, but it's actually useless.
Hope, this helped a bit:)
Assuming that your regression accounts for all source of deviation in your data, then you might argue that your regression perfectly fits the data. However, if you know all (and I mean all) of the influences in your system, then you probably don't need a regression. You likely have an analytic solution that perfectly predicts new information.
In actuality, the information you possess will fall short of this perfect level. Noise (measurement error, partial observability, etc) will cause deviation in your data. In response, a regression (or other fitting mechanism) should seek the general trend of the data while minimizing the influence of noise.
Actually, the statement is not quite correct as written. It is perfectly fine to match 100% of your data if your hypothesis function is linear. Every continuous nonlinear function may be approximated locally by a linear function which gives important information on it's local behavior.
It is also fine to match 100 points of data to a quadratic curve if that data matches 100%. You can have high confidence that you are not overfitting your data, since the data consistently shows quadratic behavior.
However, one can always get 100% fit by using a polynomial function of high enough degree. Even without the noise that others have pointed out, though, you shouldn't assume your data has some high degree polynomial behavior without having some kind of theoretical or experimental confirmation of that hypothesis. Two good indicators that polynomial behavior is indicated are:
You have some theoretical reason for expecting the data to grow as x^n in one of the directional limits.
You have data that has been supporting a fixed degree polynomial fit as more and more data has been collected.
Notice, though, that even though exponential and reciprocal relationships may have data that fits a polynomial of high enough degree, they don't tend to obey eith of the two conditions above.
The point is that your data fit needs to be useful to prediction. You always know that a linear fit will give information locally, but that information becomes more useful the more points are fit. Even if there are only two points and noise, a linear fit still gives the best theoretical look at the data collected so far, and establishes the first expectations of the data. Beyond that, though, using a quadratic fit for three points or a cubic fit for four is not validly giving more information, as it assumes both local and asymptotic behavior information with the addition of one point. You need justification for your hypothesis function. That justification can come from more points or from theory.
(A third reason that sometimes comes up is
You have theoretical and experimental reason to believe that error and noise do not contribute more than some bounds, and you can take a polynomial hypothesis to look at local derivatives and the behavior needed to match the data.
This is typically used in understanding data to build theoretical models without having a good starting point for theory. You should still strive to use the smallest polynomial degree possible, and look to substitute out patterns in the coefficients with what they may indicate (reciprocal, exponential, gaussian, etc.) in infinite series.)
Try imagining it this way. You have a function from which you pick n different values to represent a sample / training set:
y(n) = x(n), n is element of [0, 1]
But, since you want to build a robust model, you want to add a little noise to your training set, so you actually add a little noise when generating the data:
data(n) = y(n) + noise(n) = x(n) + u(n)
where by u(n) I marked a uniform random noise with a mean 0 and standard deviation 1: U(0,1). Quite simply, it's a noise signal which is most probable to take an value 0, and less likely to take a value farther it is from 0.
And then you draw, let's say, 10 points to be your training set. If there was no noise, they would all be lying on a line y = x. Since there was noise, the lowest degree of polynomial function that can represent them is probably of 10-th order, a function like: y = a_10 * x^10 + a_9 * x^9 + ... + a_1 * x + a_0.
If you consider, by just using an estimation of the information from the training set, you would probably get a simpler function than the 10-th order polynomial function, and it would have been closer to the real function.
Consider further that your real function can have values outside the [0, 1] interval but for some reason the samples for the training set could only be collected from this interval. Now, a simple estimation would probably act significantly better outside the interval of the training set, while if we were to fit the training set perfectly, we would get an overfitted function that meandered with lots of ups and downs all over :)
Overfitting is termed as bad due to the bais it has to the true solution. The solution which is overfit is 100% fitting to the training data which is used but with any small data point addition the model will change drastically. This is called variance of the model. Hence the bais-variance tradeoff where we try to have a balance between both the factors so that, the model does not change drastically on small data changes but also reasonably properly predicts the output.
Related
I have recently started working on a project. One of the problems I ran into was converting changing accelerations into velocity. Accelerations at different points in time are provided through sensors. If you get the equation of these data points, the derivative of a certain time (x) on that equation will be the velocity.
I know how to do this on the computer, but how would I get the equation to start with? I have searched around but I have not found any existing programs that can form an equation given a set of points. In the past, I have created a neural net algorithm to form an equation, but it takes an incredibly long time to run.
