Hot answers tagged

52

Let's denote the true value of interest as $\theta$ and the value estimated using some algorithm as $\hat{\theta}$. Correlation tells you how much $\theta$ and $\hat{\theta}$ are related. It gives values between $-1$ and $1$, where $0$ is no relation, $1$ is very strong, linear relation and $-1$ is an inverse linear relation (i.e. bigger values of $\theta$ ...


8

Let me explain what odds mean in general. Odds are the ratio between the probability of success over the probability of failure, that is, $\displaystyle \frac{p_{i}}{1-p_{i}}$. Let's say $p_{i}$ for a given event is 0.6, then the odds for that event is $0.6/0.4=1.5$. 1- As you said, since the logistic regression outputs probabilities based on the ...


8

First of all, as @Marc Claesen already explained, semi-supervised classification is one of the techniques to take care of the situation where you know that the classes are really distinct, but you are not certain which class the case actually belongs to. However, there are related situations as well, where the the "reality" isn't that clear, and the ...


8

Feature selection we consider a subset of attributes which has the greatest impact towards our targeted classification. This understanding is perfectly correct. PCA we generate a smaller amount of artificial set of attributes that will account for our target. This is partially correct. We are not accounting target in PCA. In layman terms, we do ...


7

I cannot quickly see what would generate your error without having a sample data set. But if you run the following code you can see that the RWeka tokonizer works. library(tm) library(RWeka) data("crude") crude <- as.VCorpus(crude) crude <- tm_map(crude, stripWhitespace) crude <- tm_map(crude, content_transformer(tolower)) crude <-...


6

This is one of the generalizations of classification that are tackled in semi-supervised learning. If you have a measurement of certainty you can use approaches that allow weighting of training instances. The higher the certainty, the larger the corresponding instance weight. Examples of such approaches include instance-weighted SVM and logistic regression. ...


6

Related answer: https://stats.stackexchange.com/a/66420/25433 If we choose to use these toolkits, besides that it is convenient, sometimes we might lose control of what is really happening. This is not at all specific to open source toolkits. In fact, in open source programs it's easier to make changes where and when necessary. Additionally, if you add ...


6

OK, here is the method for tokenizing grams in quanteda. Our view is that there is no such thing as n-grams without tokenization, since the notion implies sequences of tokens defined by some kind of adjacency. So we built in an ngrams option into our tokenize() function. ngrams takes a vector of integers, where each integer represents a size of ngram (the ...


5

Start with J4.8 since it is fastest to train and generally gives good results. Also its output is human readable therefore you can see if it makes sense. It has tree visualizers to aid understanding. It is among most used data mining algorithms. If J4.8 does not give you good enough solutions , try other algorithms. Random forests may give you better ...


5

You can use the Adjusted Rand Index or the Adjusted Mutual Information to measure the similarity (agreement) of the overall results of two clustering algorithms on an overlapping dataset. Both scores are adjusted for chance which means that 2 random clusterings will likely have an ARI or AMI close to 0.0. Furthermore you can use those measure for model ...


5

I do not use Weka, but I will try to explain how things works, and I hope you will find the way to do that in Weka. Transform the unsupervised into a supervised problem So RF knows only supervised learning. However, in order to do unsupervised learning you have to set up your problem as a supervised learning one. In order to transform you problem into a ...


4

RapidAnalytics, a cousin of the RapidMiner project, does exactly that. You model your process in RapidMiner, using Weka or native RapidMiner operators, and then publish it to the RapidAnalytics server where it is available as a web service, returning XML, HTML or JSON to you. It's impressive open source technology with a great community. http://rapid-i....


4

It sounds like you want to use a method such a LMT (logistic model trees) or perhaps Joao Gama 's functional trees. The former uses standard decision tree splitting criteria at the interior nodes and additive logistic regression functions at the leaves, and the tree is pruned using the CART pruning method. The latter also can have logistic regression ...


4

The "Class Correlation" is Pearson Correlation Coefficient between target variable and the other variables. i.e. corr(Species, sepal length) = 0.7826 Here is the R output (i do not have Weka handy): > iris2 <- data.frame(iris$Sepal.Length, iris$Sepal.Width, iris$Petal.Length, iris$Petal.Width, as.numeric(iris$Species)) > head(iris2) iris.Sepal....


