I was thinking of using this for a course in linear regression but unfortunately this doesn't include any of the mathematics behind the analysis. This can be used simply to learn how to do regression in R in a few hours without any knowledge of Comprehensiveness rating: 3 see less.
This can be used simply to learn how to do regression in R in a few hours without any knowledge of the algorithm used or the mathematics that goes into it. The material and the CSV files provided for practicing regression with data is accurate and useful. The pictures included in the body of the text are very helpful to new R learners. It could be more comprehensive as a textbook in linear regression. But it simply doesn't work as a college level textbook.
Perhaps just used for one week of the course. The writing and the pictures and the data attached are very accessible and clear. It's amazing that the content is hyperlinked,too. The material is consistently taught at the same level of difficulty which makes is accessible for high school or middle school students. The order of the material presented in the book is logical and accessible for an easy level of difficulty. The images and hyperlinked content are string areas of the book that attracted me to browse through.
More appropriate for elementary audience of linear regression. Not appropriate for a data science program. I would recommend this for middle school or high school students. It is very focused and short treatment of simple and multiple linear regression. It does not cover categorical predictors, interactions of predictors, doesn't spend much time interpreting slope coefficients, or discuss confidence intervals for It does not cover categorical predictors, interactions of predictors, doesn't spend much time interpreting slope coefficients, or discuss confidence intervals for them, and has some technical issues with the methods discussed.
So it is not a very comprehensive treatment of the topic but does provide a short introduction to simple and multiple linear regression model building with quantitative predictors and how to use R to do that.
I like Chapter 5 best and the diagram for training and validation and how this fits into an engineering perspective on statistical model building and evaluation. There are issues of bias not discussed that could be discussed with the missing data discussion and the sloppiness of this discussion propagates into issues with models and comparing models for different sets of responses that are problematic. Some of the notation is a little bit loose for the statistical modeling and inferences discussed — for example there is no distinction between population and sample estimates in the models.
There are some incorrect interpretations of p-values and the overall F-test. And some modeling choices are difficult to understand including both square-root and original versions of the same predictors — why do we need both?
How would you interpret the model with both? Doesn't this create multicollinearity issues? It presents a good introduction to the topics promised. I am not sure this would be enough for a full semester of a course — more of a unit for a course. And it is a bit sparse on tools to prepare students to try to use these methods on their own similar data sets. It is well written. Some of the technical issues cause me concern in using it for a course, but it is a nice book to read and has some interesting points covered.
The links appear to be dynamic within the document but do not seem to work on my system. It is otherwise a nice platform and formatting for the material. These issues would prevent my use of the book in my classes. The one example data set relates to computers and there is nothing else that might either possibly allow incorporation of, or cause concerns about, different cultural perspectives.
There are some technical issues with the discussions and examples that a discussion with a statistician could aid in resolving.
I think it could resonate with an engineering audience and with some modest changes be very successful at presenting the material to that audience. I really liked Chapter 5 and found that to be the most successful topic presented. Comprehensiveness rating: 5 see less.
My only problem is, the author calls variables in data sets "parameters". Within the context of linear regressions, I believe the term "parameters" should be reserved for coefficients in the model that will be estimated. By showing linear regressions with the statistical software R, the book gives a modern and hands on approach the material.
I think the best thing about this book is it's clarity. The clear and concise language of this book makes it very friendly to readers. By using one variable that is modeled throughout the entire book, it allows for a nice connectiveness between chapters. For the potential reader with little R programming and data science background, this book quickly allows someone to build a linear model from a given data set.
Also, the book has a nice introduction to training and testing a linear model. With the authors clear and easy to read explanations, this will be a text that I will refer to people to for quickly running linear regressions in R. There are basic functions such as class or typeof that should be introduced early on for any user of R. Also, A practical explanation of residual standard error or what a nonsensical model for the example used throughout the text would be Also, A practical explanation of residual standard error or what a nonsensical model for the example used throughout the text would be helpful for a beginner.
