PREMISE:
I'm really new to Computer Vision/Image Processing and Machine Learning (luckily, I'm more expert on Information retrieval), so please be kind with this filthy peasant! :D
MY APPLICATION:
We have a mobile application where the user takes a photo (the query) and the system returns the most similar picture thas was previously taken by some other user (the dataset element). Time performances are crucial, followed by precision and finally by memory usage.
MY APPROACH:
First of all, it's quite obvious that this is a 1-Nearest Neighbor problem (1-NN). LSH is a popular, fast and relatively precise solution for this problem. In particular, my LSH impelementation is about using Kernalized Locality Sensitive Hashing to achieve a good precision to translate a d-dimension vector to a s-dimension binary vector (where s<<d) and then use Fast Exact Search in Hamming Space
with Multi-Index Hashing to quickly find the exact nearest neighbor between all the vectors in the dataset (transposed to hamming space).
In addition, I'm going to use SIFT since I want to use a robust keypoint detector&descriptor for my application.
WHAT DOES IT MISS IN THIS PROCESS?
Well, it seems that I already decided everything, right? Actually NO: in my linked question I face the problem about how to represent the set descriptor vectors of a single image into a vector. Why do I need it? Because a query/dataset element in LSH is vector, not a matrix (while SIFT keypoint descriptor set is a matrix). As someone suggested in the comments, the commonest (and most efficient) solution is using the Bag of Features (BoF) model, which I'm still not confident with yet.
So, I read this article, but I have still some questions (see QUESTIONS below)!
QUESTIONS:
First and most important question: do you think that this is a reasonable approach?
Is k-means used in the BoF algorithm the best choice for such an application? What are alternative clustering algorithms?
The dimension of the codeword vector obtained by the BoF is equal to the number of clusters (so k parameter in the k-means approach)?
If 2. is correct, bigger is k then more precise is the BoF vector obtained?
There is any "dynamic" k-means? Since the query image must added to the dataset after the computation is done (remember: the dataset is formed by the images of all submitted queries) the cluster can change in time.
Given a query image, is the process to obtain the codebook vector the same as the one for a dataset image, e.g. we assign each descriptor to a cluster and the i-th dimension of the resulting vector is equal to the number of descriptors assigned to the i-th cluster?
It looks like you are building codebook from a set of keypoint features generated by SIFT.
You can try "mixture of gaussians" model. K-means assumes that each dimension of a keypoint is independent while "mixture of gaussians" can model the correlation between each dimension of the keypoint feature.
I can't answer this question. But I remember that the SIFT keypoint, by default, has 128 dimensions. You probably want a smaller number of clusters like 50 clusters.
N/A
You can try Infinite Gaussian Mixture Model or look at this paper: "Revisiting k-means: New Algorithms via Bayesian Nonparametrics" by Brian Kulis and Michael Jordan!
Not sure if I understand this question.
Hope this help!
Related
I've read something about Fisher Vector and I'm still in the learning process. It's a better representation than the classic BoF representation, exploiting GMM (or k-means, even if that's usually referred as VLAD).
However, I've seen that usually they are used for classification problem, for example with SVM.
But what about Image Retrieval? I've seen that they have been used for image retrieval too (here), but I don't understand one point: given two FV representing 2 images, how do we compute their distances and so "how similar the two images are?"
Is it reasonable to use them in such a context?
As seen in the two papers below, Euclidean distance seems to be the popular choice. There are also references to using dot-product as a similarity measure; cosine similarity (closely related) is a generally popular metric for ML similarity.
http://link.springer.com/article/10.1007/s11263-013-0636-x
http://www.robots.ox.ac.uk/~vgg/publications/2013/Simonyan13/simonyan13.pdf
Is this enough to let you choose something that meets your needs?
If one trains a model using a SVM from kernel data, the resultant trained model contains support vectors. Now consider the case of training a new model using the old data already present plus a small amount of new data as well.
