I am working on a Machine Learning Classification Model in which the user can provide label instances that should help improve the model.
More relevance needs to be given to the latest instances given by the user than for those instances that were previously available for training.
In particular, I am developing my machine learning models in python using Sklearn libraries.
So far I've only found the strategy of oversampling particular instances as a possible solution to the problem. With this strategy I would create multiple copies of the instances for which I want to give higher relevance.
Other strategy that I've found, but it seems not help under these conditions is:
Strategies that focus on giving weights for each class. This strategy is highly used in multiple libraries like Sklearn by default. However, this generalizes the idea to a class level and doesn't help me to put focus on particular instances
I've look for multiple strategies that might help provide specific weights for individual instances but most have focused on class level instead of instance level weights.
I read some suggestions to multiple the loss function by some factors for instances in tensor flow models, but this seems to be mostly applicable to neural network models in Tensor flow.
I wonder if anyone has information of other approaches that might helps with this problem
I've look for multiple strategies that might help provide specific weights for individual instances but most have focused on class level instead of instance level weights.
This is not accurate; most scikit-learn classifiers provide a sample_weight argument in their fit methods, which does exactly that. For example, here is the documentation reference for Logistic Regression:
sample_weight : array-like, shape (n_samples,) optional
Array of weights that are assigned to individual samples. If not provided, then each sample is given unit weight.
Similar arguments exist for most scikit-learn classifiers, e.g. decision trees, random forests etc, even for linear regression (not a classifier). Be sure to check the SVM: Weighted samples example in the docs.
The situation is roughly similar for other frameworks; see for example own answer in Is there in PySpark a parameter equivalent to scikit-learn's sample_weight?
What's more, scikit-learn also provides a utility function to compute sample_weight in cases of imbalanced datasets: sklearn.utils.class_weight.compute_sample_weight
Related
I am currently trying to use satellite imagery to recognize Apples orchards. And I am facing a small problem in the number of representative data for each class.
In fact my question is :
Is it possible to take randomly some different images in my "not-apples" class at each epoch because I have much more of theses (compared to the "apples" one) and I want to increase the probability my network will classify out an image unrepresentative.
Thanks in advance for your help
That is not possible in Keras. Keras will, by default, shuffle your training data and then train on it in a mini-batch fashion. However, there are still ways to re-balance your dataset.
The imbalanced training data problem that you are facing is pretty common. You have many options available to you; I list a few below:
You can adjust the relative weights of your classes using class_weight keyword of the model.fit() function.
You can "up-sample" your "apples" class or "down-sample" your "non-apples" class to have equal numbers of both classes during training.
You can generate synthetic images of your "apples" class to augment your data set. To this end, the ImageDataGenerator class in Keras can be particularly useful. This Keras tutorial is a good introduction to its usage.
In my experience, I've found #2 and #3 to be most useful. #1 is limited by the fact that the convergence of stochastic gradient descent suffers when using class weights differing by a couple orders of magnitude and smaller batch sizes.
Jason Brownlee has put together a list of tactics for dealing with imbalanced classes that might also be useful to you.
I'm using a multiclass classifier (a Support Vector Machine, via One-Vs-All) to classify data samples. Let's say I currently have n distinct classes.
However, in the scenario I'm facing, it is possible that a new data sample may belong to a new class n+1 that hasn't been seen before.
So I guess you can say that I need a form of Online Learning, as there is no distinct training set in the beginning that suits all data appearing later. Instead I need the SVM to adapt dynamically to new classes that may appear in the future.
So I'm wondering about if and how I can...
identify that a new data sample does not quite fit into the existing classes but instead should result in creating a new class.
integrate that new class into the existing classifier.
I can vaguely think of a few ideas that might be approaches to solve this problem:
If none of the binary SVM classifiers (as I have one for each class in the OVA case) predicts a fairly high probability (e.g. > 0.5) for the new data sample, I could assume that this new data sample may represent a new class.
