The extraction procedure is called sample points corresponding sample from a specific probability distribution.
This chapter mathematics involved ones, the more boring, only individuals "resampling uneven sample set of 07" focus on the last one looked under
1 sampling effect
Is a simulation of random sampling on the nature of the phenomenon, according to a given probability distribution, to simulate a random event corresponding.
2 uniformly distributed random numbers
Q1: How to program a uniformly distributed random number generator?
A1: may generally be linear congruential method for generating a pseudo-random number uniformly distributed discrete, calculated as
Xt+1 ≡ a · Xt + c (mod m)
3 common sampling method
In a sense, the probability density function of the application of reverse sampling.
Q1: sampling method set aside for a particular distribution of well-designed, say some common sampling methods or sampling strategies you know, their main ideas and specific steps are simple description.
A1: Almost all sampling methods are uniformly distributed random number as the basic operation. Generally uniformly distributed random number used to generate the linear congruential method. . . . P176-179
4 sample Gaussian distribution
5 Markov Monte Carlo sampling method
Q1: Description of the main ideas MCMC sampling method.
A1: from the name See, MCMC sampling includes two MC, i.e. Monte Carlo method and the Markov chain. Monte Carlo sampling means based on numerical approximation solution method, while the Markov chain for sampling. The basic idea is MCMC sampling method: target distribution to be sampled, construct a Markov chain, so that the stationary distribution of the Markov chain is the target profile; then, starting from any initial state, along the Markov chain state transition, the state transition sequence of the finally obtained converge to the target profile, thereby to obtain a series of samples of target distribution.
Q2: brief introduction to several common MCMC sampling method.
A2: Metropolis-Hastings sampling, Gibbs sampling P186-187
Q3: How MCMC sampling method to obtain independent samples?
A3: Unlike general Monte Carlo algorithm, the resulting sequence of samples of adjacent sampling MCMC samples are not independent, since it is after a sample obtained from the previous sample according to the particular transition probability, or probability of a certain that is, before a sample. Independent of each other only if the sample, the sample is not required. If necessary to generate independent and identically distributed samples can be run simultaneously multiple Markov chain, so that samples in different chains are separated; or several samples before selecting it in the same chain on a hidden Markov intervals, so selected out of the samples is approximately independent.
6 Bayesian network sampling
7 uneven sample set resampling
Q1: For binary classification, when the training set of positive and negative samples very uneven, how to process the data to better train the classification model?
A1: Why do so many classification model problems occur when the training data is not balanced? The reason is the nature of the evaluation criteria model optimization objective function and in training people to use when testing inconsistent. This "inconsistency" may be due to inconsistent expected when training data sample distribution and the distribution of the test sample.
The method based on the data, a method based on two methods detailed algorithm P195-196