SO:
Should the new data just be combined with the support vectors from the previously formed model to form the new training set. (If yes, then how to combine the support vectors with new graph data? I am working on libsvm)
Or:
Should the new data and the complete old data be combined together and form the new training set and not just the support vectors?
Which approach is better for retraining, more doable and efficient in terms of accuracy and memory?
You must always retrain considering the entire, newly concatenated, training set.
The support vectors from the "old" model might not be support vectors anymore in case some "new points" are closest to the decision boundary. Behind the SVM there is an optimization problem that must be solved, keep that in mind. With a given training set, you find the optimal solution (i.e. support vectors) for that training set. As soon as the dataset changes, such solution might not be optimal anymore.
The SVM training is nothing more than a maximization problem where the geometrical and functional margins are the objective function. Is like maximizing a given function f(x)...but then you change f(x): by adding/removing points from the training set you have a better/worst understanding of the decision boundary since such decision boundary is known via sampling where the samples are indeed the patterns from your training set.
I understand your concerned about time and memory efficiency, but that's a common problem: indeed training the SVMs for the so-called big data is still an open research topic (there are some hints regarding backpropagation training) because such optimization problem (and the heuristic regarding which Lagrange Multipliers should be pairwise optimized) are not easy to parallelize/distribute on several workers.
LibSVM uses the well-known Sequential Minimal Optimization algorithm for training the SVM: here you can find John Pratt's article regarding the SMO algorithm, if you need further information regarding the optimization problem behind the SVM.
Idea 1 has been already examined & assessed by research community
anyone interested in faster and smarter aproach (1) -- re-use support-vectors and add new data -- kindly review research materials published by Dave MUSICANT and Olvi MANGASARIAN on such their method referred as "Active Support Vector Machine"
MATLAB implementation: available from http://research.cs.wisc.edu/dmi/asvm/
PDF:[1] O. L. Mangasarian, David R. Musicant; Active Support Vector Machine Classification; 1999
[2] David R. Musicant, Alexander Feinberg; Active Set Support Vector Regression; IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 15, NO. 2, MARCH 2004
This is a purely theoretical thought on your question. The idea is not bad. However, it needs to be extended a bit. I'm looking here purely at the goal to sparsen the training data from the first batch.
The main problem -- which is why this is purely theoretical -- is that your data is typically not linear separable. Then the misclassified points are very important. And they will spoil what I write below. Furthermore the idea requires a linear kernel. However, it might be possible to generalise to other kernels
To understand the problem with your approach lets look at the following support vectors (x,y,class): (-1,1,+),(-1,-1,+),(1,0,-). The hyperplane is the a vertical line going trough zero. If you would have in your next batch the point (-1,-1.1,-) the max margin hyperplane would tilt. This could now be exploited for sparsening. You calculate the - so to say - minimal margin hyperplane between the two pairs ({(-1,1,+),(1,0,-)}, {(-1,-1,+),(1,0,-)}) of support vectors (in 2d only 2 pairs. higher dimensions or non-linear kernel might be more). This is basically the line going through these points. Afterwards you classify all data points. Then you add all misclassified points in either of the models, plus the support vectors to the second batch. Thats it. The remaining points can't be relevant.
Besides the C/Nu problem mentioned above. The curse of dimensionality will obviously kill you here
An image to illustrate. Red: support vectors, batch one, Blue, non-support vector batch one. Green new point batch two.
Redline first Hyperplane, Green minimal margin hyperplane which misclassifies blue point, blue new hyperplane (it's a hand fit ;) )
I got Memory Error when I was running dbscan algorithm of scikit.
My data is about 20000*10000, it's a binary matrix.
(Maybe it's not suitable to use DBSCAN with such a matrix. I'm a beginner of machine learning. I just want to find a cluster method which don't need an initial cluster number)
Anyway I found sparse matrix and feature extraction of scikit.
http://scikit-learn.org/dev/modules/feature_extraction.html
http://docs.scipy.org/doc/scipy/reference/sparse.html
But I still have no idea how to use it. In DBSCAN's specification, there is no indication about using sparse matrix. Is it not allowed?
