I have a problem with recognising a signal. Let say the signal is a quasiperiodic signal, the period time has finite limits. The "shape" of the signal must match some criteria, so the actual algorithm using signal processing techniques such as filtering, derivating the the signal, looking for maximum and minimum values. It has a good rate at finding the good signals, but the problem is it also detects wrong shapes too.
So I want to use Aritifical Intelligence - mainly Neural Networks - to overcome this problem. I thought that a multi layer network with some average inputs (the signal can be reduced) and one output whould shows the "matching" from 0..1. However the problem is that I never did such a thing, so I am asking for help, how to achive something like this? How to teach the neural network to get the expected results? (let say I have vectors for inputs which should give 1 as output)
Or this whole idea is a wrong approximation for the problem? I am open to any learning algorithms or idea to learn and use to overcome on this problem.
So here is a figure on the measured signal(s) (values and time is not a concern now) and you can see a lot "wrong" ones, the most detected signals are good as mentioned above.
Your question can be answered in a broad manner. You should consider editing it to prevent it to be closed.
But anyway, Matlab had a lot of built-in function and toolbox to support Artificial Intelligence, with a lot of sample code available, which you can modify and refer to. You can find some in Matlab FileExchange.
And I know reading a lot of technical paper for Artificial Intelligence is a daunting task, so good luck!
You can try to build a neural network using Neuroph. You can inspire from "http://neuroph.sourceforge.net/TimeSeriesPredictionTutorial.html".
On the other hand, it is possible to approximate the signal using Fourier transformation.
You can try 1D convolution. So the basic idea is you give a label 0: bad, 1: good to each signal value at each timestamp. After this, you can model
model = Sequential()
model.add(Conv1D(filters=64, kernel_size=3, activation='relu', padding = 'same', input_shape=(1,1)))
model.add(Bidirectional(LSTM(20, return_sequences=True)))
model.add(Conv1D(filters=64, kernel_size=3, activation='relu', padding = 'same'))
model.add(Dropout(0.5))
model.add(Flatten())
model.add(Dense(100, activation='sigmoid'))
model.add(Dense(2, activation='softmax'))
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
Train the model and then give it a new signal to predict. It will predict given series to 0 and 1 values. if count of 0 is more than count of 1, the signal is not good.
Related
The N2 diagram for my full problem is below.
The N2 diagram for the coupled portion of the problem is below.
I have a DirectSolver handling the coupling between LLTForces and ImplicitLiftingLine, and an LNBGS solver handling the coupling between LiftingLineGroup and TestCL.
The gist for the problem is here: https://gist.github.com/eufren/31c0e569ed703b2aea3e2ef5360610f7
I have implemented guess_nonlinear() on ImplicitLiftingLine, which should use various outputs from LLTGeometry to give a good initial guess for the vortex strengths based on a linearised form of the governing equations.
def guess_nonlinear(self, inputs, outputs, resids):
freestream_unit_vector = inputs['freestream_unit_vector']
freestream_velocity = inputs['freestream_velocity']
n = inputs['normal_vectors']
A = inputs['surface_areas']
l = inputs['bound_vortices']
ic_tot = inputs['influence_coefficients_total']
v_inf = freestream_velocity
v_inf_vec = v_inf*freestream_unit_vector
lin_numerator = np.pi * v_inf * A * np.sum(n * v_inf_vec, axis=1)
lin_denominator = (np.linalg.norm(np.cross(v_inf_vec, l), axis=1) - np.pi * v_inf * A * np.sum(np.sum(n * ic_tot, axis=2), axis=1))
lin_vtx_str = lin_numerator / lin_denominator
outputs['vortex_strengths'] = lin_vtx_str
However, when the problem is run for the first time, any inputs not explicitly set with p.set_val() are all 1s. This causes guess_nonlinear() to give a bad output and so the system fails to converge:
As far as I can tell, the execution order for the LLT group is correct, and the geometry components should be being executed before the implicit component. I'm confused as to why this doesn't seem to actually be happening when the code is run, and instead these inputs are taking their default values.
What do I need to change to get this to work properly? Additionally, I've found difficulty in getting LNBGS to converge (hence adding guess_nonlinear()) during optimisation - only DirectSolver gets all the way through the optimisation without issues, but it's very slow for large numbers of LLT nodes). How can I improve the linear and nonlinear solver selection, and improve the reliability of the iterative solver?
