I've written a GA to model a handful of stocks (4) over a period of time (5 years). It's impressive how quickly the GA can find an optimal solution to the training data, but I am also aware that this is mainly due to it's tendency to over-fit in the training phase.
However, I still thought I could take a few precautions and and get some kind of prediction on a set of unseen test stocks from the same period.
One precaution I took was:
When multiple stocks can be bought on the same day the GA only buys one from the list and it chooses this one randomly. I thought this randomness might help to avoid over-fitting?
Even if over-fitting is still occurring,shouldn't it be absent in the initial generations of the GA since it hasn't had a chance to over-fit yet?
As a note, I am aware of the no-free-lunch theorem which demonstrates ( I believe) that there is no perfect set of parameters which will produce an optimal output for two different datasets. If we take this further, does this no-free-lunch theorem also prohibit generalization?
The graph below illustrates this.
->The blue line is the GA output.
->The red line is the training data (slightly different because of the aforementioned randomness)
-> The yellow line is the stubborn test data which shows no generalization. In fact this is the most flattering graph I could produce..
The y-axis is profit, the x axis is the trading strategies sorted from worst to best ( left to right) according to there respective profits (on the y axis)
Some of the best advice I've received so far (thanks seaotternerd) is to focus on the earlier generations and increase the number of training examples. The graph below has 12 training stocks rather than just 4, and shows only the first 200 generations (instead of 1,000). Again, it's the most flattering chart I could produce, this time with medium selection pressure. It certainly looks a little bit better, but not fantastic either. The red line is the test data.
The problem with over-fitting is that, within a single data-set it's pretty challenging to tell over-fitting apart from actually getting better in the general case. In many ways, this is more of an art than a science, but here are some general guidelines:
A GA will learn to do exactly what you attach fitness to. If you tell it to get really good at predicting one series of stocks, it will do that. If you keep swapping in different stocks to predict, though, you might be more successful at getting it to generalize. There are a few ways to do this. The one that has had perhaps the most promising results for reducing over-fitting is imposing spatial structure on the population and evaluating on different test cases in different cells, as in the SCALP algorithm. You could also switch out the test cases on a time basis, but I've had more mixed results with that sort of an approach.
You are correct that over-fitting should be less of a problem early on. Generally, the longer you run a GA, the more over-fitting will be possible. Typically, people tend to assume that the general rules will be learned first, before the rote memorization of over-fitting takes place. However, I don't think I've actually ever seen this studied rigorously - I could imagine a scenario where over-fitting was so much easier than finding general rules that it happens first. I have no idea how common that is, though. Stopping early will also reduce the ability of the GA to find better general solutions.
Using a larger data-set (four stocks isn't that many) will make your GA less susceptible to over-fitting.
Randomness is an interesting idea. It will definitely hurt the GA's ability to find general rules, but it should also reduce over-fitting. Without knowing more about the specifics of your algorithm, it's hard to say which would win out.
That's a really interesting thought about the no free lunch theorem. I'm not 100% sure, but I think it does apply here to some extent - better fitting some data will make your results fit other data worse, by necessity. However, as wide as the range of possible stock behaviors is, it is much narrower than the range of all possible time series in general. This is why it is possible to have optimization algorithms at all - a given problem that we are working with tends produce data that cluster relatively closely together, relative to the entire space of possible data. So, within that set of inputs that we actually care about, it is possible to get better. There is generally an upper limit of some sort on how well you can do, and it is possible that you have hit that upper limit for your data-set. But generalization is possible to some extent, so I wouldn't give up just yet.
Bottom line: I think that varying the test cases shows the most promise (although I'm biased, because that's one of my primary areas of research), but it is also the most challenging solution, implementation-wise. So as a simpler fix you can try stopping evolution sooner or increasing your data-set.
Related
I've been working on a sentence transformation task that involves paraphrase identification as a critical step: if we are confident enough that the state of the program (a sentence repeatedly modified) has become a paraphrase of a target sentence, stop transforming. The overall goal is actually to study potential reasoning in predictive models that can generate language prior to a target sentence. The approach is just one specific way of reaching that goal. Nevertheless, I've become interested in the paraphrase identification task itself, as it's received some boost from language models recently.