If someone can link me a program or explain the process of doing this, that would be fantastic.
Sorry if this is in the wrong forum. I would post into math, but a programming background will be needed to know the realm of possibility of what a computer can do quickly.
This started out as a comment but ended up being too big.
Just to make sure you're familiar with the terminology...
Differentiation takes a function f(t) and spits out a new function f'(t) that tells you how f(t) changes with time (i.e. f'(t) gives the slope of f(t) at time t). This takes you from displacement to velocity or from velocity to acceleration.
Integreation takes a function f(t) and spits out a new function F(t) which measures the area under the function f(t) from the beginning of time up until a given point t. What's not obvious at first is that integration is actually the reverse of differentiation, a fact called the The Fundamental Theorem of Calculus. So integration takes you from acceleration to velocity or velocity to displacement.
You don't need to understand the rules of calculus to do numerical integration. The simplest (and most naive) method for integrating a function numerically is just by approximating the area by dividing it up into small slices between time points and summing the area of rectangles. This approximating sum is called a Reimann sum.
As you can see, this tends to really overshoot and undershoot certain parts of the function. A more accurate but still very simple method is the trapezoid rule, which also approximates the function with a series of slices, except the tops of the slices are straight lines between the function values rather than constant values.
Still more complicated, but yet a better approximation, is Simpson's rules, which approximates the function with parabolas between time points.
(source: tutorvista.com)
You can think of each of these methods as getting a better approximation of the integral because they each use more information about the function. The first method uses just one data point per area (a constant flat line), the second method uses two data points per area (a straight line), and the third method uses three data points per area (a parabola).
You could read up on the math behind these methods here or in the first page of this pdf.
I agree with the comments that numerical integration is probably what you want. In case you still want a function going through your data, let me further argue against doing that.
It's usually a bad idea to find a curve that goes exactly through some given points. In almost any applied math context you have to accept that there is a little noise in the inputs, and a curve going exactly through the points may be very sensitive to noise. This can produce garbage outputs. Finding a curve going exactly through a set of points is asking for overfitting to get a function that memorizes rather than understands the data, and does not generalize.
For example, take the points (0,0), (1,1), (2,4), (3,9), (4,16), (5,25), (6,36). These are seven points on y=x^2, which is fine. The value of x^2 at x=-1 is 1. Now what happens if you replace (3,9) with (2.9,9.1)? There is a sixth order polynomial passing through all 7 points,
4.66329x - 8.87063x^2 + 7.2281x^3 - 2.35108x^4 + 0.349747x^5 - 0.0194304x^6.
The value of this at x=-1 is -23.4823, very far from 1. While the curve looks ok between 0 and 2, in other examples you can see large oscillations between the data points.
Once you accept that you want an approximation, not a curve going exactly through the points, you have what is known as a regression problem. There are many types of regression. Typically, you choose a set of functions and a way to measure how well a function approximates the data. If you use a simple set of functions like lines (linear regression), you just find the best fit. If you use a more complicated family of functions, you should use regularization to penalize overly complicated functions such as high degree polynomials with large coefficients that memorize the data. If you either use a simple family or regularization, the function tends not to change much when you add or withhold a few data points, which indicates that it is a meaningful trend in the data.
Unfortunately, integrating accelerometer data to get velocity is a numerically unstable problem. For most applications, your error will diverge far too soon to get results of any practical value.
Recall that:
So:
However well you fit a function to your accelerometer data, you will still essentially be doing a piecewise interpolation of the underlying acceleration function:
Where the error terms from each integration will add!
Typically you will see wildly inaccurate results after just a few seconds.
I'm observing a sinusoidally-varying source, i.e. f(x) = a sin (bx + d) + c, and want to determine the amplitude a, offset c and period/frequency b - the shift d is unimportant. Measurements are sparse, with each source measured typically between 6 and 12 times, and observations are at (effectively) random times, with intervals between observations roughly between a quarter and ten times the period (just to stress, the spacing of observations is not constant for each source). In each source the offset c is typically quite large compared to the measurement error, while amplitudes vary - at one extreme they are only on the order of the measurement error, while at the other extreme they are about twenty times the error. Hopefully that fully outlines the problem, if not, please ask and i'll clarify.
Thinking naively about the problem, the average of the measurements will be a good estimate of the offset c, while half the range between the minimum and maximum value of the measured f(x) will be a reasonable estimate of the amplitude, especially as the number of measurements increase so that the prospects of having observed the maximum offset from the mean improve. However, if the amplitude is small then it seems to me that there is little chance of accurately determining b, while the prospects should be better for large-amplitude sources even if they are only observed the minimum number of times.