Note: Thanks for providing a testable example. It made figuring out the answer to your question a lot simpler. Your problem was a bit subtle and I would not have been able to give a good answer without runnable code
Your first question: "Why are all the inputs 1"
"Short" Answer
You have put the nonlinear solver to high in the model hierarchy, which then included a key precurser component that computed your input values. By moving the solver down to a lower level of the model, I was able to ensure that the precurser component (LTTGeometry) ran and had valid outputs before you got to the guess_nonlinear of implicit component.
Here is what you had (Notice the implicit solver included LTTGeometry even though the data cycle does not require that component:
I moved both the nonlinear solver and the linear solver down into the LTTCycle group, which then allows the LTTGeometry component to execute before getting to the nonlinear solver and guess_nonlinear step:
My fix is only partially correct, since there is a secondary cycle from the TestCL component that also needs a solver and does not have one. However, that cycle still does not involve the LTTGeometry group. So the fully correct fix is to restructure you model top run geometry first, and then put the LTTCycle and TestCL groups together so you can run a solver over just them. That was a bit more hacking than I wanted to do on your test problem, but you can see the general idea from the adjusted N2 above.
Long Answer
The guess_nonlinear sequence in OpenMDAO does NOT run the compute method of explicit components or of groups. It follows the execution hierarchy, and calls any guess_nonlinear that it finds. So that means that any explicit components you have in your model will NOT get executed, their outputs will not get updated with computed values, and those computed values will not get passed to the inputs of downstream components.
Things get a little tricky when you have deep model hierarchies. The guess_nonlinear method is called as the first step in the nonlinear solver process. If you have a NonLinearRunOnce solver at the top level, it will follow the compute chain down the line calling compute or solve_nonlinear on each child and doing a data transfer after each one. If one of those children happens to be a group with a nonlinear solver, then that solver will call guess_nonlinear on its children (grandchildren of the top group with the NonLinearRunOnce solver) as the first step. So any outputs that were computed by the siblings of this group will be valid, but none of the outputs from the grandchild level will have been computed yet.
You may be wondering why not just have the guess_nonlinear method call the compute for any explicit components? There is a difficult to balance trade off here. If you assume that all explicit components are very cheap to run, then it might make sense to run the compute methods --- or it might not. A lot depends on the cyclic data structure. If some early component in the group needs guesses from the later one, then running its compute isn't going to help you much at all. Perhaps more importantly though, not all explicit components are cheap to run. You might have a very expensive computation, and calling compute as part of the guess process would be way too costly.
The compromise here, if you need some kind of top level guess process, is that you can implement guess_nonlinear at the group level. It's less common to do, but it gives you total control over what happens. You can call whatever you need to call in whatever sequence.
So the absolute key thing to remember is that the only data you have available to you when a guess_nonlinear is called is any data that was computed before your containing solver was executed. That means any thing that was computed before you got to the model scope of the containing solver (not the scope of the component with the guess_method itself).
Your second question: "How can I speed this up when the number of nodes gets large?"
This one not possible to give a generic answer to at all. I noticed that you have already specified sparse partial derivatives. That is a great start, but if its still not fast enough for you then it means you're reaching the limits of what you can do with a DirectSolver. You note that this solver is the only one that gets you through the optimization without issues, which I will take to mean that ScipyKryloventer link description here and PetscKrylov are not converging the linear system well for you --- at least not by themselves. Thats not surprising, as krylov solvers almost always require some kind of preconditioner... and this is why I can't offer a generic answer. Setting up efficient linear solvers for larger-scale compute is a tricky subject. If you look into the literature, you'll find some good suggestions. You can also study open source implementations like VSPAero for some tips.
effectively, you've reached the limit of what simple linear solvers can offer you. From this point forward, OpenMDAO can help a bit by making it easier to implement some preconditioning, but you'll have to suffer the math side yourself.
I have to use a uninformed search technique to solve the following problem.
The game is like:
On side of the river, there is a Policeman, a Robber, a woman in a red-dress and her two children, a woman in a yellow dress and her two children. There is a boat that can carry atmost two persons. The children cannot drive the boat.
If the policeman is absent then the robber will kill the people. If the red-dress woman is absent then the yellow-dressed woman will kill the red-dressed woman’s children and vice versa.
I am confused as usual. Please help me figure it out.
The problem and how can it be solved (without programming) is shown in the video below:
https://www.youtube.com/watch?v=vSusAZBSWwg
Thank you.