If anyone knows how to use sparse matrix in DBSCAN, please tell me.
Or you can tell me a more suitable cluster method.
The scikit implementation of DBSCAN is, unfortunately, very naive. It needs to be rewritten to take indexing (ball trees etc.) into account.
As of now, it will apparently insist of computing a complete distance matrix, which wastes a lot of memory.
May I suggest that you just reimplement DBSCAN yourself. It's fairly easy, there exists good pseudocode e.g. on Wikipedia and in the original publication. It should be just a few lines, and you can then easily take benefit of your data representation. E.g. if you already have a similarity graph in a sparse representation, it's usually fairly trivial to do a "range query" (i.e. use only the edges that satisfy your distance threshold)
Here is a issue in scikit-learn github where they talk about improving the implementation. A user reports his version using the ball-tree is 50x faster (which doesn't surprise me, I've seen similar speedups with indexes before - it will likely become more pronounced when further increasing the data set size).
Update: the DBSCAN version in scikit-learn has received substantial improvements since this answer was written.
You can pass a distance matrix to DBSCAN, so assuming X is your sample matrix, the following should work:
from sklearn.metrics.pairwise import euclidean_distances
D = euclidean_distances(X, X)
db = DBSCAN(metric="precomputed").fit(D)
However, the matrix D will be even larger than X: n_samples² entries. With sparse matrices, k-means is probably the best option.
(DBSCAN may seem attractive because it doesn't need a pre-determined number of clusters, but it trades that for two parameters that you have to tune. It's mostly applicable in settings where the samples are points in space and you know how close you want those points to be to be in the same cluster, or when you have a black box distance metric that scikit-learn doesn't support.)
Yes, since version 0.16.1.
Here's a commit for a test:
https://github.com/scikit-learn/scikit-learn/commit/494b8e574337e510bcb6fd0c941e390371ef1879
Sklearn's DBSCAN algorithm doesn't take sparse arrays. However, KMeans and Spectral clustering do, you can try these. More on sklearns clustering methods: http://scikit-learn.org/stable/modules/clustering.html
I have some problems with understanding the kernels for non-linear SVM.
First what I understood by non-linear SVM is: using kernels the input is transformed to a very high dimension space where the transformed input can be separated by a linear hyper-plane.
Kernel for e.g: RBF:
K(x_i, x_j) = exp(-||x_i - x_j||^2/(2*sigma^2));
where x_i and x_j are two inputs. here we need to change the sigma to adapt to our problem.
(1) Say if my input dimension is d, what will be the dimension of the
transformed space?
(2) If the transformed space has a dimension of more than 10000 is it
effective to use a linear SVM there to separate the inputs?
Well it is not only a matter of increasing the dimension. That's the general mechanism but not the whole idea, if it were true that the only goal of the kernel mapping is to increase the dimension, one could conclude that all kernels functions are equivalent and they are not.
The way how the mapping is made would make possible a linear separation in the new space.
Talking about your example and just to extend a bit what greeness said, RBF kernel would order the feature space in terms of hyperspheres where an input vector would need to be close to an existing sphere in order to produce an activation.
So to answer directly your questions:
1) Note that you don't work on feature space directly. Instead, the optimization problem is solved using the inner product of the vectors in the feature space, so computationally you won't increase the dimension of the vectors.
2) It would depend on the nature of your data, having a high dimensional pattern would somehow help you to prevent overfitting but not necessarily will be linearly separable. Again, the linear separability in the new space would be achieved because the way the map is made and not only because it is in a higher dimension. In that sense, RBF would help but keep in mind that it might not perform well on generalization if your data is not locally enclosed.
The transformation usually increases the number of dimensions of your data, not necessarily very high. It depends. The RBF Kernel is one of the most popular kernel functions. It adds a "bump" around each data point. The corresponding feature space is a Hilbert space of infinite dimensions.
It's hard to tell if a transformation into 10000 dimensions is effective or not for classification without knowing the specific background of your data. However, choosing a good mapping (encoding prior knowledge + getting right complexity of function class) for your problem improves results.
