Methods and principles of building "computer systems that automatically improve with experience."

learn more… | top users | synonyms (1)

0
votes
1answer
26 views

How to output as unclassified object in svm multiclass classifier?

I am developing an image classifier using opencv,python.I am using svm from opencv. The image classifier classifies Animals,vehicles and Humans.It works fine.But when i give the image of 'nature ...
0
votes
1answer
17 views

How can I estimate a state-action matrix for q-learning when I do not have complete knowledge of all possible spaces and actions?

In this example of q-learning the "state-action" matrix R can be easily defined since there is a limited number of possible actions in each state and they are easy to identify. This example is very ...
1
vote
2answers
234 views

How to derive this conditional distribution function for a Restricted Boltzmann Machine?

I am following along Ian Goodfellow's new Deep Learning book and, reading the last chapter, I am confused about equations 20.7-20.9. We have a joint distribution function, $P(v,h)$, and we are ...
1
vote
0answers
21 views

AUC (and other measures) dependent on the way data is split

I am applying machine learning (XGBoost) to certain problem regarding time series classification, as input as uses some numerical values around 200 features and vectorized text (tfidf). The result I ...
2
votes
0answers
28 views

Strategy for finding optimal bagging parameters

I am using a BaggingClassifier of SVMs in sklearn. What is the best strategy for finding optimal parameters, using my training/vaildation data? When using the full dataset, I can use grid search to ...
0
votes
0answers
29 views

Training a CNN specifically for feature extraction

I am working on a multiclass multilabel image classification problem. I have been using pre-trained CNNs (from Caffe Model Zoo) to perform feature extraction. I then model the extracted feature ...
0
votes
1answer
31 views

Assigning weights to a multilabel SVM to balance classes

How is this done? I am using Sklearn to train an SVM. My classes are unbalanced. Note that my problem is multiclass, multilabel so I am using OneVsRestClassifier: ...
0
votes
0answers
16 views

Does memory ever really matter for mini-batch size selection?

I'm new to machine learning, and am confused about some aspects of stochastic gradient decent. I've read in several places that, when using vectorized code, the reason that mini-batching in ...
2
votes
1answer
32 views

Curse of dimensionality: why is it a problem that most points are near the edge?

ESL, p. 42 says: Hence most data points are closer to the boundary of the sample space than to any other data point. The reason that this presents a problem is that prediction is much more ...
0
votes
0answers
23 views

Relationship between RMSE and RSS

I'm working on simple linear regression, and I would like to understand the relationship between RMSE and RSS (residual sum of squares). In another Stackexchange question, I found some explanations, ...
0
votes
0answers
8 views

Getting and ordering the data used on a GBM object on R [migrated]

I have a GBM object in R with the underlying data saved on it and I'm having troubles using that data. The problem is that when I run x<-gbmobject$data$x.order ...
0
votes
1answer
22 views

Classification on Imbalanced and Overlapping Data

I am currently working on a project that not only is imbalanced but has a lot of overlap between the data points from the two classes. I am not quite sure how to go about solving this problem. What ...
-1
votes
0answers
18 views

Error derivative of a stochastic binary neurons

How could one backpropagate the error in a net with stochastic binary neurons activated with the probability of the inputs multiplied with the weights then summed and on this sum applied a softstep ...
2
votes
1answer
53 views

Can we gain by merging validation and test set?

Reading this, Cross-validation including training, validation, and testing. Why do we need three subsets? I realized that if we can reduce the variance of the model performance, I wouldn't need the ...
0
votes
0answers
13 views

Adaboost with constant learner weights

Consider a variant on Adaboost where instead of setting the weight $\alpha_t = 1/2\ln(\frac{1 - \epsilon_t}{\epsilon_t})$ we instead set $\alpha$ to some small fixed constant at each iteration. What ...
0
votes
0answers
32 views

Out-of-Sample forecasting with ARIMA

I am trying to forecast crude oil price using Arima. I am using in-sample data to build the Arima model and am using out-of-sample data to test if the model can generalize well to other data. I am ...
0
votes
0answers
12 views

Feature selection with ReliefF algorithm

I have a dataset consisting of around 10000 data points and 20 features. I'm using nested cross-validation for estimating the performance. Now, I want to do feature selection. Due to the nested ...
1
vote
0answers
21 views

How to approach analyzing a dataset of baby speech?

