I would like to understand SAM, the model during YOLOv4.
However, I am confused as to the meaning of the "x" in the diagram.
Please tell me what module name this "x" represents in yolov4.cfg in darknet.
Thank you in advance.
It is element-wise multiplication, so in most of the libraries quite literally the *. Source is here:
https://arxiv.org/pdf/1807.06521.pdf
Related
When coding PyTorch in torch.nn.utils I see two functions, clip_grad_norm and clip_grad_norm_.
I want to know the difference so I went to check the documentation but when I searched I only found the clip_grad_norm_ and not clip_grad_norm.
So I'm here to ask if anyone knows the difference.
Pytorch uses the trailing underscore convention for in-place operations. So the difference is that the one with an underscore modifies the tensor in place and the other one leaves the original tensor unmodified and returns a new tensor.
I have a problem with some homework, for my Advanced Digital Design course, in which I have to create the truth table and find the Boolean expression for a provided circuit (image is linked below). I was able to create the truth table and I think find the Boolean Expression for the problem using the truth table that I created, but I don't think that this is the way that we are supposed to find the Boolean Expression. I was hoping that someone could share some insight on how to find the Boolean Expression without using the truth table.
I would normally not have such an issue with this, but since there are 5 variables, and NAND gates, I am quite confused on how to simplify once I find it.
I think that the outcome is something like:
[(a(bcd)'e)']' when you look at the circuit, and not the table, but I am not entirely sure. I am also not sure on how to simplify this into a Boolean expression if this is right.
Using the truth table to find the minterms I get y= m17+m19+m21+m22+m23+m25+m29 (which I am also not sure if it is right). And if I use a K-Chart to solve this, I end up with y = ab'e + a'ce + ab'cd, which seems like a legitimate simplified Boolean expression, but I have no clue if that is right.
Since this question is worth 20 of the total 100 points, I could really use some help understanding how this works.
Here is the image we were provided:
Circuit: only circuit (a), not (b)
Thank you!
I think that the outcome is something like: [(a(bcd)'e)']'
Your first guess is correct.
You just have to remark that whatever''=whatever
f=[(a(bcd)'e)']' = a(bcd)'e
Using de Morgan (bcd)' = b'+c'+d'
Hence f=a(bcd)'e = ae(b'+c'+d') =ab'e + ac'e + ad'e
which is minimal.
I have been struggling with some questions from my study guide and really am stuck - I have asked the lecturer for help but his answer was literally "but it's been done for you" (referring to gauss_seidel code that was written) - to which I think he missed the point. I'm struggling to understand the actual question and how to approach it.
The first question reads as follows:
Define the 100x100 square matrix A and the column vector b by:
A(ij)=I(ij)+1/((i-j)2+1) b_(i)=1+2/i 1<=i j<=100
where I_(ij) is the 100x100 identity matrix (i.e 1 on the main diagonal and 0 everywhere else). Solve for x using both the Gauss-Seidel method and the A\b construct.
We have written the code for the gauss_seidel method, and i think i understand what it does mostly, however, i do not understand how the above question fits into the method. I was thinking that i'm supposed to do something like the following in the octave window then calling the gauss_seidel method:
>> A=eye(100,100);
>> b= (this is where i get slightly confused)... I've tried doing
>> for b=1:n;
>> b=1+(2/n);
That is question 1.
Question 2 I have given an answer and asked him about but he has not responded.
It reads: The Hilbert matrix is a square n x n matrix defined by:
H_(ij)n = 1/i+j+1
Define bn to be a column vector of dimension n, and with each element 1. Construct bn and then solve for x, Hn xn=bn in the cases n=4.
What i did here was simply:
>> b=ones (4,1);
>> x=hilb(4)\b;
and then it gave me the output of x values. Im not sure if what i did here was correct... since it doesnt mention using any method at all it just says solve for x.
Im not sure how to relate the lecturers reply to understanding the problem.
If you could help me by maybe letting me know what im missing or how i should be thinking about this, it would really help.
the gauss_seidel code looks like this:
function xnew=gauss_seidel(A,b,xold)
n=size(A)(1);
At=A;
xnew=xold;
for k=1:n
At(k,k)=0;
end
for k=1:n
xnew(k)=(b(k)-At(k,:)*xnew)/A(k,k);
end
endfunction
Ive been writing pseudo since last Monday and I am only a little bit clearer on what the code does.
A(ij)=I(ij)+1/((i-j)2+1), b(i)=1+2/i, 1<=i, j<=100
All this is really saying is that we have to create A and b in such a way that i>=1 and j<=100. After doing that, you simply solve using the Gauss Seidel method.
So we'd create b like this:
b=zeros(100,1);
for k=1:100
b(k) = 1+(2/k);
end
This will create a column vector with a size of 100x1 with all the values that satisfy b(i)=1+2/i where i (or in the code,'k') was greater or equal to 1.