The problem I run into is when I manipulate sentences from examples or datasets. For example, in this HuggingFace example, if I negate either sequence or change the subject to Bloomberg, I still get a majority "is paraphrase" prediction. I started going through many examples in the MSRPC training set and negating one sentence in a positive example or making one sentence in a negative example a paraphrase of the other, especially when doing so would be a few word edit. I found to my surprise that various language models, like bert-base-cased-finetuned-mrpc and textattack/roberta-base-MRPC, don't change their confidences much on these sorts of changes. It's surprising as these models claim an f1 score of 0.918+. The dataset is clearly missing a focus on negative examples and small perturbative examples.
My question is, are there datasets, techniques, or models that deal well when given small edits? I know that this is an extremely generic question, much more than is typically asked on StackOverflow, but my concern is in finding practical tools. If there is a theoretical technique, then it might not be suitable as I'm in the category of "available tools define your approach" rather than vice-versa. So I hope that the community would have a recommendation on this.
Short answer to the question: yes, they are overfitting. Most of the important NLP data sets are not actually well-crafted enough to test what they claim to test, and instead test the ability of the model to find subtle (and not-so-subtle) patterns in the data.
The best tool I know for creating data sets that help deal with this is Checklist. The corresponding paper, "Beyond Accuracy: Behavioral Testing of NLP models with CheckList" is very readable and goes into depth on this type of issue. They have a very relevant table... but need some terms:
We prompt users to evaluate each capability with
three different test types (when possible): Minimum Functionality tests, Invariance, and Directional Expectation tests... A Minimum Functionality test (MFT), is a collection of simple examples (and labels) to check a
behavior within a capability. MFTs are similar to
creating small and focused testing datasets, and are
particularly useful for detecting when models use
shortcuts to handle complex inputs without actually
mastering the capability.
...An Invariance test (INV) is when we apply
label-preserving perturbations to inputs and expect
the model prediction to remain the same.
A Directional Expectation test (DIR) is similar,
except that the label is expected to change in a certain way. For example, we expect that sentiment
will not become more positive if we add “You are
lame.” to the end of tweets directed at an airline
(Figure 1C).
I haven't been actively involved in NLG for long, so this answer will be a bit more anecdotal than SO's algorithms would like. Starting with the fact that in my corner of Europe, the general sentiment toward peer review requirements for any kind of NLG project are higher by several orders of magnitude compared to other sciences - and likely not without reason or tensor thereof.
This makes funding a bigger challenge, so wherever you are, I wish you luck on that front. I'm not sure of how big of a deal this site is in the niche, but [Ehud Reiter's Blog][1] is where I would start looking into your tooling ideas.
Maybe even reach out to them/him personally, because I can't think of another source that has an academic background and a strong propensity for practical applications of NLG, at least based on the kind of content they've been putting out over the years.
Your background, environment/funding, and seniority level/control you have over the project will eventually compose your vector decision for you. I's just how it goes on the bleeding edge of anything. What I will add, though, is not to limit yourself to a single language or technology in this phase because of those precise reasons you've mentioned. I'd recommend the same in terms of potential open source involvement but if your profile information is accurate, that probably won't happen, no matter what you do and accomplish.
But yeah, in the grand scheme of things, your question is far from too broad, in my view. It identifies a rather unmistakable problem pattern that not all branches of science are as lackadaisical to approach as NLG-adjacent fields seem to be right now. In that regard, it's not broad enough and will need to be promulgated far and wide before community-driven tooling will give you serious options on a micro level.
Blasphemy, sure, but the performance is already stacked against you As for the question potentially being too broad, I'd posit it is not broad enough, so long as we collectively remain in a "oh, I was waiting for you to start doing something about it" phase.
P.S. I'd eliminate any Rust and ECMAScript alternatives prior to looking into Python, blapshemous as this might sound to a 2021 data scientist
. Some might ARight nowccounting forr the ridicule this would receive xou sltrsfx hsbr s fszs drz zhsz s mrnzsl rcrtvidr, sz lrsdz
due to performance easons.
[1]: https://ehudreiter.com/2016/12/18/nlg-vs-templates/
In below post of Analytics Vidya, ANOVA test has been performed on COVID data, to check whether the difference in posotive cases of denser region is statistically significant.
I believe ANOVA test can’t be performed on this COVID time series data, atleast not in way as it has been done in this post.