Anyway, I wrote some code to do a least-squares fit to the data for the range of periods, and it identifies best-fit values of a, b and d quite effectively for the larger-amplitude sources. However, I see it finding a number of possible periods, and while one is the 'best' (in as much as it gives the minimum error-weighted residual) in the majority of cases the difference in the residuals for different candidate periods is not large. So what I would like to do now is quantify the possibility that the derived period is a 'false positive' (or, to put it slightly differently, what confidence I can have that the derived period is correct).
Does anybody have any suggestions on how best to proceed? One thought I had was to use a Monte-Carlo algorithm to construct a large number of sources with known values for a, b and c, construct samples that correspond to my measurement times, fit the resultant sample with my fitting code, and see what percentage of the time I recover the correct period. But that seems quite heavyweight, and i'm not sure that it's particularly useful other than giving a general feel for the false-positive rate.
And any advice for frameworks that might help? I have a feeling this is something that can likely be done in a line or two in Mathematica, but (a) I don't know it, an (b) don't have access to it. I'm fluent in Java, competent in IDL and can probably figure out other things...
This looks tailor-made for working in the frequency domain. Apply a Fourier transform and identify the frequency based on where the power is located, which should be clear for a sinusoidal source.
ADDENDUM To get an idea of how accurate is your estimate, I'd try a resampling approach such as cross-validation. I think this is the direction that you're heading with the Monte Carlo idea; lots of work is out there, so hopefully that's a wheel you won't need to re-invent.
The trick here is to do what might seem at first to make the problem more difficult. Rewrite f in the similar form:
f(x) = a1*sin(b*x) + a2*cos(b*x) + c
This is based on the identity for the sin(u+v).
Recognize that if b is known, then the problem of estimating {a1, a2, c} is a simple LINEAR regression problem. So all you need to do is use a 1-variable minimization tool, working on the value of b, to minimize the sum of squares of the residuals from that linear regression model. There are many such univariate optimizers to be found.
Once you have those parameters, it is easy to find the parameter a in your original model, since that is all you care about.
a = sqrt(a1^2 + a2^2)
The scheme I have described is called a partitioned least squares.
If you have a reasonable estimate of the size and the nature of your noise (e.g. white Gaussian with SD sigma), you can
(a) invert the Hessian matrix to get an estimate of the error in your position and
(b) should be able to easily derive a significance statistic for your fit residues.
For (a), compare http://www.physics.utah.edu/~detar/phys6720/handouts/curve_fit/curve_fit/node6.html
For (b), assume that your measurement errors are independent and thus the variance of their sum is the sum of their variances.
I've experimented with the two ways of implementing a least-squares fit (LSF) algorithm shown here.
The first code is simply the textbook approach, as described by Wolfram's page on LSF. The second code re-arranges the equation to minimize machine errors. Both codes produce similar results for my data. I compared these results with Matlab's p=polyfit(x,y,1) function, using correlation coefficients to measure the "goodness" of fit and compare each of the 3 routines. I observed that while all 3 methods produced good results, at least for my data, Matlab's routine had the best fit (the other 2 routines had similar results to each other).
Matlab's p=polyfit(x,y,1) function uses a Vandermonde matrix, V (n x 2 matrix) and QR factorization to solve the least-squares problem. In Matlab code, it looks like:
V = [x1,1; x2,1; x3,1; ... xn,1] % this line is pseudo-code
[Q,R] = qr(V,0);
p = R\(Q'*y); % performs same as p = V\y
I'm not a mathematician, so I don't understand why it would be more accurate. Although the difference is slight, in my case I need to obtain the slope from the LSF and multiply it by a large number, so any improvement in accuracy shows up in my results.
For reasons I can't get into, I cannot use Matlab's routine in my work. So, I'm wondering if anyone has a more accurate equation-based approach recommendation I could use that is an improvement over the above two approaches, in terms of rounding errors/machine accuracy/etc.
Any comments appreciated! thanks in advance.
For a polynomial fitting, you can create a Vandermonde matrix and solve the linear system, as you already done.
Another solution is using methods like Gauss-Newton to fit the data (since the system is linear, one iteration should do fine). There are differences between the methods. One possibly reason is the Runge's phenomenon.
