20

There are a couple of nuances here. Complexity question very important - ocams razor CV - is this trully the case 84%/83% (test it for train+test with CV) Given this, personal opinion: Second one. Better to catch general patterns. You already know that first model failed on that because of the train and test difference. 1% says nothing.


13

It depends mostly on the problem context. If predictive performance is all you care about, and you believe the test set to be representative of future unseen data, then the first model is better. (This might be the case for, say, health predictions.) There are a number of things that would change this decision. Interpretability / explainability. This is ...


12

I'd suggest you to try a hybrid approach: First, train your car in supervised fashion by demonstration. Just control it and use your commands as labels. This will let you get all the pros of SL. Then, fine tune your neural net using reinforcement learning. You don't need extra sensors for that: the rewards may be obtained from distance sensors (larger ...


9

Yes this is possible by treating the audio as a sequence into a Recurrent Neural Network (RNN). You can train a RNN against a target that is correct at the end of a sequence, or even to predict another sequence offset from the input. Do note however that there is a bit to learn about options that go into the construction and training of a RNN, that you will ...


9

Is this equivalent of pruning a decision tree? Though they have similar goals (i.e. placing some restrictions to the model so that it doesn't grow very complex and overfit), max_depth isn't equivalent to pruning. The way pruning usually works is that go back through the tree and replace branches that do not help with leaf nodes. If not, how could I ...


8

SVM can be used for classification (distinguishing between several groups or classes) and regression (obtaining a mathematical model to predict something). They can be applied to both linear and non linear problems. Until 2006 they were the best general purpose algorithm for machine learning. I was trying to find a paper that compared many implementations ...


8

Upsampling layer is used to increase the resolution of the image. In segmentation, we first downsample the image to get the features and then upsample the image to generate the segments. For deconvolution operation we pad the image with zeroes and then do a convolution operation on that, hence it is upsampled. For eg: - If after downsampling the images ...


7

I faced almost exactly the same scenario a year and a half ago -- basically what you have is a variation of the one-class classification (OCC) problem, specifically PU-learning (learning from Positive and Unlabelled data). You have your known, labelled positive dataset (clients) and an un-labelled dataset of prospects (some of which are client-like and some ...


7

A good rule of thumb is to look at the level of measurement of the target/response variable. If the response is measured on a nominal scale, the problem is a classification problem. Values on a nominal scale are for example labels of a categories where the categories have no natural order, like political parties in political science, species in biology, or ...


7

... someone pointed out that neural networks do not work very well with the structured data (data in tabular format) as compared to the unstructured data (like representing each pixel in an image). It's difficult to propose a universal analogy, but perhaps one moderately complex example which is easy to understand will suffice. In the link you provided ...


7

The first has an accuracy of 100% on training set and 84% on test set. Clearly over-fitted. Maybe not. It's true that 100% training accuracy is usually a strong indicator of overfitting, but it's also true that an overfit model should perform worse on the test set than a model that isn't overfit. So if you're seeing these numbers, something unusual is going ...


6

This seems like a job for Principal Component Analysis. In Scikit is PCA implemented well and it helped me many times. PCA, in a certain way, combines your features. By limiting the number of components, you fetch your model with noise-less data (in the best case). Because your model is as good as your data are. Consider below a simple example. from ...


6

I think I can answer that, since I implement such a thing in my own library, even if I really don't know how it's implemented in other libraries. Although I am confident that if there are other ways, they don't differ too much. It took my a few weeks to understand how such a graph can be drawn. Let's start with a general function $f:\mathbb{R} \times \...


6

Linear, binary classifiers can choose either class (but consistently) when the datapoint which is to classify is on the hyperplane. It just depends on how you programmed it. Also, it doesn't really matter. This is very unlikely to happen. In fact, if we had arbitrary precision computing and normal distributed features, there would be a probability of 0 (...


6

Imagine that your data is not easily separable. Your classifier isn't able to do a very good job at distinguishing between positive and negative examples, so it usually predicts the majority class for any example. In the unbalanced case, it will get 100 examples correct and 20 wrong, resulting in a 100/120 = 83% accuracy. But after balancing the classes, the ...


