I am going to work on a problem that needs to be addressed with either RNN or Deep Neural Nets. In general, the problem is predicting financial values. So, because I am given the sequence of financial data as an input, I thought that RNN would be better. On the other hand, I think that if I can fit the data into some structure, I can train with DNN much better because the training phase is easier in DNN than RNN. For example, I could get last 1-month info and keep 30 inputs and predict 31'th day while using DNN.
I don't understand the advantage of RNN over DNN in this perspective. My first question is about the proper usage of RNN or DNN in this problem.
My second questions are somehow basic. While training RNN, isn't it possible for a network to get "confused"? I mean, consider the following input: 10101111, and our inputs are one digits 0 or 1 and we have 2-sequences (1-0,1-0,1-1,1-1) Hereafter 1, comes 0 several times. And then at the end, after 1 comes 1. While training, wouldn't this become a major problem? That is, why the system not gets confused while training this sequence?
I think your question is phrased a bit problematically. First, DNNs are a class of architectures. A Convolutional Neural Network differs greatly from a Deep Belief Network or a simple Deep MLP. There are feed forward architectures (e.g. TDNN) fit for timeseries prediction but it depends on you, whether you're more interested in research or just solving your problem.
Second, RNNs are as "deep" as it gets. Considering the most basic RNN, the Elman Network: During training with Backpropagation through time (BPTT) they are unfolded in time - backpropagating over T timesteps. Since this backpropagation is done not only vertically like in a standard DNN but also horizontally over T-1 context layers, the past activations of the hidden layers from T-1 timesteps before the present are actually considered for the activation at the current timestep. This illustration of an unfolded net might help in understanding what I just wrote (source):
This makes RNNs so powerful for timeseries prediction (and should answer both your questions). If you have more questions, read about Elman Networks. LSTMs etc. will only confuse you. Understanding Elman Networks and BPTT is the needed foundation to understand any other RNN.
And one last thing you'll need to look out for: The vanishing gradient problem. While it's tempting to say let's make T=infinity and give our RNN as much memory as possible: It doesn't work. There are many ways working around this problem, LSTMs are quite popular at the moment and there are even some proper LSTM implementations around nowadays. But it's important to know that a basic Elman Network could really struggle with T=30.
As you answered yourself - RNN are for sequences. If data has sequential nature (time series) than it is preferable to use such model over DNN and other "static" models. The main reason is that RNN can model process which is responsible for each conequence, so for example given sequences
0011100
0111000
0001110
RNN will be able to build a model, that "after seeing '1' I will see two more" and correctly build a prediction when seeing
0000001**** -> 0000001110
While in the same time, for DNN (and other non sequential models) there is no relation between these three sequences, in fact the only common thing for them is that "there is 1 on forth position, so I guess it is always like that".
Regarding the second question. Why it won't get confused? Because it models sequences, because it has memory. It makes its recisions based on everything that was observed before, and assuming that your signal has any type of regularity, there is always some vent in the past that differentiate between two possible paths of signals. Once again, such phenomena are much better addressed by RNN than non-recurrent models. See for example natural language and enormous progress given by LSTM-based models in recent years.
Related
I must participate in a research project regarding a deep learning application for classification. I have a huge dataset containing over 35000 features - these are good values, taken from laboratory.
The idea is that I should create a classifier that must tell, given a new input, if the data seems to be good or not. I must use deep learning with keras and tensor flow.
The problem is that the data is not classified. I will enter a new column with 1 for good and 0 for bad. Problem is, how can I find out if an entry is bad, given the fact that the whole training set is good?
I have thought about generating some garbage data but I don't know if this is a good idea - I don't even know how to generate it. Do you have any tips?
I would start with anamoly detection. You can first reduce features with f.e. an (stacked) autoencoder and then use local outlier factor from sklearn: https://scikit-learn.org/stable/modules/outlier_detection.html
The reason why you need to reduce features first is, is because your LOF will be much more stable.
I am implementing a simple neural net from scratch, just for practice. I have got it working fine with sigmoid, tanh and ReLU activations for binary classification problems. I am now attempting to use it for multi-class, mutually exclusive problems. Of course, softmax is the best option for this.
