Can anyone point me to a source for the use of IIA in rank-ordered logit, and why this results in independence? It seems to me that the conventional formulation of IIA in logit models won't work here, because, eg, Prob[first choice]/Prob[second choice] under the logit formulation will have different sums in the denominators, and hence won't cancel, as they do in "ordinary" logit.
Thanks!
Related
Would this be a valid implementation of a cross entropy loss that takes the ordinal structure of the GT y into consideration? y_hat is the prediction from a neural network.
ce_loss = F.cross_entropy(y_hat, y, reduction="none")
distance_weight = torch.abs(y_hat.argmax(1) - y) + 1
ordinal_ce_loss = torch.mean(distance_weight * ce_loss)
I'll attempt to answer this question by first fully defining the task, since the question is a bit sparse on details.
I have a set of ordinal classes (e.g. first, second, third, fourth,
etc.) and I would like to predict the class of each data example from
among this set. I would like to define an entropy-based loss-function
for this problem. I would like this loss function to weight the loss
between a predicted class torch.argmax(y_hat) and the true class y
according to the ordinal distance between the two classes. Does the
given loss expression accomplish this?
Short answer: sure, it is "valid". You've roughly implemented L1-norm ordinal class weighting. I'd question whether this is truly the correct weighting strategy for this problem.
For instance, consider that for a true label n, the bin n response is weighted by 1, but the bin n+1 and n-1 responses are weighted by 2. This means that a lot more emphasis will be placed on NOT predicting false positives than on correctly predicting true positives, which may imbue your model with some strange bias.
It also means that examples on the edge will result in a larger total sum of weights, meaning that you'll be weighting examples where the true label is say "first" or "last" more highly than the intermediate classes. (Say you have 5 classes: 1,2,3,4,5. A true label of 1 will require distance_weight of [1,2,3,4,5], the sum of which is 15. A true label of 3 will require distance_weight of [3,2,1,2,3], the sum of which is 11.
In general, classification problems and entropy-based losses are underpinned by the assumption that no set of classes or categories is any more or less related than any other set of classes. In essence, the input data is embedded into an orthogonal feature space where each class represents one vector in the basis. This is quite plainly a bad assumption in your case, meaning that this embedding space is probably not particularly elegant: thus, you have to correct for it with sort of a hack-y weight fix. And in general, this assumption of class non-correlation is probably not true in a great many classification problems (consider e.g. the classic ImageNet classification problem, wherein the class pairs [bus,car], and [bus,zebra] are treated as equally dissimilar. But this is probably a digression into the inherent lack of usefulness of strict ontological structuring of information which is outside the scope of this answer...)
Long Answer: I'd highly suggest moving into a space where the ordinal value you care about is instead expressed in a continuous space. (In the first, second, third example, you might for instance output a continuous value over the range [1,max_place]. This allows you to benefit from loss functions that already capture well the notion that predictions closer in an ordered space are better than predictions farther away in an ordered space (e.g. MSE, Smooth-L1, etc.)
Let's consider one more time the case of the [first,second,third,etc.] ordinal class example, and say that we are trying to predict the places of a set of runners in a race. Consider two races, one in which the first place runner wins by 30% relative to the second place runner, and the second in which the first place runner wins by only 1%. This nuance is entirely discarded by the ordinal discrete classification. In essence, the selection of an ordinal set of classes truncates the amount of information conveyed in the prediction, which means not only that the final prediction is less useful, but also that the loss function encodes this strange truncation and binarization, which is then reflected (perhaps harmfully) in the learned model. This problem could likely be much more elegantly solved by regressing the finishing position, or perhaps instead by regressing the finishing time, of each athlete, and then performing the final ordinal classification into places OUTSIDE of the network training.
In conclusion, you might expect a well-trained ordinal classifier to produce essentially a normal distribution of responses across the class bins, with the distribution peak on the true value: a binned discretization of a space that almost certainly could, and likely should, be treated as a continuous space.
This has been bugging me since a long time.
Suppose I have a boolean function F defined as follows:
Now, it can be expressed in its SOP form as:
F = bar(X)Ybar(Z)+ XYZ
But I fail to understand why we always complement the 0s to express them as 1. Is it assumed that the inputs X, Y and Z will always be 1?
What is the practical application of that? All the youtube videos I watched on this topic, how to express a function in SOP form or as sum of minterms but none of them explained why we need this thing? Why do we need minterms in the first place?
As of now, I believe that we design circuits to yield and take only 1 and that's where minterms come in handy. But I couldn't get any confirmation of this thing anywhere so I am not sure I am right.
Maxterms are even more confusing. Do we design circuits that would yield and take only 0s? Is that the purpose of maxterms?
Why do we need minterms in the first place?
We do not need minterms, we need a way to solve a logic design problem, i.e. given a truth table, find a logic circuit able to reproduce this truth table.
Obviously, this requires a methodology. Minterm and sum-of-products is mean to realize that. Maxterms and product-of-sums is another one. In either case, you get an algebraic representation of your truth table and you can either implement it directly or try to apply standard theorems of boolean algebra to find an equivalent, but simpler, representation.
But these are not the only tools. For instance, with Karnaugh maps, you rewrite your truth table with some rules and you can simultaneously find an algebraic representation and reduce its complexity, and it does not consider minterms. Its main drawback is that it becomes unworkable if the number of inputs rises and it cannot be considered as a general way to solve the problem of logic design.
It happens that minterms (or maxterms) do not have this drawback, and can be used to solve any problem. We get a trut table and we can directly convert it in an equation with ands, ors and nots. Indeed minterms are somehow simpler to human beings than maxterms, but it is just a matter of taste or of a reduced number of parenthesis, they are actually equivalent.