Then to create A :
myMatrix=zeros(100,100);
for i=1:100
for j=1:100
myMatrix(i,j) = 1/(((i-j)^2) + 1);
end
end
A=eye(100) + myMatrix;
Now we have created A in such a way that it equals A(ij)=I(ij)+1/((i-j)2+1) where i was greater or equal to 1 & j was less than or equal to 100.
The rest of the question is basically asking to to solve for the values of x using the Gauss Seidel method.
So it be something like this :
y=iterative_linear_solve(A,b,x0,TOL,max_it,method);
Don't forget about creating x0 as the initial assumption, tolerance and max iterations etc.
In terms of question 2, you did exactly what I would have done. I think you're good with that.
I'm not too sure how to answer this :
If you could help me by maybe letting me know what im missing or how i
should be thinking about this, it would really help.
All I can really say is that you need to look at the problems in such a way that you see Ax=b. For example in the first question we started by making b, and then A. After that we simply applied the A\b construct or the Gauss Seidel method and got our answer.
And that's essentially what you did for the second question.
Lastly, are you a UNISA student by chance? I am, haha. I've been struggling with this on my own for a while. The study guides don't seem to give a lot of info.
I am working on a project to create a generic equation solver... envision this to take the form of 25-30 equations that will be saved in a table- variable names along with the operators.
I would then call this table for solving any equation with a missing variable and it would move operators/ other pieces to the other side of the missing variable
e.g. 2x+ 3y=z and if x were missing variable. I would call equation with values for y and z and it would convert to solve for x=(z-3y)/2
equations could be linear, polynomial, binary(yes/no result)...
i am not sure if i can get any light-weight library available or whether this needs to built from scratch... any pointers or guidance will be appreciated
See Maxima.
I rather like it for my symbolic computation needs.
If such a general black-box algorithm could be made accurate, robust and stable, pigs could fly. Solutions can be nonexistent, multiple, parametrized, etc.
Even for linear equations it gets tricky to do it right.
Your best bet is some form of Newton algorithm, but generally you tailor it to your problem at hand.
EDIT: I didn't see you wanted something symbolic, rather than numerical. It's another bag of worms.
Given maths is not my strongest point I'm implementing a bezier curve for 3D animation.
The formula is shown here, and as you can see it is quite nasty. In my programming I use descriptive names, and like to break complex lines down to smaller manageable ones.
How is the best way to handle a scenario like this?
Is it to ignore programming best practices and stick with variable names such as x, y, and t?
In my opinion when you have a predefined mathematical equation it is perfectly acceptable to use short variable names: x, y, t, P_0 etc. which correspond to the equation. Make sure to reference the formula clearly though.
if the formulas is extrated to its own function i'd certainly use the canonical maths representation, and maybe add the wiki page url in a comment
if its imbedded in code with a specific usage of the function then keeping the domain names from your code might be better
it depends
Seeing as only the mathematician in you is actually going to understand the formula, my advice would be to go with a style that a mathematician would be most comfortable with (so letters as variables etc...)
I would also definitely put a comment in there somewhere that clearly states what the formula is, and what it does, for example "This method returns a series of points along a quadratic Bezier curve". That way whenever the programmer in you revisits the code you can safely ignore the mathematical complexity with the assumption that your inner mathematician has already checked to make sure its all ok.
I'd encourage you to use mathematic's best practices and denote variables with letters. Just provide explanation for the variables above the formula. And if you can split the formula to smaller subformulas, even better.
Don't bother. Just reference the documentation (the wikipedia page in this case or even better your own documentation) and make sure the variable names match your documentation. Code comments are just not well suited (nor need them to) describe mathematical formulation.
Sometimes a reference is better than 40 lines of comments or even suggestive variable names.
Make the formula in C# (or other language of preference) resemble the mathematical formula as closely as possible, and include a reference to the formula, including a description of the variables. The idea in coding is to be readable, and if you're dealing with mathematical formulae the most readable representation is the one that looks most like mathematics.
You could key the formula into wolfram alpha ... it will try to simplify for you.
It'll also output in a mathematica friendly style ... funnily enough ;)
Kindness,
Dan
I tend to break an equation down into its root parts.
def sum(array)
array.inject(0) { |result, item| result + item }
end
def average(array)
sum(array) / array.length
end
def sum_squared_error(array)
avg = average(array)
array.inject(0) { |result, item| result + (item - avg) ** 2 }
end
def variance(array)
sum_squared_error(array) / (array.length - 1)
end
def standard_deviation(array)
Math.sqrt(variance(array))
end
You might consider using a domain-specific language to handle this. Mathematica would allow you to write out the equation just as it appears in mathematical notion.
The more your final form resembles the original equation, the more maintainable it will be in the long run (otherwise you have to interpret the code every time you see it).