1 00:00:00,000 --> 00:00:04,331 You now know a bunch about machine learning. In this video, I'd like to teach 2 00:00:04,331 --> 00:00:08,662 you a programming language, Octave, in which you will be able to very quickly, 3 00:00:08,662 --> 00:00:13,278 implement the learning algorithms we've seen already and the learning algorithms 4 00:00:13,278 --> 00:00:17,837 we'll see later in this course. In the past I've tried to teach machine learning 5 00:00:17,837 --> 00:00:22,169 using a, a large variety of different programming languages, including C plus, 6 00:00:22,169 --> 00:00:27,826 plus, Java, Python and Pi are. And also octave and what I've found is that 7 00:00:27,826 --> 00:00:33,062 students were able to learn the most productively learn the most quickly and 8 00:00:33,062 --> 00:00:38,560 prototype your albums most quickly using a relatively high level language by octave 9 00:00:38,560 --> 00:00:43,605 in fact What I often see in silicon valley is even if you need to build if you want 10 00:00:43,605 --> 00:00:48,040 to build a large scale deployment of a learning algorithm what people will often 11 00:00:48,040 --> 00:00:52,036 do is prototype in the language they Octave which is a great prototyping 12 00:00:52,036 --> 00:00:56,307 language so you can get your learning albums working quickly and then only if 13 00:00:56,307 --> 00:01:00,194 you need a very large scale deployment of it only then spend your time 14 00:01:00,194 --> 00:01:04,793 reimplementing the algorithms [inaudible] some other language like that because one 15 00:01:04,793 --> 00:01:09,118 of the lessons we've learned is that programmer time or developer time that is 16 00:01:09,118 --> 00:01:13,280 your time with the machine learning [inaudible] time is incredibly valuable. 17 00:01:13,280 --> 00:01:18,274 And if you can get your learning algorithms to work more quickly in octave 18 00:01:18,274 --> 00:01:23,206 then overall you have a huge time savings by first developing the algorithms in 19 00:01:23,206 --> 00:01:28,262 octave and then implementing [inaudible] only after we have the ideas working. The 20 00:01:28,262 --> 00:01:32,948 most common prototyping languages I see people use for machine learning are 21 00:01:32,948 --> 00:01:42,073 Octave. Mattlab-numpinr. Octave is nice this is free open source and Mat-Lab works 22 00:01:42,073 --> 00:01:47,656 well too but is expensive to many people but if you have access to copy Mat-Lab you 23 00:01:47,656 --> 00:01:53,371 can also use mat lab for this class if you know [inaudible] or if you know R I do see 24 00:01:53,371 --> 00:01:59,153 some people use it but what I see is that people usually end up developing somewhat 25 00:01:59,153 --> 00:02:04,029 more slowly in. You know more complete languages [inaudible] syntax it's just 26 00:02:04,029 --> 00:02:09,535 slightly [inaudible] clear than the octave syntax and so because of that and because 27 00:02:09,535 --> 00:02:14,728 we're releasing starter code in Octave I strongly recommend that you not try to do 28 00:02:14,728 --> 00:02:20,108 the pony exercises in this class in Empire R but that I do recommend that you instead 29 00:02:20,108 --> 00:02:25,359 do the pony exercises for this class in octave instead. Welcome to do this video 30 00:02:25,359 --> 00:02:31,006 is go through a list of commands very, very quickly and goes will quickly show 31 00:02:31,006 --> 00:02:36,942 you the range of commands and the range of things you can do in office. The course 32 00:02:36,942 --> 00:02:42,516 web site we have a transcripts of everything I do and so after watching this 33 00:02:42,516 --> 00:02:48,379 video you can refer to transcript person in the course web site when you want to 34 00:02:48,379 --> 00:02:54,098 find a command. Our feel free whatever recommend you do is fresh wash and after 35 00:02:54,098 --> 00:02:58,765 watching Now DDN then is [inaudible] octave on your computer and finally it 36 00:02:58,765 --> 00:03:03,679 goes to the course website download the transcripts of the things you see in this 37 00:03:03,679 --> 00:03:08,354 section and type in whatever commands seem interesting to you into octaves 38 00:03:08,354 --> 00:03:13,446 [inaudible] on your own computer so that you can see it for yourself. And, with 39 00:03:13,446 --> 00:03:19,421 that let's get started. Here's my Windows desktop, and I'm start in Octave. And I'm 40 00:03:19,421 --> 00:03:23,849 now in Octave and, and there's my Octave prompt. Let me first show you the 41 00:03:23,849 --> 00:03:28,458 elementary operations you