Sample data has been consider randomly from different groups(denser1, denser2…denser4). The data is time series so it is more likely that number of positive cases in random sample of groups will be from different point of time.
There might be the case denser1 has random data from early covid time and another region has random data from another point of time. If this is the case, then F-Statistics will high certainly.
Can anyone explain if you have other opinions?
https://www.analyticsvidhya.com/blog/2020/06/introduction-anova-statistics-data-science-covid-python/
ANOVA should not be applied to time-series data, as the independence assumption is violated. The issue with independence is that days tend to correlate very highly. For example, if you know that today you have 1400 positive cases, you would expect tomorrow to have a similar number of positive cases, regardless of any underlying trends.
It sounds like you're trying to determine causality of different treatments (ie mask mandates or other restrictions etc) and their effects on positive cases. The best way to infer causality is usually to perform A-B testing, but obviously in this case it would not be reasonable to give different populations different treatments. One method that is good for going back and retro-actively inferring causality is called "synthetic control".
https://economics.mit.edu/files/17847
Above is linked a basic paper on the methodology. The hard part of this analysis will be in constructing synthetic counterfactuals or "controls" to test your actual population against.
If this is not what you're looking for, please reply with a clarifying question, but I think this should be an appropriate method that is well-suited to studying time-series data.
I want to test for heteroskedasticity and autocorrelation in a large unbalanced panel dataset.
I do so using the following code:
* Heteroskedasticity test
// iterated GLS with only heteroskedasticity produces
// maximum-likelihood parameter estimates
xtgls adjusted_volume ibn.rounded_time i.id i.TRD_EVENT_DT, igls panels(heteroskedastic)
estimates store hetero
* Autocorrelation
findit xtserial
net sj 3-2 st0039
net install st0039
xtserial adjusted_volume ibn.rounded_time i.id i.TRD_EVENT_DT
Though I use the calculation power of high process center, because of the iteration method, this procedure takes more than 15 hours.
What is the most efficient program to perform these tests using Stata?
This question is borderline off-topic and quite broad, but i suspect still of
considerable interest to new users. As such, here i will try to consolidate our
conversation in the comments as an answer.
I strongly advise in the future to refrain from using highly subjective
words such as 'best', which can mean different things to different people. Or
terms like 'efficient', which can have a different meaning in a different context.
It is also difficult to provide specific advice regarding the use of commands
when we know nothing about what you are trying to do.
In my view, the 'best' choice, is the choice that gets the job done as accurately
as possible given the available data. Speed is an important consideration nowadays, but accuracy is still the most fundamental one. As you continue to use Stata, you will see that it has a considerable number of commands, often with overlapping functionality. Depending on the use case, sometimes opting for one implementation over another can be 'better', in the sense that it may be more practical or faster in achieving the desired end result.
Case in point, your comment in your previous post where the noconstant option is unavailable in rreg. In that particular context you can get a reasonably good alternative using regress with the vce(robust) option. In fact, this alternative may often be adequate for several use cases.
In this particular example, xtgls will be considerably faster if the igls
option is not used. This will be especially true with larger and more 'difficult' datasets. In cases where MLE is necessary, the iterate option will allow you to specify a fixed number of iterations, which could speed things up but can be a recipe for disaster if you don't know what you are doing and is thus not recommended. This option is usually used for other purposes. However, is xtgls the only command you could use? Read here why this may in fact not necessarily be the case.
Regarding speed, Stata in general is slow, at least when the ado language is used. This is because it is an interpreted language. The only realistic option for speed gains here is through parallelisation if you have Stata MP. Even in this case, whether any gains are achieved it will depend on a number of factors,
including which command you use.
Finally, xtserial is a community-contributed command, something which you
fail to make clear in your question. It is customary and useful to provide this
information right from the start, so others know that you do not refer to an
official, built-in command.
I am currently working on a search ranking algorithm which will be applied to elastic search queries (domain: e-commerce). It assigns scores on several entities returned and finally sorts them based on the score assigned.
My question is: Has anyone ever tried to introduce a certain level of randomness to any search algorithm and has experienced a positive effect of it. I am thinking that it might be useful to reduce bias and promote the lower ranking items to give them a chance to be seen easier and get popular if they deserve it. I know that some machine learning algorithms are introducing some randomization to reduce the bias so I thought it might be applied to search as well.
