I am currently working with some Video data so rather than use a standard feed-forward model, I want to use a Long short-term memory (LSTM) architecture such that the context/memory persists in the reasoning. I looked at the Flux.jl docs but they don't say much about how to actually use the LSTM function beyond the input variables such as: LSTM(in::Integer, out::Integer).
How can I make use of the LSTM architecture with flux.jl? What do the in and out integers represent in this case?
Related
My goal is to deploy a Mask RCNN model trained with the well known Matterport's repo with Nvidia deepstream.
To do so, first I have to convert the generated .h5 model into a .uff. This operation is decribed here.
After the conversion, I have run the generated .uff model with TensoRT and deepstream and it has a very poor performance compared to the .h5model (almost never detects/masks the objects).
Before the conversion, I have done the corresponding changes to handle NCWH models and configured the number of classes and backbone (in this case resnet50).
I don't know how to continue. Any advice could really healp me. Thanks!
To solve the problem one must use the same configuration for the training and the conversion.
In particular, since most of models start from tranfering learning from the pretrained coco model, one has to use its very same config.
In adition, the input images sizes have to be coherent with the trainning configuration.
I'm using the Caret package from R to create prediction models for maximum energy demand. What i need to use is neural network multilayer perceptron, but in the Caret package i found out there's 2 of the mlp method, which is "mlp" and "mlpML". what is the difference between the two?
I have read description from a book (Advanced R Statistical Programming and Data Models: Analysis, Machine Learning, and Visualization) but it still doesnt answer my question.
Caret has 238 different models available! However many of them are just different methods to call the same basic algorithm.
Besides mlp there are 9 other methods of calling a multi-layer-perceptron one of which is mlpML. The real difference is only in the parameters of the function call and which model you need depends on your use case and what you want to adapt about the basic model.
Chances are, if you don't know what mlpML or mlpWeightDecay,etc. does you are fine to just use the basic mlp.
Looking at the official documentation we can see that:
mlp(size) while mlpML(layer1,layer2,layer3) so in the first method you can only tune the size of the multi-layer-perceptron while in the second call you can tune each layer individually.
Looking at the source code here:
https://github.com/topepo/caret/blob/master/models/files/mlp.R
and here:
https://github.com/topepo/caret/blob/master/models/files/mlpML.R
It seems that the difference is that mlpML allows several hidden layers:
modelInfo <- list(label = "Multi-Layer Perceptron, with multiple layers",
while mlp has one single layer with hidden units.
The official documentation also hints at this difference. In my opinion, it is not particularly useful to have many different models that differ only very slightly, and the documentation does not explain those slight differences well.
I have a model where some of the input features are calculated from the training dataset (e.g. average or median of a value). I am trying to perform n-fold cross validation on this model, but that means that the values for these features would be different depending on the samples selected for training/validation for each fold. Is there a way in h2o (I'm using it in R) to perhaps pass a funtion that calculates those features once the training set has been determined?
It seems like a pretty intuitive feature to have, but I have not been able to find any documentation on something like it out-of-the-box. Does it exist? If so, could someone point me to a resource?
There's no way to do this while using the built-in cross-validation in H2O. If H2O were written in pure R or Python, then it would be easy to extend it to allow a user to pass in a function to create custom features within the cross-validation loop, however the core of H2O is written in Java, so automatically translating an arbitrary user-defined function from R or Python, first into a REST call and then into Java is not trivial.
Instead, what you'd have to do is write a loop to do the cross-validation yourself and compute the features within the loop.
It sounds like you may be doing target encoding (or something similar), and if that's the case, you'll be interested in this PR to add target encoding in H2O. In the discussion, we talk about the same issue that you're having.
There exist a very large own-collected dataset of size [2000000 12672] where the rows shows the number of instances and the columns, the number of features. This dataset occupies ~60 Gigabyte on the local hard disk. I want to train a linear SVM on this dataset. The problem is that I have only 8 Gigabyte of RAM! so I cannot load all data once. Is there any solution to train the SVM on this large dataset? Generating the dataset is on my own desire, and currently are is HDF5 format.
Thanks
Welcome to machine learning! One of the hard things about working in this space is the compute requirements. There are two main kinds of algorithms, on-line and off-line.
Online: supports feeding in examples one at a time, each one improving the model slightly
Offline: supports feeding in the entire dataset at once, achieving higher accuracy than an On-line model
Many typical algorithms have both on-line, and off-line implementations, but an SVM is not one of them. To the best of my knowledge, SVMs are traditionally an off-line only algorithm. The reason for this is a lot of the fine details around "shattering" the dataset. I won't go too far into the math here, but if you read into it it should become apparent.
It's also worth noting that the complexity of an SVM is somewhere between n^2 and n^3, meaning that even if you could load everything into memory it would take ages to actually train the model. It's very typical to test with a much smaller portion of your dataset before moving to the full dataset.
When moving to the full dataset you would have to run this on a much larger machine than your own, but AWS should have something large enough for you, though at your size of data I highly advise using something other than an SVM. At large data sizes, neural net approaches really shine, and can be trained in a more realistic amount of time.
As alluded to in the comments, there's also the concept of an out-of-core algorithm that can operate directly on objects stored on disk. The only group I know with a good offering of out-of-core algorithms is dato. It's a commercial product, but might be your best solution here.
A stochastic gradient descent approach to SVM could help, as it scales well and avoids the n^2 problem. An implementation available in R is RSofia, which was created by a team at Google and is discussed in Large Scale Learning to Rank. In the paper, they show that compared to a traditional SVM, the SGD approach significantly decreases the training time (this is due to 1, the pairwise learning method and 2, only a subset of the observations end up being used to train the model).
Note that RSofia is a little more bare bones than some of the other SVM packages available in R; for example, you need to do your own centering and scaling of features.
As to your memory problem, it'd be a little surprising if you needed the entire dataset - I would expect that you'd be fine reading in a sample of your data and then training your model on that. To confirm this, you could train multiple models on different samples and then estimate performance on the same holdout set - the performance should be similar across the different models.
You don't say why you want Linear SVM, but if you can consider another model that often gives superior results then check out the hpelm python package. It can read an HDF5 file directly. You can find it here https://pypi.python.org/pypi/hpelm It trains on segmented data, that can even be pre-loaded (called async) to speed up reading from slow hard disks.
I am trying to build a basic Emotion detector from speech using MFCCs, their deltas and delta-deltas. A number of papers talk about getting a good accuracy by training GMMs on these features.
I cannot seem to find a ready made package to do the same. I did play around with scilearn in Python, Voicebox and similar toolkits in Matlab and Rmixmod, stochmod, mclust, mixtools and some other packages in R. What would be the best library to calculate GMMs from trained data?
Challenging problem is training data, which contains the emotion information, embedded in feature set. The same features that encapsulate emotions should be used in the test signal. The testing with GMM will only be good as your universal background model. In my experience typically with GMM you can only separate male female and a few unique speakers. Simply feeding the MFCC’s into GMM would not be sufficient, since GMM does not hold time varying information. Since emotional speech would contain time varying parameters such as pitch and changes in pitch over periods in addition to the frequency variations MFCC parameters. I am not saying it not possible with current state of technology but challenging in a good way.
If you want to use Python, here is the code in the famous speech recognition toolkit Sphinx.
http://sourceforge.net/p/cmusphinx/code/HEAD/tree/trunk/sphinxtrain/python/cmusphinx/gmm.py