6

Try this: Make the data stationary (remove trends and seasonality). Implement PACF analysis on the label data (For eg: Load) and find out the optimal lag value. Usually, you need to know how to interpret PACF plots. Apply the sliding window on the whole data (t+o, t-o) where o is the optimal lag value. Apply walk forward validation to train and test the ...


6

Good question and welcome to Datascience Imagine you have the tree as follows. Machine Learning Models | ---------------------------------------------------- | | Supervised Unsupervised |...


5

What does this sentence mean? It means that the next vector should be perpendicular to all the previous ones with respect to a matrix. It's like how the natural basis vectors are perpendicular to each other, with the added twist of a matrix: $\mathrm {x^T A y} = 0$ instead of $\mathrm{x^T y} = 0$ And what is line search mentioned in the webpage? Line ...


5

'Training' data is really just splitting data you have already collected into test or training sets. For example, if you want to build a classifier for handwritten numbers, you collect thousands of samples of handwritten numbers like the MNIST database. When you think you have enough data to build a model, you then split it into train and test sets (usually ...


5

Accuracy is probably not a good metric for your problem. For the original dataset, if the model just makes a dummy prediction that all samples belong to the bigger class, the accuracy will be 83% (100/120). But that's usually not what we want to predict in an imbalanced dataset. Let's take a fraud detection problem. The probability that a transaction is a ...


5

Looks like to me this is a classic imbalance binary classification problem (see comments above). What loss are you using ? It looks like your model is predicting the non-membership class because it’s minimising it’s averaging loss. Here are some techniques you might wanna try to solve this issue: use regularization over sampling the membership class under ...


5

Depends. The first thing that has to be clear is that you are running an experiment, which means you need to measure both with the same metric. Which one? Depends on which underlying problem you are solving, if what you are doing is to determine which algorithm is better, your conclusion will only be applicable to your specific dataset Accuracy: Is ...


5

The tradeoff between bias and variance summarizes the "tug of war" game between fitting a model that predicts the underlying training dataset well (low bias) and producing a model that doesn't change much with the training dataset (low variance). What statisticians/mathematicians a while ago realized is that any model can be made to perfectly fit the ...


5

Audio .wav codec file has a 44 byte header which will give you critical data like bit depth ( CD quality audio is 16 bits per sample), sample rate ( CD quality uses 44,100 audio samples per second ), number of channels, etc ... the balance of bytes in a .wav file is the payload which is the audio curve stored as a set of integers which define the height of ...


4

The best way to combine features is through ensemble methods. Basically there are three different methods: bagging, boosting and stacking. You can either use Adabbost augmented with feature selection (in this consider both sparse and dense features) or stacking based (random feature - random subspace) I prefer the second option you can train a set of base ...


4

Data typically exists. What typically does not exist, is ground truth (in the case of classification). Such ground truth is typically always collected manually and crowd sourcing plays an important role. For example, think about Face recognition that Facebook does. Before automatic tagging was available, Facebook allowed users to manually add tags to ...


4

The main difference between supervised and unsupervised learning is the following: In supervised learning you have a set of labelled data, meaning that you have the values of the inputs and the outputs. What you try to achieve with machine learning is to find the true relationship between them, what we usually call the model in math. There are many ...


4

Semi-Supervised Learning The combination of unsupervised learning and supervised learning is referred to as semi-supervised learning, which is the concept that I believe you are searching for. Label propagation is often cited when outlining the heuristics of semi-supervised learning. The essence is to employ clustering, but to use a tiny set of known ...


4

Likelihood-ratio tests are a mainstay of classical hypothesis testing. The idea is to form the likelihoods of the two hypotheses under consideration, and choose the one with the highest likelihood if their ratio is sufficiently large. Hypotheses come in two flavors: simple, and composite. Simple tests are those for which the hypothesis uniquely defines the ...


4

I think what you're talking about is called compound features, and it's extremely important because sometimes it captures interactions between certain features which are not readily apparent in considering each column vector (or column) independently. The answer to your question is, it may or may not, there's really no way to know without looking at the ...


Only top voted, non community-wiki answers of a minimum length are eligible