Problems like River Crossing Puzzle, Sokoban or Lemmings are solved normally with Brute-Force-Search in the gametree. The domain is specified declarative as rules (moves are possible or not), and a function which determines the amount of points which are reached by a policy (policy = plan through the gametree). The solver has the aim to find a good policy. The best hardware for doing this is a quantum computer with unlimited speed for testing as much as possble moves per second.
The reason why this is not practicaly is because of a phenomenon which is called "combinatorial explosion", first introduced by James Lighthill in year 1973 for prooving that artifical intelligence is not ready for use in realworld. The answer to that problem is, to use alternative strategies which have noting to do with brute-force-search.
One possibility is to use heuristics which are hardcoded into programcode. Often these heuristics are called macroaction or motion primitives. An example would be "bring-robber-to-other-side". This subfunction executes a predefined number of actions. Another macroaction could be "check-if-two-woman-are-on-the-same-side". To implement such kind of strategy for the complete game is a hard task. Not because of high cpu usage, but because of every detail has to coded into software.
I've read some material about quantum computers and quantum circuits. A certain number of already known algorithms (Simon's algorithm, period-finding algorithm, Grover's algorithm, …) have the following form:
Suppose I have an unknown classical function f: {0,1}^n -> {0, 1}^m satisfying a certain number of statements. I can associate the (unknown) quantum circuit U_f to it and plug the |0.. 0> input state. Now let us define circuit X and show that when appended to U_f, the global output can be measured to extract some information about f.
Wait a minute... What is the relation with classical circuits? A classical problem would refer to an unknown input that satisfies certain properties, this input representing a state coming from outside (user action, file system, database, server, whatever). In case this state is rather generated by another circuit/algorithm the logic applies to the input before. In the end we don't reason about unknown circuits but about unknown inputs. The circuits (the algorithms/ the functions) are the known/chosen components.
Here I came to realize that the common name "circuit" was somehow misleading. In a classical world gate inputs can be thought as values coexisting with outputs. But quantum gates seem to require a temporal interpretation: inputs and outputs represent temporal evolution of the same qubits.
Now this does not really explain how you transform a given bunch of a priori unknown classical input bits (that I believe your keyboard will keep on generating in the future except maybe if Schrödinger's cat is sitting on it) into a "black box quantum circuit" transforming |0…0> into something to be reversed. For example Grover's algorithm propose, for a quantum circuit corresponding to a function f: {0, 1}^n -> {0, 1} that yields 1 for a single unknown input, an efficient method to determine this input. Nice! But how and why would you have to start with such a circuit in the first place?
In practice, the 'unknown function black box' is just a circuit that checks if its input meets some criteria or is the solution to a problem.
This is useful because it's easier to make a circuit that detects a solution than it is to actually find a solution. Grover's algorithm then augments our detector circuit into a solver circuit:
The classical equivalent of Grover's algorithm is a brute-force search function like this:
def bruteForceSearch(min, max, predicate):
for i in min..max:
if predicate(i):
return i
return none
which you would use it like so:
let mersennePrimeWithFiveDigitExponent = bruteForceSearch(
10000,
99999,
i => isMersennePrime(2**i - 1))
Notice how the brute force search turns our knowledge of how to recognize something into a mechanism for finding something. Grover's algorithm does the same thing, but quadratically faster and with the caveat that the recognizer must be implemented as a reversible circuit.
I need to implement Minesweeper solver. I have started to implement rule based agent.
I have implemented certain rules. I have a heuristic function for choosing best matching rule for current cell (with info about surrounding cells) being treated. So for each chosen cell it can decide for 8 surroundings cells to open them, to mark them or to do nothing. I mean. at the moment, the agent gets as an input some revealed cell and decides what to do with surrounding cells (at the moment, the agent do not know, how to decide which cell to treat).
My question is, what algorithm to implement for deciding which cell to treat?
Suppose, for, the first move, the agent will reveal a corner cell (or some other, according to some rule for the first move). What to do after that?
I understand that I need to implement some kind of search. I know many search algorithms (BFS, DFS, A-STAR and others), that is not the problem, I just do not understand how can I use here these searches.
I need to implement it in a principles of Artificial Intelligence: A modern approach.
BFS, DFS, and A* are probably not appropriate here. Those algorithms are good if you are trying to plan out a course of action when you have complete knowledge of the world. In Minesweeper, you don't have such knowledge.