4

I like to determine it based on a hypothesis. Take a look at the dataset, then keep in mind the hypothesis you have of the data.


4

This is a regression problem, meaning that you are trying to approximate a function, as opposed to a classification problem in which you would be trying to reproduce a discrete category. I think the first step should be to use something simple like linear regression. Did you try that, and if so, what was unsatisfactory with the results?


4

The method you link to in your comment should work, if you choose to follow the neural-network survival analysis approach in the article I linked to in my comment. For each patient in the model that approach uses a list of probabilities of being alive at each time of interest: 1/0 for patients known to have died, and for "censored" cases a 1 until last ...


4

Sequential Minimal Optimization (SMO) is one way to solve the SVM training problem that is more efficient than standard QP solvers. SMO uses heuristics to partition the training problem into smaller problems that can be solved analytically. Whether or not it works well depends largely on the assumptions behind the heuristics (working set selection). ...


4

The problem is that the distribution of classes in your training dataset is dramatically different than the distribution of your classes in your testing dataset. In your training dataset if you were to always predict class B, then you would have about a 50 percent accuracy; however, in your testing dataset notice that class B is very under represented as ...


4

It appears that you have a lot of background reading to do. First, it is not usually appropriate to develop classifiers because this involves using an arbitrary loss/utility/cost function. It is usually better to develop a probability estimator (this is what logistic regression does). For details see http://www.fharrell.com/2017/01/classification-vs-...


4

Read the first RF article for implementation of RF in the R-project ("R-News"). I commonly use 500 to 5000 trees, but Breiman recommend using more in his original Machine Learning paper (he said: "Don't be stingy"). I have written the following for my own codes: Number of features used for training at each node split, jtry. A unique characteristic of RF ...


3

I only complete two mentioned measures with third that is also widely used in this situation, Normalized Variation of Information, which was proposed in the last 10 years. It is also scaled to the unit interval. In addition, I recommend that you use any ensemble clustering technique that can assign to more partitions, so called consensus partitions, where ...


3

k-fold cross classification is about estimating the accuracy, not improving the accuracy. Increasing the k can improve the accuracy of the measurement of your accuracy (yes, think Inception), but it does not actually improve the original accuracy you are trying to measure. Most implementations of k-fold cross validation give you an estimate of how ...


3

An easy way to build an ensemble is by using a random forest. I'm fairly sure weka has a random forest algorithm, and if other tree-based models are performing well it's worth trying out. You could also build your own ensemble by training multiple (say 50 or 100) J48 decision trees and using them to "vote" on the classification of each object. For example,...


3

In the weka explorer, under the classify tab. Once you have chosen the J48 classifier and have clicked the start button, the classifier output displays the confusion matrix. Just under the start button there is the result list, right click the most recent classifier and look for the visualise tree option. Note that if things do not display well, you can ...


3

The methods appear to be very similar in performance and that's perhaps the main story. But I don't think we can say much to help you decide. Which is more important practically, making the right classification or reducing error? It can easily happen with mean error measures that even one odd observation pushes them up a fair bit. I don't think that ...


3

Ridge is a regularization technique. In simple words, adding ridge to Logistic Regression means that you want to make a model that is not overfitting with the training data and hopefully generalizes well in the test data. It is done by adding penalty on weights learned. This way, the learner will try to find the weights that are close to zero and not the ...


3

I would suggest a different approach. Instead of sweeping across all possible classifiers, stop and think about your problem. How does your feature space look like? For the case of binary classification, are there two large clusters with some boundary, or is your feature space "segmented" and contains many clusters? In the former case, an SVM would be a ...


3

Your description of the confusion matrix is correct assuming alive people are defined as a positive outcome. Those entries are the correct order. TP | FN FP | TN I do not like how Weka labels the columns. TP Rate (for example) is based on that row being the positive. So the second entry under TP Rate (0.626) is actually the TN Rate. The other columns ...


Only top voted, non community-wiki answers of a minimum length are eligible