lecture 2: a crash course in r

Transcription

1 lecture 2: a crash course in r STAT 545: Introduction to computational statistics Vinayak Rao Department of Statistics, Purdue University August 20, 2018

2 The programming language From the manual, is a system for statistical computation and graphics provides a programming language, high level graphics, interfaces to other languages and debugging facilities It is possible to go far using interactively Better: Organize code for debugging/reproducibility/homework 1/39

3 The programming language John Chambers: Everything that exists is an object Everything that happens is a function call gives the type or internal storage mode of an object provides a summary of the object returns the object s class 2/39

4 Atomic vectors Collections of objects of the same type Common types include: has no scalars, just vectors of length 1 3/39

5 Creating vectors One-dimensional vectors: 4/39

6 Creating vectors 5/39

7 Indexing elements of a vector Use brackets to index subelements of a vector 6/39

8 Indexing elements of a vector Negative numbers exclude elements 6/39

9 Indexing elements of a vector Index with logical vectors 6/39

10 Indexing elements of a vector Example: sample 100 Gaussian random variables and find the mean of the positive elements 6/39

11 Indexing elements of a vector Example: sample 100 Gaussian random variables and find the mean of the positive elements More terse 6/39

12 Replacing elements of a vector Can assign single elements 7/39

13 Replacing elements of a vector Can assign single elements or multiple elements 7/39

14 Replacing elements of a vector Can assign single elements or multiple elements or assign multiple elements a single value (more on this when we look at recycling) 7/39

15 Replacing elements of a vector What if we assign to an element outside the vector? We have increased the vector length by 1 In general, this is an inefficient way to go about things Much more efficient is to first allocate the entire vector 8/39

16 Recycling 9/39

17 Recycling Can repeat explicitly too 9/39

18 Recycling Can repeat explicitly too 9/39

19 Some useful functions, Comparison operators: Logical operators: Coercion often happens implicitly in function calls: 10/39

20 Lists (generic vectors) in Elements of a list can be any object (including other lists) Lists are created using : 11/39

21 Lists (generic vectors) in Elements of a list can be any object (including other lists) Lists are created using : Can have nested lists: 11/39

22 Indexing elements of a list Use brackets and double brackets Brackets return a sublist of indexed elements 12/39

23 Indexing elements of a list Use brackets and double brackets Brackets return a sublist of indexed elements 12/39

24 Indexing elements of a list Use brackets and double brackets Double brackets return element of list 12/39

25 Named lists Can assign names to elements of a list 13/39

26 Named lists Can assign names to elements of a list 13/39

27 Object attributes is an instance of an object attribute These store useful information about the object 14/39

28 Object attributes is an instance of an object attribute These store useful information about the object Other common attributes:, and. Many have specific accessor functions e.g. or You can create your own 14/39

29 Matrices and arrays Are two- and higher-dimensional collections of objects These have an appropriate attribute 15/39

30 Matrices and arrays Are two- and higher-dimensional collections of objects These have an appropriate attribute Equivalently (and better) 15/39

31 Matrices and arrays Are two- and higher-dimensional collections of objects These have an appropriate attribute 15/39

32 Matrices and arrays Useful functions include Matrix multiplication is carried out with the %*% operator Simple * is elementwise multiplication 16/39

33 Matrices and arrays A vector/list is NOT an 1-d matrix (no attribute) 17/39

34 Matrices and arrays A vector/list is NOT an 1-d matrix (no attribute) Use to eliminate empty dimensions 17/39

35 Matrices and arrays A vector/list is NOT an 1-d matrix (no attribute) Use to eliminate empty dimensions 17/39

36 Indexing matrices and arrays 18/39

37 Indexing matrices and arrays Excluding an index returns the entire dimension 18/39

38 Indexing matrices and arrays Excluding an index returns the entire dimension Usual ideas from indexing vectors still apply 18/39

39 Column-major order We saw how to create a matrix from an array 19/39

40 Column-major order We saw how to create a matrix from an array In matrices are stored in column-major order (like, and unlike and ) 19/39