Instead, I would suggest trying to use some of the logical inference techniques from Section III of the book, particularly using SAT or the techniques from Chapter 10. This will let you draw conclusions about where the mines are using facts like "one of the following eight squares is a mine, and exactly two of the following eight squares is a mine." Doing this at each step will help you identify where the mines are, or realize that you must guess before continuing.
Hope this helps!
I ported this (with a bit of help). Here is the link to it working: http://robertleeplummerjr.github.io/smartSweepers.js/ . Here is the project: https://github.com/robertleeplummerjr/smartSweepers.js
Have fun!
Not unlike a clap detector ("Clap on! clap clap Clap off! clap clap Clap on, clap off, the Clapper! clap clap ") I need to detect when a door closes. This is in a vehicle, which is easier than a room or household door:
Listen: http://ubasics.com/so/van_driver_door_closing.wav
Look:
It's sampling at 16bits 4khz, and I'd like to avoid lots of processing or storage of samples.
When you look at it in audacity or another waveform tool it's quite distinctive, and almost always clips due to the increase in sound pressure in the vehicle - even when the windows and other doors are open:
Listen: http://ubasics.com/so/van_driverdoorclosing_slidingdoorsopen_windowsopen_engineon.wav
Look:
I expect there's a relatively simple algorithm that would take readings at 4kHz, 8 bits, and keep track of the 'steady state'. When the algorithm detects a significant increase in the sound level it would mark the spot.
What are your thoughts?
How would you detect this event?
Are there code examples of sound pressure level calculations that might help?
Can I get away with less frequent sampling (1kHz or even slower?)
Update: Playing with Octave (open source numerical analysis - similar to Matlab) and seeing if the root mean square will give me what I need (which results in something very similar to the SPL)
Update2: Computing the RMS finds the door close easily in the simple case:
Now I just need to look at the difficult cases (radio on, heat/air on high, etc). The CFAR looks really interesting - I know I'm going to have to use an adaptive algorithm, and CFAR certainly fits the bill.
-Adam
Looking at the screenshots of the source audio files, one simple way to detect a change in sound level would be to do a numerical integration of the samples to find out the "energy" of the wave at a specific time.
A rough algorithm would be:
Divide the samples up into sections
Calculate the energy of each section
Take the ratio of the energies between the previous window and the current window
If the ratio exceeds some threshold, determine that there was a sudden loud noise.
Pseudocode
samples = load_audio_samples() // Array containing audio samples
WINDOW_SIZE = 1000 // Sample window of 1000 samples (example)
for (i = 0; i < samples.length; i += WINDOW_SIZE):
// Perform a numerical integration of the current window using simple
// addition of current sample to a sum.
for (j = 0; j < WINDOW_SIZE; j++):
energy += samples[i+j]
// Take ratio of energies of last window and current window, and see
// if there is a big difference in the energies. If so, there is a
// sudden loud noise.
if (energy / last_energy > THRESHOLD):
sudden_sound_detected()
last_energy = energy
energy = 0;
I should add a disclaimer that I haven't tried this.
This way should be possible to be performed without having the samples all recorded first. As long as there is buffer of some length (WINDOW_SIZE in the example), a numerical integration can be performed to calculate the energy of the section of sound. This does mean however, that there will be a delay in the processing, dependent on the length of the WINDOW_SIZE. Determining a good length for a section of sound is another concern.
How to Split into Sections
In the first audio file, it appears that the duration of the sound of the door closing is 0.25 seconds, so the window used for numerical integration should probably be at most half of that, or even more like a tenth, so the difference between the silence and sudden sound can be noticed, even if the window is overlapping between the silent section and the noise section.
For example, if the integration window was 0.5 seconds, and the first window was covering the 0.25 seconds of silence and 0.25 seconds of door closing, and the second window was covering 0.25 seconds of door closing and 0.25 seconds of silence, it may appear that the two sections of sound has the same level of noise, therefore, not triggering the sound detection. I imagine having a short window would alleviate this problem somewhat.
However, having a window that is too short will mean that the rise in the sound may not fully fit into one window, and it may apppear that there is little difference in energy between the adjacent sections, which can cause the sound to be missed.
I believe the WINDOW_SIZE and THRESHOLD are both going to have to be determined empirically for the sound which is going to be detected.
For the sake of determining how many samples that this algorithm will need to keep in memory, let's say, the WINDOW_SIZE is 1/10 of the sound of the door closing, which is about 0.025 second. At a sampling rate of 4 kHz, that is 100 samples. That seems to be not too much of a memory requirement. Using 16-bit samples that's 200 bytes.