Closest I can get here is this but not exactly what I am hoping to get answers for:
Randomness in Artificial Intelligence & Machine Learning
I don't see this mentioned in your post... Elasticsearch offers a random scoring feature: https://www.elastic.co/guide/en/elasticsearch/guide/master/random-scoring.html
As the owner of the website, you want to give your advertisers as much exposure as possible. With the current query, results with the same _score would be returned in the same order every time. It would be good to introduce some randomness here, to ensure that all documents in a single score level get a similar amount of exposure.
We want every user to see a different random order, but we want the same user to see the same order when clicking on page 2, 3, and so forth. This is what is meant by consistently random.
The random_score function, which outputs a number between 0 and 1, will produce consistently random results when it is provided with the same seed value, such as a user’s session ID
Your intuition is right - randomization can help surface results that get a lower than deserved score due to uncertainty in the estimation. Empirically, Google search ads seemed to have sometimes been randomized, and e.g. this paper is hinting at it (see Section 6).
This problem describes an instance of a class of problems called Explore/Exploit algorithms, or Multi-Armed Bandit problems; see e.g. http://en.wikipedia.org/wiki/Multi-armed_bandit. There is a large body of mathematical theory and algorithmic approaches. A general idea is to not always order by expected, "best" utility, but by an optimistic estimate that takes the degree of uncertainty into account. A readable, motivating blog post can be found here.
Closed. This question is off-topic. It is not currently accepting answers.
Want to improve this question? Update the question so it's on-topic for Stack Overflow.
Closed 11 years ago.
Improve this question
Basically, I have a reasonably large list (a year's worth of data) of times that a single discrete event occurred (for my current project, a list of times that someone printed something). Based on this list, I would like to construct a statistical model of some sort that will predict the most likely time for the next event (the next print job) given all of the previous event times.
I've already read this, but the responses don't exactly help out with what I have in mind for my project. I did some additional research and found that a Hidden Markov Model would likely allow me to do so accurately, but I can't find a link on how to generate a Hidden Markov Model using just a list of times. I also found that using a Kalman filter on the list may be useful but basically, I'd like to get some more information about it from someone who's actually used them and knows their limitations and requirements before just trying something and hoping it works.
Thanks a bunch!
EDIT: So by Amit's suggestion in the comments, I also posted this to the Statistics StackExchange, CrossValidated. If you do know what I should do, please post either here or there
I'll admit it, I'm not a statistics kind of guy. But I've run into these kind of problems before. Really what we're talking about here is that you have some observed, discrete events and you want to figure out how likely it is you'll see them occur at any given point in time. The issue you've got is that you want to take discrete data and make continuous data out of it.
The term that comes to mind is density estimation. Specifically kernel density estimation. You can get some of the effects of kernel density estimation by simple binning (e.g. count the number events in a time interval such as every quarter hour or hour.) Kernel density estimation just has some nicer statistical properties than simple binning. (The produced data is often 'smoother'.)
That only takes care of one of your problems, though. The next problem is still the far more interesting one -- how do you take a time line of data (in this case, only printer data) and produced a prediction from it? First thing's first -- the way you've set up the problem may not be what you're looking for. While the miracle idea of having a limited source of data and predicting the next step of that source sounds attractive, it's far more practical to integrate more data sources to create an actual prediction. (e.g. maybe the printers get hit hard just after there's a lot of phone activity -- something that can be very hard to predict in some companies) The Netflix Challenge is a rather potent example of this point.
Of course, the problem with more data sources is that there's extra legwork to set up the systems that collect the data then.
Honestly, I'd consider this a domain-specific problem and take two approaches: Find time-independent patterns, and find time-dependent patterns.
An example time-dependent pattern would be that every week day at 4:30 Suzy prints out her end of the day report. This happens at specific times every day of the week. This kind of thing is easy to detect with fixed intervals. (Every day, every week day, every weekend day, every Tuesday, every 1st of the month, etc...) This is extremely simple to detect with predetermined intervals -- just create a curve of the estimated probability density function that's one week long and go back in time and average the curves (possibly a weighted average via a windowing function for better predictions).
If you want to get more sophisticated, find a way to automate the detection of such intervals. (Likely the data wouldn't be so overwhelming that you could just brute force this.)