Unfortunately, I have had a lot of trouble understanding how to implement softmax, cross-entropy loss and their derivatives in backprop. Even after asking a couple of questions here and on Cross Validated, I can't get any good guidance.
Before I try to go further with implementing softmax, is it possible to somehow use sigmoid for multi-class problems (I am trying to predict 1 of n characters, which are encoded as one-hot vectors)? And if so, which loss function would be best? I have been using the squared error for all binary classifications.
Your question is about the fundamentals of neural networks and therefore I strongly suggest you start here ( Michael Nielsen's book ).
It is python-oriented book with graphical, textual and formulated explanations - great for beginners. I am confident that you will find this book useful for your understanding. Look for chapters 2 and 3 to address your problems.
Addressing your question about the Sigmoids, it is possible to use it for multiclass predictions, but not recommended. Consider the following facts.
Sigmoids are activation functions of the form 1/(1+exp(-z)) where z is the scalar multiplication of the previous hidden layer (or inputs) and a row of the weights matrix, in addition to a bias (reminder: z=w_i . x + b where w_i is the i-th row of the weight matrix ). This activation is independent of the others rows of the matrix.
Classification tasks are regarding categories. Without any prior knowledge ,and even with, most of the times, categories have no order-value interpretation; predicting apple instead of orange is no worse than predicting banana instead of nuts. Therefore, one-hot encoding for categories usually performs better than predicting a category number using a single activation function.
To recap, we want an output layer with number of neurons equals to number of categories, and sigmoids are independent of each other, given the previous layer values. We also would like to predict the most probable category, which implies that we want the activations of the output layer to have a meaning of probability disribution. But Sigmoids are not guaranteed to sum to 1, while softmax activation does.
Using L2-loss function is also problematic due to vanishing gradients issue. Shortly, the derivative of the loss is (sigmoid(z)-y) . sigmoid'(z) (error times the derivative), that makes this quantity small, even more when the sigmoid is closed to saturation. You can choose cross entropy instead, or a log-loss.
EDIT:
Corrected phrasing about ordering the categories. To clarify, classification is a general term for many tasks related to what we used today as categorical predictions for definite finite sets of values. As of today, using softmax in deep models to predict these categories in a general "dog/cat/horse" classifier, one-hot-encoding and cross entropy is a very common practice. It is reasonable to use that if the aforementioned is correct. However, there are (many) cases it doesn't apply. For instance, when trying to balance the data. For some tasks, e.g. semantic segmentation tasks, categories can have ordering/distance between them (or their embeddings) with meaning. So please, choose wisely the tools for your applications, understanding what their doing mathematically and what their implications are.
What you ask is a very broad question.
As far as I know, when the class become 2, the softmax function will be the same as sigmoid, so yes they are related. Cross entropy maybe the best loss function.
For the backpropgation, it is not easy to find the formula...there
are many ways.Since the help of CUDA, I don't think it is necessary to spend much time on it if you just want to use the NN or CNN in the future. Maybe try some framework like Tensorflow or Keras(highly recommand for beginers) will help you.
There is also many other factors like methods of gradient descent, the setting of hyper parameters...
Like I said, the topic is very abroad. Why not trying the machine learning/deep learning courses on Coursera or Stanford online course?
Many of the papers I have read so far have this mentioned "pre-training network could improve computational efficiency in terms of back-propagating errors", and could be achieved using RBMs or Autoencoders.
If I have understood correctly, AutoEncoders work by learning the
identity function, and if it has hidden units less than the size of
input data, then it also does compression, BUT what does this even have
anything to do with improving computational efficiency in propagating
error signal backwards? Is it because the weights of the pre
trained hidden units does not diverge much from its initial values?
Assuming data scientists who are reading this would by theirselves
know already that AutoEncoders take inputs as target values since
they are learning identity function, which is regarded as
unsupervised learning, but can such method be applied to
Convolutional Neural Networks for which the first hidden layer is
feature map? Each feature map is created by convolving a learned
kernel with a receptive field in the image. This learned kernel, how
could this be obtained by pre-training (unsupervised fashion)?
