Estimated P&I (Subject To Lock Of Interest Rate By Lender: Fill & Download for Free

You remember the story of the pandavas. Well you got an estimate. :P

A candidate who really impresses me would be knowledgeable in:Statistics: Confidence intervals, parameter estimation, p-value, hypothesis testing.Common metrics: Engagement / retention rate, conversion, similar products / duplicates matching, how to measure them.Useful cost functions: Log-loss, other entopy-based, DCG/NDCG, etc.Basic machine learning: Classification / regression / ranking problems, overfitting, convex optimization, trees, ensembles, boosting, collaborative filtering, etc.Tools: R / Python / Mathematica, Weka & similar. Code up something yourself would help too, Kaggle is very useful.Mathematics and complexities: Eigenvectors, singular values, PCA, LDA, Gibbs Sampling, Information Bottleneck et. al.Real-life numbers and intuition: Expected user behavior, reasonable ranges for user signup / retention rate, session length / count, registered / unregistered users, deep / top-level engagement, spam rate, complaint rate, ads efficiency.

A generative model describes how data is generated, in terms of a probabilistic model.In the scenario of supervised learning, a generative model estimates the joint probability distribution of data P(X, Y) between the observed data X and corresponding labels Y [1]. Another requirement of a generative model, which is less frequently stated, is that it provides a way to sample X, Y pairs [2].Examples of popular generative models are:Naive BayesHidden Markov ModelsLatent Dirichlet AllocationBoltzmann MachinesYou can use a generative model to perform prediction.[math] argmax_y P(Y=y|X=x) = [/math][math] argmax_y P(Y=y, X=x) / P(X=x)[/math]and since P(X=x) is constant on the RHS, this equals[math] argmax_y P(Y=y, X=x). [/math]One of the most important things to understand about generative models is that they are capable of more than just prediction, i.e. maximizing P(Y|X=x). By estimating P(Y, X) and being able to sample X, Y pairs - a generative model can be used to impute missing data, compress your dataset or generate unseen data.If your primary goal is prediction then discriminative models, which directly estimate P(Y|X), are found to be empirically superior because they attack the problem directly. However, you gain little understanding about the data from discriminative models. A further discussion on the comparison between generative and discriminative models can be found here: What are some benefits and drawbacks of discriminative and generative models? and a good textbook reference on the subject can be found here: Chapter 3: Generative and Discriminative Classifiers.[1] This is often accomplished by modeling P(X|Y) and P(Y) separately, for example Naive Bayes does this.[2] I think this is an important point in order to understand why we don't consider Kernel Density Estimation a generative model, despite the fact it can be used to estimate P(X,Y).