I've been collecting speech data for my baby brother (who is now 6 months old) with the intention of doing computational analysis of the development of his speech patterns. I haven't much deep ...
4
votes
1answer
37 views

Do word vectors obtained via word embedding techniques really form a vector space?

Word embedding refers to feature learning techniques in natural language processing where words are mapped to vectors of real numbers in a low-dimensional space, the embedding space. Similar ...
0
votes
0answers
12 views

How is it correct to optimize a binary classifier output threshold with ROC and LPOCV?

Hello everyone and thank you in advance for you help! I'm building a screening tool with a machine learning algorithm. The model provides a probabilistic prediction (i.e. logistic regression, ...
1
vote
1answer
32 views

How to proceed with building an ensemble classifier using Naive Bayes, TAN and Logistic Regression in R

I'm relatively new to machine learning (started about 5 months ago), and I'm looking at potentially implementing an ensemble classifier as part of my research. I have built 3 models that I use to ...
0
votes
0answers
21 views

How to build a model that maps strings to lists of strings?

I have a mapping from strings to rows in a data table. Each strings maps to exactly one row in a table but the opposite is not generally true (one row can be "bound" to different strings. For example ...
0
votes
0answers
11 views

How can additional data with measurement error be used to improve the predicitve accuracy of a model on data without measurement error?

I have two data sets with 7 categorical variables. The first small dataset (100 samples) is the goldstandard and contains only values that are exactly like they are in the data. The second dataset ...
0
votes
0answers
32 views

Classification of multivariate time series datasets

I have data where each feature is a multivariate time series dataset with a known class label. Each feature is of dimension 4xn and contains per-second measurements of 4 different variables A, B, C ...
0
votes
1answer
17 views

Bayes decision rule and thresholding

The best possible classification is for a set of samples drawn from any probability distribution is given by the Bayes decision rule. For any distribution, the rule is given by $f(x) = 1 ~if ...
-1
votes
1answer
23 views

K-means with learning proces

I have a data set in which I already know the cluster to which each individual belong just by empirical observation but I want to predict, given the characteristics of a new individual in which ...
0
votes
0answers
10 views

What is the meaning of laplace, eps and threshold in NaiveBayes package in R e1071 lib?

I am using NaiveBayes for text classification, I am interested on tagging a text (like a blog post). What I am finding is that normally I have results in which a tag has a probability of 0.9999 of ...
0
votes
0answers
9 views

Image convolution vs signal convolution

What are the similarity and differences between image convolution (e.g., convolution for filtering) and time signal convolution? Apparently I am confused about the formulation of them. Thanks!
0
votes
1answer
25 views

Tikhonov regularization in the context of deconvolution

I came across "Tikhonov regularization" and I have bare knowledge on it. It seems that it is a type of regularization that is important for deconvolution. Are there any good resources and examples? ...
2
votes
0answers
32 views

MCMC efficiency and nonlinear reparametrizations

The efficiency (e.g., effective sample size per density evaluation) of most MCMC methods depends on the parametrization. However, so far I have come across little work in the MCMC literature that ...
0
votes
0answers
24 views

Why is Training Error lower than Testing Error during the First Epoch?

I am training a deep neural network for classification (specifically, a convolutional neural network for object recognition). I use mini-batches for my training, because I cannot fit the entire ...
3
votes
1answer
56 views

Additive bias in xgboost (and its correction?)

I am taking part in a competition right now. I know it is my job to do that well, but maybe somebody wants to discuss my problem and its solution here as this could be helfull for others in their ...
1
vote
0answers
42 views

Confusing about centering variables in regression analysis

Thanks in advance for someone who can help. I realize that, in regression analysis, if there is an interaction term included, it is recommended to center the variables (subtracting variable's ...
0
votes
0answers
16 views

Matrices and vectors as inputs and outputs in linear regression

Suppose I have a matrix with the shape (200 x 40000). Call it features. For every (200 x 100) block, the values capture pixel information about an individual image, ...
0
votes
0answers
27 views

Predicting of revenue. Penalized regression (Ridge regression)

I have data of sales. I've selected one point of sales to check a possibility of predicting revenue using regression method (I don't know what can I use in this task). First of all I've tried to find ...
2
votes
3answers
147 views

What is the difference between Machine Learning and Deep Learning?

