I was looking for this information in the tensorflow_decision_forests docs (https://github.com/tensorflow/decision-forests) (https://www.tensorflow.org/decision_forests/api_docs/python/tfdf/keras/wrappers/CartModel) and yggdrasil_decision_forests docs (https://github.com/google/yggdrasil-decision-forests).
I've also taken a look at the code of these two libraries, but I didn't find that information.
I'm also curious if I can specify an impurity index to use.
I'm looking for some analogy to sklearn decision tree, where you can specify the impurity index with criterion parameter.
https://scikit-learn.org/stable/modules/generated/sklearn.tree.DecisionTreeClassifier.html
For TensorFlow Random Forest i found only a parameter uplift_split_score:
uplift_split_score: For uplift models only. Splitter score i.e. score
optimized by the splitters. The scores are introduced in "Decision trees
for uplift modeling with single and multiple treatments", Rzepakowski et
al. Notation: p probability / average value of the positive outcome,
q probability / average value in the control group.
- KULLBACK_LEIBLER or KL: - p log (p/q)
- EUCLIDEAN_DISTANCE or ED: (p-q)^2
- CHI_SQUARED or CS: (p-q)^2/q
Default: "KULLBACK_LEIBLER".
I'm not sure if it's a good lead.
No, you shouldn't use uplift_split_score, because it is For uplift models only.
Uplift modeling is used to estimate treatment effect or other tasks in causal inference
Related
I am given a data that consists of N sequences of variable lengths of hidden variables and their corresponding observed variables (i.e., I have both the hidden variables and the observed variables for each sequence).
Is there a way to find the order K of the "best" HMM model for this data, without exhaustive search? (justified heuristics are also legitimate).
I think there may be a confusion about the word "order":
A first-order HMM is an HMM which transition matrix depends only on the previous state. A 2nd-order HMM is an HMM which transition matrix depends only on the 2 previous states, and so on. As the order increases, the theory gets "thicker" (i.e., the equations) and very few implementations of such complex models are implemented in mainstream libraries.
A search on your favorite browser with the keywords "second-order HMM" will bring you to meaningful readings about these models.
If by order you mean the number of states, and with the assumptions that you use single distributions assigned to each state (i.e., you do not use HMMs with mixtures of distributions) then, indeed the only hyperparameter you need to tune is the number of states.
You can estimate the optimal number of states using criteria such as the Bayesian Information Criterion, the Akaike Information Criterion, or the Minimum Message Length Criterion which are based on model's likelihood computations. Usually, the use of these criteria necessitates training multiple models in order to be able to compute some meaningful likelihood results to compare.
If you just want to get a blur idea of a good K value that may not be optimal, a k-means clustering combined with the percentage of variance explained can do the trick: if X clusters explain more than, let say, 90% of the variance of the observations in your training set then, going with an X-state HMM is a good start. The 3 first criteria are interesting because they include a penalty term that goes with the number of parameters of the model and can therefore prevent some overfitting.
These criteria can also be applied when one uses mixture-based HMMs, in which case there are more hyperparameters to tune (i.e., the number of states and the number of component of the mixture models).
I have a combinatorial optimization problem for which I have a genetic algorithm to approximate the global minima.
Given X elements find: min f(X)
Now I want to expand the search over all possible subsets and to find the one subset where its global minimum is maximal compared to all other subsets.
X* are a subset of X, find: max min f(X*)
The example plot shows all solutions of three subsets (one for each color). The black dot indicates the highest value of all three global minima.
image: solutions over three subsets
The main problem is that evaluating the fitness between subsets runs agains the convergence of the solution within a subset. Further the solution is actually a local minimum.
How can this problem be generally described? I couldn't find a similar problem in the literature so far. For example if its solvable with a multi-object genetic algorithm.
Any hint is much appreciated.
While it may not always provide exactly the highest minima (or lowest maxima), a way to maintain local optima with genetic algorithms consists in implementing a niching method. These are ways to maintain population diversity.
For example, in Niching Methods for Genetic Algorithms by Samir W. Mahfoud 1995, the following sentence can be found:
Using constructed models of fitness sharing, this study derives lower bounds on the population size required to maintain, with probability gamma, a fixed number of desired niches.
If you know the number of niches and you implement the solution mentioned, you could theoretically end up with the local optima you are looking for.