Using vocabulary to help student differentiate between an assumed model and a prediction equation would be helpful if you are planning to use this as a classroom text. Depending on how you are used to teaching regression, you may find many problematic uses of vocabulary or you may find none. The main vocab is touched on and explained well, minus some possible misuse of terminology depending upon how one teaches regression. As for technical R vocab, the use of 'row' early on in the text to describe the header of a data frame could also be problematic since the first row of a data frame typically refers to the first row of data, not the names of the columns.
The text follows the typical presentation of a traditional look at regression, which makes for a text that is clear and well organized. I think all is fine. I wouldn't see any particular computer processor feeling like they have been misrepresented or purposely left out. I hope this review actually goes through this time???
It is my third attempt at trying to complete this before Qualtrics times me out. Sorry that I am so slow Also, below is a paragraph style review. I wrote this before seeing the actual format was going to be a survey type of setup.
The author does use an example throughout that many can understand at least to some degree influencers of computer performance , which exposes the reader to the useful concept that knowledge about a data set can be extremely useful. Further, the introduction of functions like attach and update are examples of how the author has nicely woven into the content a practical approach to how coding is part of analysis.
The exploratory use of plot to visualize the data before introducing a one-factor regression is another positive example of this. However, there are some places throughout the book that might make you seriously question whether you could teach a course using this book either as a stand-alone resource or just a supplementary one. The wording in some places can be confusing or even contradictory depending upon how you present regression, especially as an introduction to the topic where consistent use of vocabulary can be crucial.
For example, consider in Section 3. Maybe this example isn't problematic for you, nonetheless I still suggest you carefully look through the entire book before adopting it as a resource for your students. This is a tutorial that covers basic areas and ideas of linear regression.
Now as feature first we will be considering only the population of the city in s , x1. Since y , profit depends only on one variable x1 , population, we call it simple linear regression. We square it to negate the sign and put a half before for a mathematical convenience. Here the blue ones are correctly predicted by the line and thus have no error and the red ones are with errors.
Our task is to find w0 , w1 so that the error e is minimized. We are going to use gradient descent algorithm to find these values. The function is minimized at a point where the slope is zero.
We randomly start from any value of w0 or w1 and eventually reach the minimum. Lets see how that is done! We could either start at w00 or at w but we wish to reach w0 From w00 , we have to increase the value and move right and here at this point slope of the tangent is negative.
From w , we have to decrease the value and move left and here at this point slope of the tangent is positive. Its interesting to note that, more the distance from the point to the minimum is the value is slope is larger.
We thus can change the weights proportionate to the slope. Again too high value will cause divergence. Note, each time the error in prediction is multiplied by 1 for w0 or multiplied by the feature xj for wj. Lets assume there are n features. Note that each of the values of this vector E has to be multiplied by corresponding features and then summed.
That is now easily achieved by the following multiplication:. We can do the whole thing without any loops! The error function is the trick! Now at minimum the differentiation of this error function must be 0. All weights to be found by a single line containing some matrix operations!
This problem can be formulated as a classification problem. We can apply logistic regression when the data is linearly separable. The relationship will be predicted as:. A little trick required! This linear classifier divides instances based on the local wrt the line, on the right positive, negative on the left Any point on the line satisfies the equation. Any point on the right 3,1 yields positive result and any point on the left 1,1 yields negative result. This is not a continuous function and thus not differentiable.
We need to find an alternate! Good things about sigmoid! Lets try! Open navigation menu. Close suggestions Search Search. User Settings. Skip carousel. Carousel Previous. Carousel Next. What is Scribd? Explore Ebooks. Bestsellers Editors' Picks All Ebooks. Explore Audiobooks. Bestsellers Editors' Picks All audiobooks. Explore Magazines. Editors' Picks All magazines. Explore Podcasts All podcasts. Difficulty Beginner Intermediate Advanced.
0コメント