But I fail to understand why we always complement the 0s to express them as 1. Is it assumed that the inputs X, Y and Z will always be 1?
Assume that we have a truth table, with only a given output at 1. For instance, as line 3 of your table. It means that when x=0, y=1 and z=0 , the output will be zero. So, can I express that in boolean logic? With the SOP methodology, we say that we want a solution for this problem that is an "and" of entries or of their complement. And obviously the solution is "x must be false and y must be true and z must be false" or "(not x) must be true and y must be true and (not z) must be true", hence the minterm /x.y./z. So complementing when we have a 0 and leaving unchanged when we have a 1 is way to find the equation that will be true when xyz=010
If I have another table with only one output at 1 (for instance line 8 of your table), we can find similarly that I can implement this TT with x.y.z.
Now if I have a TT with 2 lines at 1, one can use the property of OR gates and do the OR of the previous circuits. when the output of the first one is 1, it will force this behavior and ditto for the second. And we directly get the solution for your table /xy/z+xyz
This can be extended to any number of ones in the TT and gives a systematic way to find an equation equivalent to a truth table.
So just think of minterms and maxterms as a tool to translate a TT into equations. What is important is the truth table (that describes the behaviour of what you want to do) and the equations (that give you a way to realize it).
I tried these two methods with pycaffe:
loss_weigth=100 in prototxt;
net.blobs['fc'].diff[...] = A_loss + 100*B_loss.
I thought they do the same thing in BP theory, model loss shows the opposite results however.
I want to know what is the difference between those two methods? How should I deal with loss weights if there are multiple losses?
I am just getting start with deep reinforcement learning and i am trying to crasp this concept.
I have this deterministic bellman equation
When i implement stochastacity from the MDP then i get 2.6a
My equation is this assumption correct. I saw this implementation 2.6a without a policy sign on the state value function. But to me this does not make sense due to i am using the probability of which different next steps i could end up in. Which is the same as saying policy, i think. and if yes 2.6a is correct, can i then assume that the rest (2.6b and 2.6c) because then i would like to write the action state function like this:
The reason why i am doing it like this is because i would like to explain myself from a deterministic point of view to a non-deterministic point of view.
I hope someone out there can help on this one!
Best regards Søren Koch
No, the value function V(s_t) does not depend on the policy. You see in the equation that it is defined in terms of an action a_t that maximizes a quantity, so it is not defined in terms of actions as selected by any policy.
In the nondeterministic / stochastic case, you will have that sum over probabilities multiplied by state-values, but this is still independent from any policy. The sum only sums over different possible future states, but every multiplication involves exactly the same (policy-independent) action a_t. The only reason why you have these probabilities is because in the nondeterministic case a specific action in a specific state can lead to one of multiple different possible states. This is not due to policies, but due to stochasticity in the environment itself.
There does also exist such a thing as a value function for policies, and when talking about that a symbol for the policy should be included. But this is typically not what is meant by just "Value function", and also does not match the equation you have shown us. A policy-dependent function would replace the max_{a_t} with a sum over all actions a, and inside the sum the probability pi(s_t, a) of the policy pi selecting action a in state s_t.
Yes, your assumption is completely right. In the Reinforcement Learning field, a value function is the return obtained by starting for a particular state and following a policy π . So yes, strictly speaking, it should be accompained by the policy sign π .
The Bellman equation basically represents value functions recursively. However, it should be noticed that there are two kinds of Bellman equations:
Bellman optimality equation, which characterizes optimal value functions. In this case, the value function it is implicitly associated with the optimal policy. This equation has the non linear maxoperator and is the one you has posted. The (optimal) policy dependcy is sometimes represented with an asterisk as follows:
Maybe some short texts or papers omit this dependency assuming it is obvious, but I think any RL text book should initially include it. See, for example, Sutton & Barto or Busoniu et al. books.
Bellman equation, which characterizes a value function, in this case associated with any policy π:
In your case, your equation 2.6 is based on the Bellman equation, therefore it should remove the max operator and include the sum over all actions and possible next states. From Sutton & Barto (sorry by the notation change wrt your question, but I think it's understable):
I am working on an multi-class image recognition problem. The task is to have the correct answer being in the top 3 output probabilities. So I was thinking that maybe there exists a clever cost function that prioritizes the correct answer being in the top K and doesn't penalize much in between these top K.
This can be achieved by class-weighted cross-entropy loss, which essentially assigns the weight to the errors associated with each class. This loss is used in research, e.g. see the paper "Multi-task learning and Weighted Cross-entropy for DNN-based Keyword" by S. Panchapagesan at al. Before computing the cross-entropy, you can check if the predicted distribution satisfies your condition (e.g., ground truth class is in top-k of the predicted classes) and assign the zero (or near zero) weights accordingly, if it does.
There are open questions though: when the correct class is in top-k, should you penalize the k-1 incorrectly predicted classes? What if, for example, the prediction is (0.9, 0.05, 0.01, ...), the third class is correct and it is in top-3 -- is this prediction good enough or not? Should you care what exactly k-1 incorrect classes are?
All these question arise because this kind of loss doesn't have probabilistic interpretation, unlike standard cross-entropy. That's why I wouldn't recommend using it in practice, but reformulate the goal instead.
E.g., if the original problem is that for some inputs several classes are equally good, the best way to deal with it is to use soft labels, e.g. (0.33, 0.33, 0.33, 0, 0, 0, ...) instead of one-hot (note that this totally agrees with probabilistic interpretation). It will force the network to learn features associated with all three good classes, and generally lead to the same goal, but with better control over target classes.