A Fine Slice Of SQL Server
Evgeni Chasnovski has a new CRAN package:
I am glad to announce that my latest, long written R package ‘pdqr’ is accepted to CRAN. It provides tools for creating, transforming and summarizing custom random variables with distribution functions (as base R ‘p*()’, ‘d*()’, ‘q*()’, and ‘r*()’ functions). You can read a brief overview in one of my previous posts.
Click through for a description of the package.
Comments closedSebastian Sauer takes us through two of the most important assumptions of linear models:
Additivity and linearity as the second most important assumptions in linear models
We assume that \(y\) is a linear function of the predictors. If y is not a linear function of the predictors, we cannot expect the model to deliver correct insights (predictions, causal coefficients). Let’s check an example.
Read on to understand what this means, as well as the most important assumption.
Comments closedThe R-hub blog has an interesting post on creating mocks in R for unit testing:
In some of these cases, the programming concept you’re after is mocking, i.e. making a function act as if something were a certain way! In this blog post we shall offer a round-up of resources around mocking, or not mocking, when unit testing an R package.
It’s interesting watching data scientists work through the same sorts of problems which traditional developers have hit, whether that be testing, deployment, or source control management. H/T R-bloggers
Comments closedJohn Mount has a new introduction to rquery:
rqueryis a data wrangling system designed to express complex data manipulation as a series of simple data transforms. This is in the spirit ofR’sbase::transform(), ordplyr’sdplyr::mutate()and uses a pipe in the style popularized inRwithmagrittr. The operators themselves follow the selections in Codd’s relational algebra, with the addition of the traditionalSQL“window functions.” More on the background and context ofrquerycan be found here.The
R/rqueryversion of this introduction is here, and thePython/data_algebraversion of this introduction is here.
Check it out.
Comments closedJohn Mount shares an extract from Mount and Nina Zumel’s Practical Data Science with R, 2nd Edition:
This section reflects an important design decision in the book: teach model evaluation first, and as a step separate from model construction.
It is funny, but it takes some effort to teach in this way. New data scientists want to dive into the details of model construction first, and statisticians are used to getting model diagnostics as a side-effect of model fitting. However, to compare different modeling approaches one really needs good model evaluation that is independent of the model construction techniques.
Click through for that extract. I liked the first edition of the book, so I’m looking forward to the 2nd.
Comments closedBeth Ebersole takes us through creating training, validation, and test data sets using SAS Viya:
Training data are used to fit each model. Training a model involves using an algorithm to determine model parameters (e.g., weights) or other logic to map inputs (independent variables) to a target (dependent variable). Model fitting can also include input variable (feature) selection. Models are trained by minimizing an error function.
For illustration purposes, let’s say we have a very simple ordinary least squares regression model with one input (independent variable, x) and one output (dependent variable, y). Perhaps our input variable is how many hours of training a dog or cat has received, and the output variable is the combined total of how many fingers or limbs we will lose in a single encounter with the animal.
Read on for some good notes, including the difference between mean squared error and average squared error.
Comments closedBoth results show that evaluating two tests on the same family of data will lead to a ~10% chance that a researcher will claim a “significant” result if they look for either test to reject the null. Any claim there is a maximum 5% false positive rate would be mistaken. As an exercise, verify that doing the same on \(m=4\) tests will lead to an ~18% chance!
A bad testing platform would be one that claims a maximum 5% false positive rate when any one of multiple tests on the same family of data show significance at the 5% level. Clearly, if a researcher is going to claim that the FWER is no more than \(\alpha\), then they must control for the FWER and carefully consider how individual tests reject the null.
This is worth taking some time to read carefully. H/T R-Bloggers
Comments closedStephanie Glen disambiguates three commonly confused but quite different terms:
In a nutshell, here are the definitions for all three.
1. Significance level: In a hypothesis test, the significance level, alpha, is the probability of making the wrong decision when the null hypothesis is true.
2. Confidence level: The probability that if a poll/test/survey were repeated over and over again, the results obtained would be the same. A confidence level = 1 – alpha.
3. Confidence interval: A range of results from a poll, experiment, or survey that would be expected to contain the population parameter of interest. For example, an average response. Confidence intervals are constructed using significance levels / confidence levels.
Read on for several examples and more elaboration.
Comments closedAbhinav Choudhary shows us how to implement Principal Component Analysis in Python:
Principal Component Analysis (PCA) is an unsupervised statistical technique used to examine the interrelation among a set of variables in order to identify the underlying structure of those variables. In simple words, suppose you have 30 features column in a data frame so it will help to reduce the number of features making a new feature which is the combined effect of all the feature of the data frame. It is also known as factor analysis.
PCA is quite useful in practice, though it has the unfortunate side effect of making it harder to interpret which factors are driving your solution.
Comments closedOne reason that the proper residual graph (for a well fit model) should smooth out to the line
y=0is known as reversion to mediocrity, or regression to the mean.Imagine that you have an ideal process that always produces a single value y. You don’t actually observe this “true value”; instead, what you observe is y plus (IID, zero mean) noise. You can build a “model” for this process that predicts the mean of the observations, in this case the value 0.1033149. Then you can calculate the residuals of your “model” in the usual way.
This post went in a direction I wasn’t expecting, and it was all the better for it.
Comments closed