An example time-independent pattern is that every time Mike in accounting prints out an invoice list sheet, he goes over to Johnathan who prints out a rather large batch of complete invoice reports a few hours later. This kind of thing is harder to detect because it's more free form. I recommend looking at various intervals of time (e.g. 30 seconds, 40 seconds, 50 seconds, 1 minute, 1.2 minutes, 1.5 minutes, 1.7 minutes, 2 minutes, 3 minutes, .... 1 hour, 2 hours, 3 hours, ....) and subsampling them via in a nice way (e.g. Lanczos resampling) to create a vector. Then use a vector-quantization style algorithm to categorize the "interesting" patterns. You'll need to think carefully about how you'll deal with certainty of the categories, though -- if your a resulting category has very little data in it, it probably isn't reliable. (Some vector quantization algorithms are better at this than others.)
Then, to create a prediction as to the likelihood of printing something in the future, look up the most recent activity intervals (30 seconds, 40 seconds, 50 seconds, 1 minute, and all the other intervals) via vector quantization and weight the outcomes based on their certainty to create a weighted average of predictions.
You'll want to find a good way to measure certainty of the time-dependent and time-independent outputs to create a final estimate.
This sort of thing is typical of predictive data compression schemes. I recommend you take a look at PAQ since it's got a lot of the concepts I've gone over here and can provide some very interesting insight. The source code is even available along with excellent documentation on the algorithms used.
You may want to take an entirely different approach from vector quantization and discretize the data and use something more like a PPM scheme. It can be very much simpler to implement and still effective.
I don't know what the time frame or scope of this project is, but this sort of thing can always be taken to the N-th degree. If it's got a deadline, I'd like to emphasize that you worry about getting something working first, and then make it work well. Something not optimal is better than nothing.
This kind of project is cool. This kind of project can get you a job if you wrap it up right. I'd recommend you do take your time, do it right, and post it up as function, open source, useful software. I highly recommend open source since you'll want to make a community that can contribute data source providers in more environments that you have access to, will to support, or time to support.
Best of luck!
I really don't see how a Markov model would be useful here. Markov models are typically employed when the event you're predicting is dependent on previous events. The canonical example, of course, is text, where a good Markov model can do a surprisingly good job of guessing what the next character or word will be.
But is there a pattern to when a user might print the next thing? That is, do you see a regular pattern of time between jobs? If so, then a Markov model will work. If not, then the Markov model will be a random guess.
In how to model it, think of the different time periods between jobs as letters in an alphabet. In fact, you could assign each time period a letter, something like:
A - 1 to 2 minutes
B - 2 to 5 minutes
C - 5 to 10 minutes
etc.
Then, go through the data and assign a letter to each time period between print jobs. When you're done, you have a text representation of your data, and that you can run through any of the Markov examples that do text prediction.
If you have an actual model that you think might be relevant for the problem domain, you should apply it. For example, it is likely that there are patterns related to day of week, time of day, and possibly date (holidays would presumably show lower usage).
Most raw statistical modelling techniques based on examining (say) time between adjacent events would have difficulty capturing these underlying influences.
I would build a statistical model for each of those known events (day of week, etc), and use that to predict future occurrences.
I think the predictive neural network would be a good approach for this task.
http://en.wikipedia.org/wiki/Predictive_analytics#Neural_networks
This method is also used for predicting f.x. weather forecasting, stock marked, sun spots.
There's a tutorial here if you want to know more about how it works.
http://www.obitko.com/tutorials/neural-network-prediction/
Think of a markov chain like a graph with vertex connect to each other with a weight or distance. Moving around this graph would eat up the sum of the weights or distance you travel. Here is an example with text generation: http://phpir.com/text-generation.
A Kalman filter is used to track a state vector, generally with continuous (or at least discretized continuous) dynamics. This is sort of the polar opposite of sporadic, discrete events, so unless you have an underlying model that includes this kind of state vector (and is either linear or almost linear), you probably don't want a Kalman filter.
It sounds like you don't have an underlying model, and are fishing around for one: you've got a nail, and are going through the toolbox trying out files, screwdrivers, and tape measures 8^)
My best advice: first, use what you know about the problem to build the model; then figure out how to solve the problem, based on the model.