OK, I know there is a lot of topic regarding this in the internet, and trust me, I've googled it. But things are getting more and more confused for me. From my understanding, Deep Learning (DL) is ...
0
votes
0answers
13 views

Predicting an arbitrary number of next states given a sequence

I have a problem where I have an input sequence $S_I$ of $N$ states, and I need to predict the next $d$ states in $S_I$. For example, I'm given as input ['A', 'A', 'B', 'B', 'C'], and I then need ...
0
votes
0answers
28 views

How to transform feature with peak at zero to normal distribution?

I have a feature in my dataset which has lot of zero values, i.e. a big peak at zero (the zeroes are valid and valuable information). The histogram is the following: I want to transform all my ...
0
votes
0answers
12 views

R-programming limma design matrices

I have a really basic understanding of how the package limma works in that it fits a linear model to each row/sample in a micro-array dataset. What I do not understand is how to use a design matrix to ...
0
votes
0answers
27 views

Neural Network does polynomials but not xor

my standard vanilla backprop feedforward network models {x} {2x,2x+0.1} (1 input 2 outputs) perfectly, but keeps outputting 0.5 for xor on a 2 input 1 output 2 depth 2 hidden layer network. Any advice ...
2
votes
1answer
34 views

Do neural networks usually take a while to “kick in” during training?

I am trying to train a deep neural network for classification, using back propagation. Specifically, I am using a convolutional neural network for image classification, using the Tensor Flow library. ...
3
votes
0answers
43 views

What is the intuition behind a Long Short Term Memory (LSTM) recurrent neural network?

The idea behind Recurrent Neural Network (RNN) is clear to me. I understand it in the following way. We have a sequence of observations ($\vec o_1, \vec o_2, \dots, \vec o_n$) (or, in other words, ...
1
vote
0answers
26 views

Large sample with many little groups of dependent observations

I work with traffic crash data and my sample consists of about 165,000 injured people distributed over roughly 107,000 crashes. The prevalent approach in traffic crash analysis is to look at every ...
0
votes
0answers
24 views

No change in accuracy big vs small training set size ConvNet

I am doing some small experiments with image classification in Caffe using the AlexNet architecture. I use a dataset of 50 classes with each class containing 1,000 training images. After about 2k ...
1
vote
0answers
32 views

Reporting of Neural Network Accuracy for Academic Publications

I'm an academic researcher, working with Convolutional Neural Networks, particularly for image classification. In academic publications, a typical metric for evaluating the performance of a ...
0
votes
1answer
27 views

Need for removing correlated and near-zero variance features despite feature selection?

I'm doing classification with two classes. Before I apply a classifier, I'm doing some preprocessing steps like removing near-zero variance features or highly correlated features (for those ...
0
votes
0answers
50 views

Standardization with mean/std or median/IQR?

I have a dataset with 10000 data points and 20 features. The features are not normally distributed (most of them have a generalized extreme value or burr distribution and all values are greater or ...
0
votes
0answers
22 views

The feature space from Gaussian kernel is infinite-dimensional, are there countably or uncountably many basis?

My attempt: Let $x,y\in\mathbb{R}^d$. We already know the Fourier transform of a Gaussian function is a Gaussian function.If substituting $x-y$ for the variable after Fourier transform, we have $$ ...
1
vote
0answers
12 views

Training set for Bernoulli Process: retain number of “1” examples in proportion to process?

Given a Bernoulli Process, should my training set have a number of "1" examples in proportion to the process? For example, a Bernoulli Process is "1" 10% of the time and "0" otherwise. In a training ...
0
votes
0answers
20 views

Q-Learning: Should I give the sink states a reward, or a Q-value?

I'm implementing Q-learning via neural network to learn the game of Othello/Reversi. Currently, a win gives a reward of 1, a lose gives -1. However, I've run into a dilemma. I don't know whether I ...