I am modeling a perceptual process in tensorflow. In the setup I am interested in, the modeled agent is playing a resource game: it has to choose 1 out of n resouces, by relying only on the label that a classifier gives to the resource. Each resource is an ordered pair of two reals. The classifier only sees the first real, but payoffs depend on the second. There is a function taking first to second.
Anyway, ideally I'd like to train the classifier in the following way:
In each run, the classifier give labels to n resources.
The agent then gets the payoff of the resource corresponding to the highest label in some predetermined ranking (say, A > B > C > D), and randomly in case of draw.
The loss is taken to be the normalized absolute difference between the payoff thus obtained and the maximum payoff in the set of resources. I.e., (Payoff_max - Payoff) / Payoff_max
For this to work, one needs to run inference n times, once for each resource, before calculating the loss. Is there a way to do this in tensorflow? If I am tackling the problem in the wrong way feel free to say so, too.
I don't have much knowledge in ML aspects of this, but from programming point of view, I can see doing it in two ways. One is by copying your model n times. All the copies can share the same variables. The output of all of these copies would go into some function that determines the the highest label. As long as this function is differentiable, variables are shared, and n is not too large, it should work. You would need to feed all n inputs together. Note that, backprop will run through each copy and update your weights n times. This is generally not a problem, but if it is, I heart about some fancy tricks one can do by using partial_run.
Another way is to use tf.while_loop. It is pretty clever - it stores activations from each run of the loop and can do backprop through them. The only tricky part should be to accumulate the inference results before feeding them to your loss. Take a look at TensorArray for this. This question can be helpful: Using TensorArrays in the context of a while_loop to accumulate values
I use the python implementation of XGBoost. One of the objectives is rank:pairwise and it minimizes the pairwise loss (Documentation). However, it does not say anything about the scope of the output. I see numbers between -10 and 10, but can it be in principle -inf to inf?
good question. you may have a look in kaggle competition:
Actually, in Learning to Rank field, we are trying to predict the relative score for each document to a specific query. That is, this is not a regression problem or classification problem. Hence, if a document, attached to a query, gets a negative predict score, it means and only means that it's relatively less relative to the query, when comparing to other document(s), with positive scores.
It gives predicted score for ranking.
However, the scores are valid for ranking only in their own groups.
So we must set the groups for input data.
For esay ranking, refer to my project xgboostExtension
If I understand your questions correctly, you mean the output of the predict function on a model fitted using rank:pairwise.
Predict gives the predicted variable (y_hat).
This is the same for reg:linear / binary:logistic etc. The only difference is that reg:linear builds trees to Min(RMSE(y, y_hat)), while rank:pairwise build trees to Max(Map(Rank(y), Rank(y_hat))). However, output is always y_hat.
Depending on the values of your dependent variables, output can be anything. But I typically expect output to be much smaller in variance vs the dependent variable. This is usually the case as it is not necessary to fit extreme data values, the tree just needs to produce predictors that are large/small enough to be ranked first/last in the group.
I would like to perform feature analysis in WEKA. I have a data set of 8 features and 65 instances.
I would like to perform feature selection and optimization functionalities that are available for machine learning methods like SVM.
For example in Weka I would like to know how I can display which of the features contribute best to the classification result.
I think that WEKA provides a nice graphical user interface and allows a very detailed analysis of the influence of single features. But I dont know how to use it. Any help?
You have two options:
You can perform attribute selection using filters. For instance you can use the AttributeSelection tab (or filter) with the search method Ranker and the attribute evaluation metric InfoGainAttributeEval. This way you get a ranked list of the most predictive features according to its Information Gain score. I have done this many times with good results. Sometimes it helps even to increase the accuracy of SVMs, which are known not to need (too much) of feature selection. You can try with other search methods in order to find subgroups of coupled predictors, and with other metrics.
You can just look at the coefficients in the SVM output. For instance, in linear SVMs, the classifier is a polynomial like a1.f1 + a2.f2 + ... + an.fn + fn+1 > 0, being ai the attribute values for an instance, and fi the "weights" obtained in the SVM training algorithm. In consequence, those weights with values close to 0 represent attributes that do not count too much, thus being bad predictors; extreme weights (either positive or negative) represent good predictors.
Additionally, you can check the visualization options available for a particular classifier (e.g. J48 is a decision tree, the attribute used in the root test is for the best predictor). You can check the AttributeSelection tab visualization options as well.