I could train a new binary classifier for that new class and add it to the multiclass SVM.
However, these are just my naive thoughts. I'm wondering if there is some "proper" approach for this instead, e.g. using a Clustering algorithms to find all classes.
Or maybe my approach of trying to use an SVM for this is not even appropriate for this kind of problem?
Help on this is greatly appreciated.
As in any other machine learning problem, if you do not have a quality criterion, you suck.
When people say "classification", they have supervised learning in mind: there is some ground truth against which you can train and check your algorithms. If new classes can appear, this ground truth is ambiguous. Imagine one class is "horse", and you see many horses: black horses, brown horses, even white ones. And suddenly you see a zebra. Whoa! Is it a new class or just an unusual horse? The answer will depend on how you are going to use your class labels. The SVM itself cannot decide, because SVM does not use these labels, it only produces them. The decision is up to a human (or to some decision-making algorithm which knows what is "good" and "bad", that is, has its own "loss function" or "utility function").
So you need a supervisor. But how can you assist this supervisor? Two options come to mind:
Anomaly detection. This can help you with early occurences of new classes. After the very first zebra your algorithm sees it can raise an alarm: "There is something unusual!". For example, in sklearn various algorithms from random forest to one-class SVM can be used to detect unusial observations. Then your supervisor can look at them and decide whether they deserve to form an entirely new class.
Clustering. It can help you to make decision about splitting your classes. For example, after the first zebra, you decided it is not worth making a new class. But over time, your algorithm has accumulated dozens of their images. So if you run a clustering algorithm on all the observations labeled as "horses", you might end up with two well-separated clusters. And it will be again up to the supervisor to decide, whether the striped horses should be detached from the plain ones into a new class.
If you want this decision to be purely authomatic, you can split classes if the ratio of within-cluster mean distance to between-cluster distance is low enough. But it will work well only if you have a good distance metric in the first place. And what is "good" is again defined by how you use your algorithms and what your ultimate goal is.
I am a deep-learning newbie and working on creating a vehicle classifier for images using Caffe and have a 3-part question:
Are there any best practices in organizing classes for training a
CNN? i.e. number of classes and number of samples for each class?
For example, would I be better off this way:
(a) Vehicles - Car-Sedans/Car-Hatchback/Car-SUV/Truck-18-wheeler/.... (note this could mean several thousand classes), or
(b) have a higher level
model that classifies between car/truck/2-wheeler and so on...
and if car type then query the Car Model to get the car type
(sedan/hatchback etc)
How many training images per class is a typical best practice? I know there are several other variables that affect the accuracy of
the CNN, but what rough number is good to shoot for in each class?
Should it be a function of the number of classes in the model? For
example, if I have many classes in my model, should I provide more
samples per class?
How do we ensure we are not overfitting to class? Is there way to measure heterogeneity in training samples for a class?
Thanks in advance.
Well, the first choice that you mentioned corresponds to a very challenging task in computer vision community: fine-grained image classification, where you want to classify the subordinates of a base class, say Car! To get more info on this, you may see this paper.
According to the literature on image classification, classifying the high-level classes such as car/trucks would be much simpler for CNNs to learn since there may exist more discriminative features. I suggest to follow the second approach, that is classifying all types of cars vs. truck and so on.
Number of training samples is mainly proportional to the number of parameters, that is if you want to train a shallow model, much less samples are required. That also depends on your decision to fine-tune a pre-trained model or train a network from scratch. When sufficient samples are not available, you have to fine-tune a model on your task.
Wrestling with over-fitting has been always a problematic issue in machine learning and even CNNs are not free of them. Within the literature, some practical suggestions have been introduced to reduce the occurrence of over-fitting such as dropout layers and data-augmentation procedures.
May not included in your questions, but it seems that you should follow the fine-tuning procedure, that is initializing the network with pre-computed weights of a model on another task (say ILSVRC 201X) and adapt the weights according to your new task. This procedure is known as transfer learning (and sometimes domain adaptation) in community.