41 Recycling Column-major order explains recycling to fill larger matrices 20/39

42 Recycling Column-major order explains recycling to fill larger matrices 20/39

43 Recycling Column-major order explains recycling to fill larger matrices 20/39

44 Data frames Very common and convenient data structures Used to store tables: Columns are variables and rows are observations Age PhD GPA Alice 25 TRUE 3.6 Bob 24 TRUE 3.4 Carol 21 FALSE 3.8 An data frame is a list of equal length vectors and special convenience syntax 21/39

45 Data frames 22/39

46 Data frames Since data frames are lists, we can use list indexing Can also use matrix indexing (more convenient) list functions apply as usual matrix functions are also interpreted intuitively 23/39

47 Data frames Many datasets are data frames and many packages expect dataframes 24/39

48 Allow conditional execution of statements 25/39

49 Logical operators : logical negation and : logical and and : logical or and perform elementwise comparisons on vectors and : evaluate from left to right look at first element of each vector evaluation proceeds only until the result is determined 26/39

50 explicit looping: and 27/39

51 explicit looping: and 27/39

52 explicit looping: and 27/39

53 explicit looping: and 27/39

54 Vectorization Vectorization allows concise and fast loop-free code Example: Entropy H(p) = p i=1 p i log p i of a prob. distrib. 28/39

55 Vectorization Vectorization allows concise and fast loop-free code Example: Entropy H(p) = p i=1 p i log p i of a prob. distrib. 28/39

56 Vectorization Vectorization allows concise and fast loop-free code Example: Entropy H(p) = p i=1 p i log p i of a prob. distrib. 28/39

57 Vectorization Vectorization allows concise and fast loop-free code Example: Entropy H(p) = p i=1 p i log p i of a prob. distrib. 28/39

58 While loops If doesn t affect, we loop forever. Then, we need a statement. Useful if many conditions 29/39

59 The *apply family Useful functions for repeated operations on vectors, lists etc. Sample usage: Stackexchange has a nice summary: [url] Note (Circle 4 of the R inferno): These are not vectorized operations but are loop-hiding Cleaner code, but comparable speeds to explicit loops 30/39

60 functions comes with its own suite of built-in functions An important part of learning is learning the vocabulary See e.g. Non-trivial applications require you build your own functions Reuse the same set of commands Apply the same commands to different inputs Cleaner, more modular code Easier testing/debugging 31/39

61 Creating functions Create functions using : The above statement creates a function called formal_arguments comma separated names describes inputs expects function_body a statement or a block describes what does with inputs 32/39

62 An example function 33/39

63 Argument matching Proceeds by a three-pass process Exact matching on tags Partial matching on tags: multiple matches gives an error Positional matching Any remaining unmatched arguments triggers an error 34/39

64 Plotting in base R 35/39

65 Plotting in ggplot 36/39

66 Plotting in ggplot Of course, has intelligent defaults 36/39

67 Plotting in ggplot Of course, has intelligent defaults There s also further abbreviations via (I find it confusing) 36/39

68 Layers ggplot produces an object that is rendered into a plot This object consists of a number of layers Each layer can get own inputs or share arguments to 37/39

69 Layers ggplot produces an object that is rendered into a plot This object consists of a number of layers Each layer can get own inputs or share arguments to Add another layer to previous plot: 37/39

70 More Examples 38/39

71 A more complicated example Moscou Kowno Wilna Smorgoni Moiodexno Gloubokoe Polotzk Bobr Studienska Witebsk Orscha Smolensk Dorogobouge Chjat Wixma Mojaisk Tarantino Malo Jarosewii direction A R survivors 1e+05 2e+05 3e Minsk Mohilow A Layered Grammar of Graphics, Hadlay Wickham, Journal of Computational and Graphical Statistics, 2010 documentation: Search ggplot on Google Images for inspiration Play around to make your own figures 39/39