One thing to note is that autoencoders try to learn the non-trivial identify function, not the identify function itself. Otherwise they wouldn't have been useful at all. Well the pre-training helps moving the weight vectors towards a good starting point on the error surface. Then the backpropagation algorithm, which is basically doing gradient descent, is used improve upon those weights. Note that gradient descent gets stuck in the closes local minima.
[Ignore the term Global Minima in the image posted and think of it as another, better, local minima]
Intuitively speaking, suppose you are looking for an optimal path to get from origin A to destination B. Having a map with no routes shown on it (the errors you obtain at the last layer of the neural network model) kind of tells you where to to go. But you may put yourself in a route which has a lot of obstacles, up hills and down hills. Then suppose someone tells you about a route a a direction he has gone through before (the pre-training) and hands you a new map (the pre=training phase's starting point).
This could be an intuitive reason on why starting with random weights and immediately start to optimize the model with backpropagation may not necessarily help you achieve the performance you obtain with a pre-trained model. However, note that many models achieving state-of-the-art results do not use pre-training necessarily and they may use the backpropagation in combination with other optimization methods (e.g. adagrad, RMSProp, Momentum and ...) to hopefully avoid getting stuck in a bad local minima.
Here's the source for the second image.
I don't know a lot about autoencoder theory, but I've done a bit of work with RBMs. What RBMs do is they predict what the probability is of seeing the specific type of data in order to get the weights initialized to the right ball park- it is considered an (unsupervised) probabilistic model, so you don't correct using the known labels. Basically, the idea here is that having a learning rate that is too big will never lead to convergence but having one that is too small will take forever to train. Thus, by "pretraining" in this way you find out the ball park of the weights and then can set the learning rate to be small in order to get them down to the optimal values.
As for the second question, no, you don't generally prelearn kernels, at least not in an unsupervised fashion. I suspect that what is meant by pretraining here is a bit different than in your first question- this is to say, that what is happening is that they are taking a pretrained model (say from model zoo) and fine tuning it with a new set of data.
Which model you use generally depends on the type of data you have and the task at hand. Convnets I've found to train faster and efficiently, but not all data has meaning when convolved, in which case dbns may be the way to go. Unless say, you have a small amount of data then I'd use something other than neural networks entirely.
Anyways, I hope this helps clear some of your questions.
One of the most popular questions regarding Neural Networks seem to be:
Help!! My Neural Network is not converging!!
See here, here, here, here and here.
So after eliminating any error in implementation of the network, What are the most common things one should try??
I know that the things to try would vary widely depending on network architecture.
But tweaking which parameters (learning rate, momentum, initial weights, etc) and implementing what new features (windowed momentum?) were you able to overcome some similar problems while building your own neural net?
Please give answers which are language agnostic if possible. This question is intended to give some pointers to people stuck with neural nets which are not converging..
If you are using ReLU activations, you may have a "dying ReLU" problem. In short, under certain conditions, any neuron with a ReLU activation can be subject to a (bias) adjustment that leads to it never being activated ever again. It can be fixed with a "Leaky ReLU" activation, well explained in that article.
For example, I produced a simple MLP (3-layer) network with ReLU output which failed. I provided data it could not possibly fail on, and it still failed. I turned the learning rate way down, and it failed more slowly. It always converged to predicting each class with equal probability. It was all fixed by using a Leaky ReLU instead of standard ReLU.
If we are talking about classification tasks, then you should shuffle examples before training your net. I mean, don't feed your net with thousands examples of class #1, after thousands examples of class #2, etc... If you do that, your net most probably wouldn't converge, but would tend to predict last trained class.
I had faced this problem while implementing my own back prop neural network. I tried the following:
Implemented momentum (and kept the value at 0.5)
Kept the learning rate at 0.1
Charted the error, weights, input as well as output of each and every neuron, Seeing the data as a graph is more helpful in figuring out what is going wrong
Tried out different activation function (all sigmoid). But this did not help me much.
Initialized all weights to random values between -0.5 and 0.5 (My network's output was in the range -1 and 1)
I did not try this but Gradient Checking can be helpful as well
If the problem is only convergence (not the actual "well trained network", which is way to broad problem for SO) then the only thing that can be the problem once the code is ok is the training method parameters. If one use naive backpropagation, then these parameters are learning rate and momentum. Nothing else matters, as for any initialization, and any architecture, correctly implemented neural network should converge for a good choice of these two parameters (in fact, for momentum=0 it should converge to some solution too, for a small enough learning rate).
