I just wrote a simple Unix command line utility that could be implemented a lot more efficiently. I can measure its performance by just running it on a number of inputs and measuring the time it takes. This will produce a set of pairs of numbers, s t, where s is the input size and t the processing time. In order to determine the performance characteristics of my utility, I need to fit a function through these data points. I can do this manually, but I prefer to be lazy and let a utility do it for me.
Does such a utility exist?
Its input is a sequence of pairs of numbers.
Its output is a formula that expresses how the second number depends as a function on the first, plus an error measure.
One step of the way is to have a utility that does this just for polynomials.
This has been discussed here but it didn't produce a ready-to-use solution.
The next step is to extend the utility to try non-polynomial terms: negative-degree polynomials (as in y = 1/x) and logarithmic terms (as in y = x log x) will need to be tried as well. One idea to cope with the non-polynomial terms is to just surround the polynomial fitting with x and y scale transformations. I don't know whether that will do. This question is related but not exactly the same.
As I said, I'm lazy: I'm not looking for ideas on how to to write this myself, I'm looking for a reliable result of a project that has already done it for me. Any suggestions?
I believe that SAS has this, RS/1 has this, I think that Mathematica has this, Execel and most spreadsheets have a primitive form of this and usually there are add-ons available for more advanced forms. There are lots of Lab analysis and Statistical analysis tools that have stuff like this.
RE., Command Line Tools:
SAS, RS/1 and Minitab were all command line tools 20 years ago when I used them. I bet at least one of them still has this capability.
Related
I have a regression that I can run for example as
reghdfe y, a(x1_est=x1 x2_est=x2)
which will store the estimated coefficients in x1_est and x2_est. Now, the issue is that using absorb() does not allow me to get the standard errors for these coefficients. If I understand it correctly, no postestimation method of reghdfe allows me to retrieve those.
Luckily, I only care about the standard errors of x1. So, I could instead run
reg y i.x1, a(x2)
and inspect _se[x1]. Unfortunately, x1 has so many different levels that it is not possible to store it as integer, it has to be double. The previous regression hence will fail with x1: factor variables may not contain noninteger values.
What could be another approach to get standard errors for x1?
With large number of fixed effects, STATA's default approaches won't work. One angle is to bootstrap fixed effects and generate standard errors. Again, the issue is that there are so many FE, such that standard bootstrapping methods won't work (cannot return such a large matrix in each bootstrap).
Essentially, to bootstrap the FE, one would (for a large number of iterations)
preserve
bsample
run the regression, reghdfe y, a(x1_est=x1 x2_est-x2)
Store x1_est in a .dta file
restore
After the loop is done, iteratively append all the .dta files, and compute standard errors.
It seems to me that the natural way to integrate hyperband with a bayesian optimization search is to have the search algorithm determine each bracket and have the hyperband scheduler run the bracket. That is to say, the bayesian optimization search runs only once per bracket. Looking at Tune's source code for this, it's not clear to me whether the Tune library applies this strategy or not.
In particular, I want to know how the Tune library handles passing between the search algorithm and trial scheduler. For instance, how does this work if I call SkOptSearch and AsyncHyperBandScheduler (or HyperBandScheduler) together as follows:
sk_search = SkOptSearch(optimizer,
['group','dimensions','normalize','sampling_weights','batch_size','lr_adam','loss_weight'],
max_concurrent=4,
reward_attr="neg_loss",
points_to_evaluate=current_params)
hyperband = AsyncHyperBandScheduler(
time_attr="training_iteration",
reward_attr="neg_loss",
max_t=50,
grace_period=5,
reduction_factor=2,
brackets=5
)
run(Trainable_Dense,
name='hp_search_0',
stop={"training_iteration": 9999,
"neg_loss": -0.2},
num_samples=75,
resources_per_trial={'cpu':4,'gpu':1},
local_dir='./tune_save',
checkpoint_freq=5,
search_alg=sk_search,
scheduler=hyperband,
verbose=2,
resume=False,
reuse_actors=True)
Based on the source code linked above and the source code here, it seems to me that sk_search would return groups of up to 4 trials at a time, but hyperband should be querying the sk_search algorithm for N_sizeofbracket trials at a time.