I've a Regression model that is most suitably solved using elastic net.
It has a very large number of predictors that I need to select only subset of them. Moreover, there could be correlation between the predictors, so Elastic net was the choice)
My question is:
If I have knowledge that a specific subset of the predictors must exist in the output (they shouldn't be penalized), how can this information be added to the elastic net?
Or even to the Regression model if elastic net is suitable in this case.
I need advise about papers that propose such solutions if possible.
I'm using Scikit-learn in Python, but I'm concerned more about the algorithm more than just how to do it.
If you're using the glmnet package in R, the penalty.factor argument addresses this.
From ?glmnet:
penalty.factor
Separate penalty factors can be applied to each coefficient. This is a number that multiplies lambda to allow differential shrinkage. Can be 0 for some variables, which implies no shrinkage, and that variable is always included in the model. Default is 1 for all variables (and implicitly infinity for variables listed in exclude). Note: the penalty factors are internally rescaled to sum to nvars, and the lambda sequence will reflect this change.
It depends on the kind of knowledge that you have. Regularization is a kind of adding prior knowledge to your model. For example Ridge regression encodes the knowledge that your coefficients should be small. Lasso regression encodes the knowledge that not all predictors are important. Elastic net is a more complicated prior that combine both of the assumptions in your model. There are other regularizers that you may check for example if you know that your predictors are grouped in certain groups you may check grouped Lasso. Also, if they interact in certain way (maybe some predictors are correlated with each other). You may also check Bayesian regression if you need more control over your prior.
I've been studying neural networks for a bit and recently learned about the dropout training algorithm. There are excellent papers out there to understand how it works, including the ones from the authors.
So I built a neural network with dropout training (it was fairly easy) but I'm a bit confused about how to perform model selection. From what I understand, looks like dropout is a method to be used when training the final model obtained through model selection.
As for the test part, papers always talk about using the complete network with halved weights, but they do not mention how to use it in the training/validation part (at least the ones I read).
I was thinking about using the network without dropout for the model selection part. Say that makes me find that the net performs well with N neurons. Then, for the final training (the one I use to train the network for the test part) I use 2N neurons with dropout probability p=0.5. That assures me to have exactly N neurons active on average, thus using the network at the right capacity most of the time.
Is this a correct approach?
By the way, I'm aware of the fact that dropout might not be the best choice with small datasets. The project I'm working on has academic purposes, so it's not really needed that I use the best model for the data, as long as I stick with machine learning good practices.
First of all, model selection and the training of a particular model are completely different issues. For model selection, you would usually need a data set that is completely independent of both training set used to build the model and test set used to estimate its performance. So if you're doing for example a cross-validation, you would need an inner cross-validation (to train the models and estimate the performance in general) and an outer cross-validation to do the model selection.
To see why, consider the following thought experiment (shamelessly stolen from this paper). You have a model that makes a completely random prediction. It has a number of parameters that you can set, but have no effect. If you're trying different parameter settings long enough, you'll eventually get a model that has a better performance than all the others simply because you're sampling from a random distribution. If you're using the same data for all of these models, this is the model you will choose. If you have a separate test set, it will quickly tell you that there is no real effect because the performance of this parameter setting that achieves good results during the model-building phase is not better on the separate set.
Now, back to neural networks with dropout. You didn't refer to any particular paper; I'm assuming that you mean Srivastava et. al. "Dropout: A Simple Way to Prevent Neural Networks from Overfitting". I'm not an expert on the subject, but the method to me seems to be similar to what's used in random forests or bagging to mitigate the flaws an individual learner may exhibit by applying it repeatedly in slightly different contexts. If I understood the method correctly, essentially what you end up with is an average over several possible models, very similar to random forests.
This is a way to make an individual model better, but not for model selection. The dropout is a way of adjusting the learned weights for a single neural network model.
To do model selection on this, you would need to train and test neural networks with different parameters and then evaluate those on completely different sets of data, as described in the paper I've referenced above.