In particular - there is a good heuristic approach called "resillient backprop" which is in fact parameterless appraoch, which should (almost) always converge (assuming correct implementation).
after you've tried different meta parameters (optimization / architecture), the most probable place to look at is - THE DATA
as for myself - to minimize fiddling with meta parameters, i keep my optimizer automated - Adam is by opt-of-choice.
there are some rules of thumb regarding application vs architecture... but its really best to crunch those on your own.
to the point:
in my experience, after you've debugged the net (the easy debugging), and still don't converge or get to an undesired local minima, the usual suspect is the data.
weather you have contradictory samples or just incorrect ones (outliers), a small amount can make the difference from say 0.6-acc to (after cleaning) 0.9-acc..
a smaller but golden (clean) dataset is much better than a big slightly dirty one...
with augmentation you can tweak results even further.
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 4 years ago.
Improve this question
Suppose I'm working on some classification problem. (Fraud detection and comment spam are two problems I'm working on right now, but I'm curious about any classification task in general.)
How do I know which classifier I should use?
Decision tree
SVM
Bayesian
Neural network
K-nearest neighbors
Q-learning
Genetic algorithm
Markov decision processes
Convolutional neural networks
Linear regression or logistic regression
Boosting, bagging, ensambling
Random hill climbing or simulated annealing
...
In which cases is one of these the "natural" first choice, and what are the principles for choosing that one?
Examples of the type of answers I'm looking for (from Manning et al.'s Introduction to Information Retrieval book):
a. If your data is labeled, but you only have a limited amount, you should use a classifier with high bias (for example, Naive Bayes).
I'm guessing this is because a higher-bias classifier will have lower variance, which is good because of the small amount of data.
b. If you have a ton of data, then the classifier doesn't really matter so much, so you should probably just choose a classifier with good scalability.
What are other guidelines? Even answers like "if you'll have to explain your model to some upper management person, then maybe you should use a decision tree, since the decision rules are fairly transparent" are good. I care less about implementation/library issues, though.
Also, for a somewhat separate question, besides standard Bayesian classifiers, are there 'standard state-of-the-art' methods for comment spam detection (as opposed to email spam)?
First of all, you need to identify your problem. It depends upon what kind of data you have and what your desired task is.
If you are Predicting Category :
You have Labeled Data
You need to follow Classification Approach and its algorithms
You don't have Labeled Data
You need to go for Clustering Approach
If you are Predicting Quantity :
You need to go for Regression Approach
Otherwise
You can go for Dimensionality Reduction Approach
There are different algorithms within each approach mentioned above. The choice of a particular algorithm depends upon the size of the dataset.
Source: http://scikit-learn.org/stable/tutorial/machine_learning_map/
Model selection using cross validation may be what you need.
Cross validation
What you do is simply to split your dataset into k non-overlapping subsets (folds), train a model using k-1 folds and predict its performance using the fold you left out. This you do for each possible combination of folds (first leave 1st fold out, then 2nd, ... , then kth, and train with the remaining folds). After finishing, you estimate the mean performance of all folds (maybe also the variance/standard deviation of the performance).
How to choose the parameter k depends on the time you have. Usual values for k are 3, 5, 10 or even N, where N is the size of your data (that's the same as leave-one-out cross validation). I prefer 5 or 10.
Model selection
Let's say you have 5 methods (ANN, SVM, KNN, etc) and 10 parameter combinations for each method (depending on the method). You simply have to run cross validation for each method and parameter combination (5 * 10 = 50) and select the best model, method and parameters. Then you re-train with the best method and parameters on all your data and you have your final model.
There are some more things to say. If, for example, you use a lot of methods and parameter combinations for each, it's very likely you will overfit. In cases like these, you have to use nested cross validation.
Nested cross validation
In nested cross validation, you perform cross validation on the model selection algorithm.