There is now a Bayesian Optimization HyperBand implementation in Tune - https://ray.readthedocs.io/en/latest/tune-searchalg.html#bohb.
For standard search algorithms and schedulers, the search algorithm currently only sees the result of a trial if it is completed.
I want to fit a curve to data obtained from an FFT. While working on this, I remembered that an FFT gives binned data, and therefore I wondered if I should treat this differently with curve-fitting.
If the bins are narrow compared to the structure, I think it should not be necessary to treat the data differently, but for me that is not the case.
I expect the right way to fit binned data is by minimizing not the difference between values of the bin and fit, but between bin area and the area beneath the fitted curve, for each bin, such that the energy in each bin matches the energy in the range of the bin as signified by the curve.
So my question is: am I thinking correctly about this? If not, how should I go about it?
Also, when looking around for information about this subject, I encountered the "Maximum log likelihood" for example, but did not find enough information about it to understand if and how it applied to my situation.
PS: I have no clue if this is the right site for this question, please let me know if there is a better place.
For an unwindowed FFT, the correct interpolation between bins is by using a Sinc (sin(x)/x) or periodic Sinc (Dirichlet) interpolation kernel. For an FFT of samples of a band-limited signal, thus will reconstruct the continuous spectrum.
A very simple and effective way of interpolating the spectrum (from an FFT) is to use zero-padding. It works both with and without windowing prior to the FFT.
Take your input vector of length N and extend it to length M*N, where M is an integer
Set all values beyond the original N values to zeros
Perform an FFT of length (N*M)
Calculate the magnitude of the ouput bins
What you get is the interpolated spectrum.
Best regards,
Jens
This can be done by using maximum log likelihood estimation. This is a method that finds the set of parameters that is most likely to have yielded the measured data - the technique originates in statistics.
I have finally found an understandable source for how to apply this to binned data. Sadly I cannot enter formulas here, so I refer to that source for a full explanation: slide 4 of this slide show.
EDIT:
For noisier signals this method did not seem to work very well. A method that was a bit more robust is a least squares fit, where the difference between the area is minimized, as suggested in the question.
I have not found any literature to defend this method, but it is similar to what happens in the maximum log likelihood estimation, and yields very similar results for noiseless test cases.
I've been interested in compiler/interpreter design/implementation for as long as I've been programming (only 5 years now) and it's always seemed like the "magic" behind the scenes that nobody really talks about (I know of at least 2 forums for operating system development, but I don't know of any community for compiler/interpreter/language development). Anyways, recently I've decided to start working on my own, in hopes to expand my knowledge of programming as a whole (and hey, it's pretty fun :). So, based off the limited amount of reading material I have, and Wikipedia, I've developed this concept of the components for a compiler/interpreter:
Source code -> Lexical Analysis -> Abstract Syntax Tree -> Syntactic Analysis -> Semantic Analysis -> Code Generation -> Executable Code.
(I know there's more to code generation and executable code, but I haven't gotten that far yet :)
And with that knowledge, I've created a very basic lexer (in Java) to take input from a source file, and output the tokens into another file. A sample input/output would look like this:
Input:
int a := 2
if(a = 3) then
print "Yay!"
endif
Output (from lexer):
INTEGER
A
ASSIGN
2
IF
L_PAR
A
COMP
3
R_PAR
THEN
PRINT
YAY!
ENDIF
Personally, I think it would be really easy to go from there to syntactic/semantic analysis, and possibly even code generation, which leads me to question: Why use an AST, when it seems that my lexer is doing just as good a job? However, 100% of my sources I use to research this topic all seem adamant that this is a necessary part of any compiler/interpreter. Am I missing the point of what an AST really is (a tree that shows the logical flow of a program)?