For example, the MNIST database of handwritten digits contains 60K training examples and 10K test examples with 28x28 binary images.
Linear SVM has ~8.5% test error.
Polynomial SVM has ~ 1% test error.
Your question is a very natural one that almost everyone who's learned about kernel methods has asked some variant of. However, I wouldn't try to understand what's going on with a non-linear kernel in terms of the implied feature space in which the linear hyperplane is operating, because most non-trivial kernels have feature spaces that it is very difficult to visualise.
Instead, focus on understanding the kernel trick, and think of the kernels as introducing a particular form of non-linear decision boundary in input space. Because of the kernel trick, and some fairly daunting maths if you're not familiar with it, any kernel function satisfying certain properties can be viewed as operating in some feature space, but the mapping into that space is never performed. You can read the following (fairly) accessible tutorial if you're interested: from zero to Reproducing Kernel Hilbert Spaces in twelve pages or less.
Also note that because of the formulation in terms of slack variables, the hyperplane does not have to separate points exactly: there's an objective function that's being maximised which contains penalties for misclassifying instances, but some misclassification can be tolerated if the margin of the resulting classifier on most instances is better. Basically, we're optimising a classification rule according to some criteria of:
how big the margin is
the error on the training set
and the SVM formulation allows us to solve this efficiently. Whether one kernel or another is better is very application-dependent (for example, text classification and other language processing problems routinely show best performance with a linear kernel, probably due to the extreme dimensionality of the input data). There's no real substitute for trying a bunch out and seeing which one works best (and make sure the SVM hyperparameters are set properly---this talk by one of the LibSVM authors has the gory details).
I have a collection of documents related to a particular domain and have trained the centroid classifier based on that collection. What I want to do is, I will be feeding the classifier with documents from different domains and want to determine how much they are relevant to the trained domain. I can use the cosine similarity for this to get a numerical value but my question is what is the best way to determine the threshold value?
For this, I can download several documents from different domains and inspect their similarity scores to determine the threshold value. But is this the way to go, does it sound statistically good? What are the other approaches for this?
Actually there is another issue with centroids in sparse vectors. The problem is that they usually are significantly less sparse than the original data. For examples, this increases computation costs. And it can yield vectors that are themselves actually atypical because they have a different sparsity pattern. This effect is similar to using arithmetic means of discrete data: say the mean number of doors in a car is 3.4; yet obviously no car exists that actually has 3.4 doors. So in particular, there will be no car with an euclidean distance of less than 0.4 to the centroid! - so how "central" is the centroid then really?
Sometimes it helps to use medoids instead of centroids, because they actually are proper objects of your data set.
Make sure you control such effects on your data!
A simple method to try would be to employ various machine-learning algorithms - and in particular, tree-based ones - on the distances from your centroids.
As mentioned in another answer(#Anony-Mousse), this won't necessarily provide you with good or usable answers, but it just might. Using a ML framework for this procedure, E.g. WEKA, will also help you with estimating your accuracy in a more rigorous manner.
Here are the steps to take, using WEKA:
Generate a train set by finding a decent amount of documents representing each of your classes (to get valid estimations, I'd recommend at least a few dozens per class)
Calculate the distance from each document to each of your centroids.
Generate a feature vector for each such document, composed of the distances from this document to the centroids. You can either use a single feature - the distance to the nearest centroid; or use all distances, if you'd like to try a more elaborate thresholding scheme. For example, if you chose the simpler method of using a single feature, the vector representing a document with a distance of 0.2 to the nearest centroid, belonging to class A would be: "0.2,A"
Save this set in ARFF or CSV format, load into WEKA, and try classifying, e.g. using a J48 tree.
The results would provide you with an overall accuracy estimation, with a detailed confusion matrix, and - of course - with a specific model, e.g. a tree, you can use for classifying additional documents.
These results can be used to iteratively improve the models and thresholds by collecting additional train documents for problematic classes, either by recreating the centroids or by retraining the thresholds classifier.