I have a set of examples, which are each annotated with feature data. The examples and features describe the settings of an experiment in an arbitrary domain (e.g. number-of-switches, number-of-days-performed, number-of-participants, etc.). Certain features are fixed (i.e. static), while others I can manually set (i.e. variable) in future experiments. Each example also has a "reward" feature, which is a continuous number bounded between 0 and 1, indicating the success of the experiment as determined by an expert.
Based on this example set, and given a set of static features for a future experiment, how would I determine the optimal value to use for a specific variable so as to maximise the reward?
Also, does this process have a formal name? I've done some research, and this sounds similar to regression analysis, but I'm still not sure if it's the same thing.
The process is called "design of experiments." There are various techniques that can be used depending on the number of parameters, and whether you are able to do computations between trials or if you have to pick all your treatments in advance.
full factorial - try each combination, the brute force method
fractional factorial - eliminate some of the combinations in a pattern and use regression to fill in the missing data
Plackett-Burman, response surface - more sophisticated methods, trading off statistical effort for experimental effort
...and many others. This is an active area of statistical research.
Once you've built a regression model from the data in your experiments, you can find an optimum by applying the usual numerical optimization techniques.
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/
I have a object with many fields. Each field has different range of values. I want to use hypothesis to generate different instances of this object.
Is there a limit to the number of combination of field values Hypothesis can handle? Or what does the search tree hypothesis creates look like? I don't need all the combinations but I want to make sure that I get a fair number of combinations where I test many different values for each field. I want to make sure Hypothesis is not doing a DFS until it hits the max number of examples to generate
TLDR: don't worry, this is a common use-case and even a naive strategy works very well.
The actual search process used by Hypothesis is complicated (as in, "lead author's PhD topic"), but it's definitely not a depth-first search! Briefly, it's a uniform distribution layered on a psudeo-random number generator, with a coverage-guided fuzzer biasing that towards less-explored code paths, with strategy-specific heuristics on top of that.
In general, I trust this process to pick good examples far more than I trust my own judgement, or that of anyone without years of experience in QA or testing research!
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.
In the context of a large-scale data mining benchmarking study, I am comparing 15 algorithms over 9 data sets, leading to an overall 135 algorithm/dataset combinations. The study is done using WEKA.
My last analysis is concerned with the influence of feature selection. I am aware, that there is no such thing as the perfect feature selection algorithm but the optimal choice rather depends on both algorithm to be deployed and the data set to which it will be applied.
Although the problem is to large to find the optimal feature selection algorithm for each combination, I am looking for ones that are considered to show a good performance in general, 'allrounder' so to say.
So far I have found recommendation for CFS (Correlation-based feature selection), ReliefF and Consistency-based subset evaluation (Hall / Holmes 2002) as a generally good choice as well as the note from a survey, that methods as simple as Rankers (e.g. Correlation coefficient) proved quiet effective (Guyon / Ellissef 2003).
Is there a good benchmark study some other research indicating which methods to use or which ones to use in practice?
From a Text Classification point of view, there is one article by Yang etal. comparing different feature selection algorithms (chi square, document frequency and Information Gain).
Although it is focus on text (i.e., the document frequency won't apply to you at all) the others might, depending on the nature of your features (i.e., binary or not, always present, ...)
I hope this helps.
I want to manually implement a classifier for certain short strings of words, getting a "goodness" rank for each of them. I have made a naive Bayesian classifier which is basically spam-filter-like and scores strings based on previous "good"/"bad" ratings. So far so good.
Now, there are two problems that I want to solve (by properly understanding things)...
The question is - what would be good introductory material for below, not of "cookbook" variety but more systematic, and yet ideally shorter than a university statistics course :) Set of articles that is shorter than the book, or a good book. Aimed at programmers ideally.
The problems are:
first, in my system there are actually 3 types of user feedback - "good", "bad", and "neutral". Most items are neutral, and right now I simply don't include them in the ranking. I am wondering how these things are properly handled (I still need to obtain a single "goodness probability" per item, so if I calculate probability of good and bad separately, are there any pitfalls/proper methods to combining those).
Then, I want to remove the naive part from my classifier (i.e. take relations between words into account), so some different classifier may be in order. Or, I could add all pairs-triples-etc. of words as features, since the strings are short - this feels like a hack, but then again my CS/maths background is rusty enough and/or insufficient to say whether this is a valid technique.