I am looking for a general algorithm to help in situations with similar constraints as this example :
I am thinking of a system where images are constructed based on a set of operations. Each operation has a set of parameters. The total "gene" of the image is then the sequential application of the operations with the corresponding parameters. The finished image is then given a vote by one or more real humans according to how "beautiful" it is.
The question is what kind of algorithm would be able to do better than simply random search if you want to find the most beautiful image? (and hopefully improve the confidence over time as votes tick in and improve the fitness function)
Given that the operations will probably be correlated, it should be possible to do better than random search. So for example operation A with parameters a1 and a2 followed by B with parameters b1 could generally be vastly superior to B followed by A. The order of operations will matter.
I have tried googling for research papers on random walk and markov chains as that is my best guesses about where to look, but so far have found no scenarios similar enough. I would really appreciate even just a hint of where to look for such an algorithm.
I think what you are looking for fall in a broad research area called metaheuristics (which include many non-linear optimization algorithms such as genetic algorithms, simulated annealing or tabu search).
Then if your raw fitness function is just giving a statistical value somehow approximating a real (but unknown) fitness function, you can probably still use most metaheuristics by (somehow) smoothing your fitness function (averaging results would do that).
Do you mean the Metropolis algorithm?
This approach uses a random walk, weighted by the fitness function. It is useful for locating local extrema in complicated fitness landscapes, but is generally slower than deterministic approaches where those will work.
You're pretty much describing a genetic algorithm in which the sequence of operations represents the "gene" ("chromosome" would be a better term for this, where the parameter[s] passed to each operation represents a single "gene", and multiple genes make up a chromosome), the image produced represents the phenotypic expression of the gene, and the votes from the real humans represent the fitness function.
If I understand your question, you're looking for an alternative algorithm of some sort that will evaluate the operations and produce a "beauty" score similar to what the real humans produce. Good luck with that - I don't think there really is any such thing, and I'm not surprised that you didn't find anything. Human brains, and correspondingly human evaluations of aesthetics, are much too staggeringly complex to be reducible to a simplistic algorithm.
Interestingly, your question seems to encapsulate the bias against using real human responses as the fitness function in genetic-algorithm-based software. This is a subject of relevance to me, since my namesake software is specifically designed to use human responses (or "votes") to evaluate music produced via a genetic process.
Simple Markov Chain
Markov chains, which you mention, aren't a bad way to go. A Markov chain is just a state machine, represented as a graph with edge weights which are transition probabilities. In your case, each of your operations is a node in the graph, and the edges between the nodes represent allowable sequences of operations. Since order matters, your edges are directed. You then need three components:
A generator function to construct the graph of allowed transitions (which operations are allowed to follow one another). If any operation is allowed to follow any other, then this is easy to write: all nodes are connected, and your graph is said to be complete. You can initially set all the edge weights to 1.
A function to traverse the graph, crossing N nodes, where N is your 'gene-length'. At each node, your choice is made randomly, but proportionally weighted by the values of the edges (so better edges have a higher chance of being selected).
A weighting update function which can be used to adjust the weightings of the edges when you get feedback about an image. For example, a simple update function might be to give each edge involved in a 'pleasing' image a positive vote each time that image is nominated by a human. The weighting of each edge is then normalised, with the currently highest voted edge set to 1, and all the others correspondingly reduced.
This graph is then a simple learning network which will be refined by subsequent voting. Over time as votes accumulate, successive traversals will tend to favour the more highly rated sequences of operations, but will still occasionally explore other possibilities.
Advantages
The main advantage of this approach is that it's easy to understand and code, and makes very few assumptions about the problem space. This is good news if you don't know much about the search space (e.g. which sequences of operations are likely to be favourable).
It's also easy to analyse and debug - you can inspect the weightings at any time and very easily calculate things like the top 10 best sequences known so far, etc. This is a big advantage - other approaches are typically much harder to investigate ("why did it do that?") because of their increased abstraction. Although very efficient, you can easily melt your brain trying to follow and debug the convergence steps of a simplex crawler!
Even if you implement a more sophisticated production algorithm, having a simple baseline algorithm is crucial for sanity checking and efficiency comparisons. It's also easy to tinker with, by messing with the update function. For example, an even more baseline approach is pure random walk, which is just a null weighting function (no weighting updates) - whatever algorithm you produce should perform significantly better than this if its existence is to be justified.