TL;DR: Currently in route to develop a compiler, finished the lexer, seems to me like the output would make for easy syntactic analysis/semantic analysis, rather than doing an AST. So why use one? Am I missing the point of one?
Thanks!
First off, one thing about your list of components does not make sense. Building an AST is (pretty much) the syntactic analysis, so it either shouldn't be in there, or at least come before the AST.
What you got there is a lexer. All it gives you are individual tokens. In any case, you will need an actual parser, because regular languages aren't any fun to program in. You can't even (properly) nest expressions. Heck, you can't even handle operator precedence. A token stream doesn't give you:
An idea where statements and expressions start and end.
An idea how statements are grouped into blocks.
An idea Which part of the expression has which precedence, associativity, etc.
A clear, uncluttered view at the actual structure of the program.
A structure which can be passed through a myriad of transformations, without every single pass knowing and having code to accomodate that the condition in an if is enclosed by parentheses.
... more generally, any kind of comprehension above the level of a single token.
Suppose you have two passes in your compiler which optimize certain kinds of operators applies to certain arguments (say, constant folding and algebraic simplifications like x - x -> 0). If you hand them tokens for the expression x - x * 1, these passes are cluttered with figuring out that the x * 1 part comes first. And they have to know that, lest the transformation is incorrect (consider 1 + 2 * 3).
These things are tricky enough to get right as it is, so you don't want to be pestered by parsing problems as well. That's why you solve the parsing problem first, in a separate parsing step. Then you can, say, replace a function call with its definition, without worrying about adding parenthesis so the meaning remains the same. You save time, you separate concerns, you avoid repetition, you enable simpler code in many other places, etc.
A parser figures all that out, and builds an AST which consequently holds all that information. Without any further data on the nodes, the shape of the AST alone gives you no. 1, 2, 3, and much more, for free. None of the bazillion passes that follow have to worry about it anymore.
That's not to say you always have to have an AST. For sufficiently simple languages, you can do a single-pass compiler. Instead of generating an AST or some other intermediate representation during parsing, you emit code as you go. However, this becomes harder for less simple languages and you can't reasonably do a lot of stuff (such as 70% of all optimizations and diagnostics -- and yes I just made that number up). Generally, I wouldn't advise you to do this. There are good reasons single-pass compilers are mostly dead. Even languages which permit them (e.g. C) are nowadays implemented with multiple passes and ASTs. It's a simple way to get started, but will severely limit you (and the language, if you design it) later.
You've got the AST at the wrong point in your flow diagram. Typically, the output of the lexer is a series of tokens (as you have in your output), and these are fed to the parser/syntactic analyzer, which generates the AST. So the output of your lexer is different from an AST because they are used at different points in the compilation process and fulfill different purposes.
The next logical question is: What, then, is an AST? Well, the purpose of parsing/syntactic analysis is to turn the series of tokens generated by the lexer into an AST (or parse tree). The AST is an intermediate representation that captures the relationship between syntactical elements in a way that is easier to work with programmatically. One way of thinking about this is that a text program is a one dimensional construct, and can only represent ideas as a sequence of elements, while the AST is freed from this constraint, and can represent the underlying relationships between those elements in 2 dimensions (as typically drawn), or any higher dimension space if you so choose to think about it that way.
For instance, a binary operator has two operands, let's call them A and B. In code, this may be spelled 'A * B' (assuming an infix operator - another advantage of an AST is to hide such distinctions that may be important syntactically, but not semantically), but for the compiler to "understand" this expression, it must read 5 characters sequentially, and this logic can quickly become cumbersome, given the many possibilities in even a small language. In an AST representation, however, we have a "binary operator" node whose value is '*', and that node has two children, values 'A' and 'B'.
As your compiler project progresses, I think you will begin to see the advantages of this representation.
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.