Advantages / Disadvantages
The advantage of this method is that processing can be performed with simple integer arithmetic if the source audio is fed in as integers. The catch is, as mentioned already, that real-time processing will have a delay, depending on the size of the section that is integrated.
There are a couple of problems that I can think of to this approach:
If the background noise is too loud, the difference in energy between the background noise and the door closing will not be easily distinguished, and it may not be able to detect the door closing.
Any abrupt noise, such as a clap, could be regarded as the door is closing.
Perhaps, combining the suggestions in the other answers, such as trying to analyze the frequency signature of the door closing using Fourier analysis, which would require more processing but would make it less prone to error.
It's probably going to take some experimentation before finding a way to solve this problem.
You should tap in to the door close switches in the car.
Trying to do this with sound analysis is overengineering.
There are a lot of suggestions about different signal processing
approaches to take, but really, by the time you learn about detection
theory, build an embedded signal processing board, learn the processing
architecture for the chip you chose, attempt an algorithm, debug it, and then
tune it for the car you want to use it on (and then re-tune and re-debug
it for every other car), you will be wishing you just stickey taped a reed
switch inside the car and hotglued a magnet to the door.
Not that it's not an interesting problem to solve for the dsp experts,
but from the way you're asking this question, it's clear that sound
processing isn't the route you want to take. It will just be such a nightmare
to make it work right.
Also, the clapper is just an high pass filter fed into a threshold detector. (plus a timer to make sure 2 claps quickly enough together)
There is a lot of relevant literature on this problem in the radar world (it's called detection theory).
You might have a look at "cell averaging CFAR" (constant false alarm rate) detection. Wikipedia has a little bit here. Your idea is very similar to this, and it should work! :)
Good luck!
I would start by looking at the spectral. I did this on the two audio files you gave, and there does seem to be some similarity you could use. For example the main difference between the two seems to be around 40-50Hz. My .02.
UPDATE
I had another idea after posting this. If you can, add an accelerometer onto the device. Then correlate the vibrational and acoustic signals. This should help with cross vehicle door detection. I'm thinking it should be well correlated since the sound is vibrationally driven, wheres the stereo for example, is not. I've had a device that was able to detect my engine rpm with a windshield mount (suction cup), so the sensitivity might be there. (I make no promises this works!)
(source: charlesrcook.com)
%% Test Script (Matlab)
clear
hold all %keep plots open
dt=.001
%% Van driver door
data = wavread('van_driver_door_closing.wav');
%Frequency analysis
NFFT = 2^nextpow2(length(data));
Y = fft(data(:,2), NFFT)/length(data);
freq = (1/dt)/2*linspace(0,1,NFFT/2);
spectral = [freq' 2*abs(Y(1:NFFT/2))];
plot(spectral(:,1),spectral(:,2))
%% Repeat for van sliding door
data = wavread('van_driverdoorclosing.wav');
%Frequency analysis
NFFT = 2^nextpow2(length(data));
Y = fft(data(:,2), NFFT)/length(data);
freq = (1/dt)/2*linspace(0,1,NFFT/2);
spectral = [freq' 2*abs(Y(1:NFFT/2))];
plot(spectral(:,1),spectral(:,2))
The process for finding distinct spike in audio signals is called transient detection. Applications like Sony's Acid and Ableton Live use transient detection to find the beats in music for doing beat matching.
The distinct spike you see in the waveform above is called a transient, and there are several good algorithms for detecting it. The paper Transient detection and classification in energy matters describes 3 methods for doing this.
I would imagine that the frequency and amplitude would also vary significantly from vehicle to vehicle. Best way to determine that would be taking a sample in a Civic versus a big SUV. Perhaps you could have the user close the door in a "learning" mode to get the amplitude and frequency signature. Then you could use that to compare when in usage mode.
You could also consider using Fourier analysis to eliminate background noises that aren't associated with the door close.
Maybe you should try to detect significant instant rise in air pressure that should mark a door close. You can pair it with this waveform and sound level analysis and these all might give you a better result.
On the issue of less frequent sampling, the highest sound frequency which can be captured is half of the sampling rate. Thus, if the car door sound was strongest at 1000Hz (for example) then a sampling rate below 2000Hz would lose that sound entirely
A very simple noise gate would probably do just fine in your situation. Simply wait for the first sample whose amplitude is above a specified threshold value (to avoid triggering with background noise). You would only need to get more complicated than this if you need to distinguish between different types of noise (e.g. a door closing versus a hand clap).