This idea of baselining is very important if you want to evaluate the quality of your algorithm's output empirically. In climate modelling, for example, a simple test is "does my fancy simulation do any better at predicting the weather than one where I simply predict today's weather will be the same as yesterday's?" Since weather is often correlated on a timescale of several days, this baseline can give surprisingly good predictions!
Limitations
One disadvantage of the approach is that it is slow to converge. A more agressive choice of update function will push promising results faster (for example, weighting new results according to a power law, rather than the simple linear normalisation), at the cost of giving alternatives less credence.
This is equivalent to fiddling with the mutation rate and gene pool size in a genetic algorithm, or the cooling rate of a simulated annealing approach. The tradeoff between 'climbing hills or exploring the landscape' is an inescapable "twiddly knob" (free parameter) which all search algorithms must deal with, either directly or indirectly. You are trying to find the highest point in some fitness search space. Your algorithm is trying to do that in less tries than random inspection, by looking at the shape of the space and trying to infer something about it. If you think you're going up a hill, you can take a guess and jump further. But if it turns out to be a small hill in a bumpy landscape, then you've just missed the peak entirely.
Also note that since your fitness function is based on human responses, you are limited to a relatively small number of iterations regardless of your choice of algorithmic approach. For example, you would see the same issue with a genetic algorithm approach (fitness function limits the number of individuals and generations) or a neural network (limited training set).
A final potential limitation is that if your "gene-lengths" are long, there are many nodes, and many transitions are allowed, then the size of the graph will become prohibitive, and the algorithm impractical.
I'm reading data from a device which measures distance. My sample rate is high so that I can measure large changes in distance (i.e. velocity) but this means that, when the velocity is low, the device delivers a number of measurements which are identical (due to the granularity of the device). This results in a 'stepped' curve.
What I need to do is to smooth the curve in order to calculate the velocity. Following that I then need to calculate the acceleration.
How to best go about this?
(Sample rate up to 1000Hz, calculation rate of 10Hz would be ok. Using C# in VS2005)
The wikipedia entry from moogs is a good starting point for smoothing the data. But it does not help you in making a decision.
It all depends on your data, and the needed processing speed.
Moving Average
Will flatten the top values. If you are interrested in the minimum and maximum value, don't use this. Also I think using the moving average will influence your measurement of the acceleration, since it will flatten your data (a bit), thereby acceleration will appear to be smaller. It all comes down to the needed accuracy.
Savitzky–Golay
Fast algorithm. As fast as the moving average. That will preserve the heights of peaks. Somewhat harder to implement. And you need the correct coefficients. I would pick this one.
Kalman filters
If you know the distribution, this can give you good results (it is used in GPS navigation systems). Maybe somewhat harder to implement. I mention this because I have used them in the past. But they are probably not a good choice for a starter in this kind of stuff.
The above will reduce noise on your signal.
Next you have to do is detect the start and end point of the "acceleration". You could do this by creating a Derivative of the original signal. The point(s) where the derivative crosses the Y-axis (zero) are probably the peaks in your signal, and might indicate the start and end of the acceleration.
You can then create a second degree derivative to get the minium and maximum acceleration itself.
You need a smoothing filter, the simplest would be a "moving average": just calculate the average of the last n points.
The question here is, how to determine n, can you tell us more about your application?
(There are other, more complicated filters. They vary on how they preserve the input data. A good list is in Wikipedia)
Edit!: For 10Hz, average the last 100 values.
Moving averages are generally terrible - but work well for white noise. Both moving averages & Savitzky-Golay both boil down to a correlation - and therefore are very fast and could be implemented in real time. If you need higher order information like first and second derivatives - SG is a good right choice. The magic of SG lies in the constant correlation coefficients needed for the filter - once you have decided the length and degree of polynomial to fit locally, the coefficients need only to be found once. You can compute them using R (sgolay) or Matlab.
You can also estimate a noisy signal's first derivative via the Savitzky-Golay best-fit polynomials - these are sometimes called Savitzky-Golay derivatives - and typically give a good estimate of the first derivative.
Kalman filtering can be very effective, but it's heavier computationally - it's hard to beat a short convolution for speed!
Paul
CenterSpace Software
In addition to the above articles, have a look at Catmull-Rom Splines.
You could use a moving average to smooth out the data.
In addition to GvSs excellent answer above you could also consider smoothing / reducing the stepping effect of your averaged results using some general curve fitting such as cubic or quadratic splines.