Again, you first split your data into k folds. After each step, you choose k-1 as your training data and the remaining one as your test data. Then you run model selection (the procedure I explained above) for each possible combination of those k folds. After finishing this, you will have k models, one for each combination of folds. After that, you test each model with the remaining test data and choose the best one. Again, after having the last model you train a new one with the same method and parameters on all the data you have. That's your final model.
Of course, there are many variations of these methods and other things I didn't mention. If you need more information about these look for some publications about these topics.
The book "OpenCV" has a great two pages on this on pages 462-463. Searching the Amazon preview for the word "discriminative" (probably google books also) will let you see the pages in question. These two pages are the greatest gem I have found in this book.
In short:
Boosting - often effective when a large amount of training data is available.
Random trees - often very effective and can also perform regression.
K-nearest neighbors - simplest thing you can do, often effective but slow and requires lots of memory.
Neural networks - Slow to train but very fast to run, still optimal performer for letter recognition.
SVM - Among the best with limited data, but losing against boosting or random trees only when large data sets are available.
Things you might consider in choosing which algorithm to use would include:
Do you need to train incrementally (as opposed to batched)?
If you need to update your classifier with new data frequently (or you have tons of data), you'll probably want to use Bayesian. Neural nets and SVM need to work on the training data in one go.
Is your data composed of categorical only, or numeric only, or both?
I think Bayesian works best with categorical/binomial data. Decision trees can't predict numerical values.
Does you or your audience need to understand how the classifier works?
Use Bayesian or decision trees, since these can be easily explained to most people. Neural networks and SVM are "black boxes" in the sense that you can't really see how they are classifying data.
How much classification speed do you need?
SVM's are fast when it comes to classifying since they only need to determine which side of the "line" your data is on. Decision trees can be slow especially when they're complex (e.g. lots of branches).
Complexity.
Neural nets and SVMs can handle complex non-linear classification.
As Prof Andrew Ng often states: always begin by implementing a rough, dirty algorithm, and then iteratively refine it.
For classification, Naive Bayes is a good starter, as it has good performances, is highly scalable and can adapt to almost any kind of classification task. Also 1NN (K-Nearest Neighbours with only 1 neighbour) is a no-hassle best fit algorithm (because the data will be the model, and thus you don't have to care about the dimensionality fit of your decision boundary), the only issue is the computation cost (quadratic because you need to compute the distance matrix, so it may not be a good fit for high dimensional data).
Another good starter algorithm is the Random Forests (composed of decision trees), this is highly scalable to any number of dimensions and has generally quite acceptable performances. Then finally, there are genetic algorithms, which scale admirably well to any dimension and any data with minimal knowledge of the data itself, with the most minimal and simplest implementation being the microbial genetic algorithm (only one line of C code! by Inman Harvey in 1996), and one of the most complex being CMA-ES and MOGA/e-MOEA.
And remember that, often, you can't really know what will work best on your data before you try the algorithms for real.
As a side-note, if you want a theoretical framework to test your hypothesis and algorithms theoretical performances for a given problem, you can use the PAC (Probably approximately correct) learning framework (beware: it's very abstract and complex!), but to summary, the gist of PAC learning says that you should use the less complex, but complex enough (complexity being the maximum dimensionality that the algo can fit) algorithm that can fit your data. In other words, use the Occam's razor.
Sam Roweis used to say that you should try naive Bayes, logistic regression, k-nearest neighbour and Fisher's linear discriminant before anything else.
My take on it is that you always run the basic classifiers first to get some sense of your data. More often than not (in my experience at least) they've been good enough.
So, if you have supervised data, train a Naive Bayes classifier. If you have unsupervised data, you can try k-means clustering.
Another resource is one of the lecture videos of the series of videos Stanford Machine Learning, which I watched a while back. In video 4 or 5, I think, the lecturer discusses some generally accepted conventions when training classifiers, advantages/tradeoffs, etc.
You should always keep into account the inference vs prediction trade-off.
If you want to understand the complex relationship that is occurring in your data then you should go with a rich inference algorithm (e.g. linear regression or lasso). On the other hand, if you are only interested in the result you can go with high dimensional and more complex (but less interpretable) algorithms, like neural networks.
Selection of Algorithm is depending upon the scenario and the type and size of data set.
There are many other factors.
This is a brief cheat sheet for basic machine learning.