can do in Octave. So, you can type in five plus six, 42 00:03:28,458 --> 00:03:34,720 that gives you the answer of eleven. Z minus two. Five times eight. One over two. 43 00:03:34,720 --> 00:03:41,045 Two to the power of six. Is 64. So that the [inaudible], elementary, math 44 00:03:41,045 --> 00:03:46,426 operations. You can also do logical operations. So one=2 of this evaluation 45 00:03:46,426 --> 00:03:52,703 that follows. The percent command here is, means a comment. So one=2, evaluates the 46 00:03:52,703 --> 00:03:58,905 [inaudible], which is represented by zero. One [inaudible]= to two. This is true, so 47 00:03:58,905 --> 00:04:04,866 that returns one. You know that a not equal sign is this. To their equals symbol 48 00:04:05,098 --> 00:04:11,973 and not bang equals which some of the [inaudible] languages use. The theological 49 00:04:11,973 --> 00:04:18,230 operations one and zero use a double ampersand sign to denote the logical and. 50 00:04:18,230 --> 00:04:24,021 That evaluates the false, one or zero, z or operation. And that evaluates the true. 51 00:04:24,021 --> 00:04:29,813 And I can x over one and zero, and that evaluates the one. This thing over on the 52 00:04:29,813 --> 00:04:35,388 left, this Octive 324 [inaudible] eleven, this is the default Octive prompt. It 53 00:04:35,388 --> 00:04:40,962 shows what version of Octive and so on. If you don't want that prompt, this is 54 00:04:40,962 --> 00:04:46,485 somewhat cryptic to come on. Quote particular then and so on that you can use 55 00:04:46,485 --> 00:04:52,121 to change the prop and I guess this quoted string in the middle you know quote 56 00:04:52,121 --> 00:04:57,686 greater than greater than state, that's what I prefer my octave prompt to look 57 00:04:57,686 --> 00:05:03,179 like, so if I hit enter, oh excuse me, like so, PS1 like so now my octave prompt 58 00:05:03,179 --> 00:05:09,100 has changed to the greater than greater than sign which you know looks quite a bit 59 00:05:09,100 --> 00:05:15,295 better. Next lets talk about octave variables. I can take the variable A and 60 00:05:15,295 --> 00:05:21,181 assign it to three and hit enter. And now A is equal to three. You want to sign a 61 00:05:21,181 --> 00:05:26,008 variable but you don't want to print out the result if you put a semicolon. The 62 00:05:26,008 --> 00:05:32,546 semicolon. Suppresses, the, print, the, the print output. So to do that enter, it 63 00:05:32,546 --> 00:05:38,639 doesn't print anything. Whereas A=3, you know, makes it printed out. Whereas A=3; 64 00:05:38,639 --> 00:05:44,798 doesn't print anything. I can do string assignment, B=high. If I just enter b, it 65 00:05:44,798 --> 00:05:51,726 prints out the variable b. So b is the string high. C equals three grade integral 66 00:05:51,726 --> 00:05:58,220 one. So now c evaluates the true. If you want to print out a display of your 67 00:05:58,220 --> 00:06:05,029 variable, here's how you go about it. Let me set a equals pi. And, if I wanna print 68 00:06:05,029 --> 00:06:10,341 A, I can just type A, like so, and it prints it out. For more complex printing, 69 00:06:10,341 --> 00:06:15,795 there's also the [inaudible] command, which, sends the displays. [inaudible] 70 00:06:15,795 --> 00:06:21,107 display A, it just prints out A like so. You can also display the strings, so 71 00:06:21,107 --> 00:06:27,571 [inaudible] F, two decimals. And percent 0.2f, a like so. And this will print out 72 00:06:27,571 --> 00:06:34,076 the string two decimal colon 3.14. This is kind of a [inaudible] old style C-syntax. 73 00:06:34,311 --> 00:06:40,502 For those of you that have program C before this is [inaudible] syntax used to 74 00:06:40,502 --> 00:06:46,522 print strings. So sprintf generates a sprint, generates a string. Does this and 75 00:06:46,522 --> 00:06:53,020 two decimals 3.1 pulls the string this percent 0.2F means substitute A into here 76 00:06:53,020 --> 00:06:58,976 showing it with two digits after the decimal point and this takes the string 77 00:06:58,976 --> 00:07:05,086 this generated by the sprint F command S print f the sprint F command and this 78 00:07:05,086 --> 00:07:11,378 actually displays the string and to show you an example, sprint F six decimals. >> 79 00:07:11,378 --> 00:07:17,660 Percent 0.6F comma A. And this should print pi with six decimal, to, to, six 80 00:07:17,660 --> 00:07:24,287 decimal places. Finally, I'll explain A like so, look like this, there, there are 81 00:07:24,287 --> 00:07:30,656 useful shortcuts you take, formats long, it causes strings to, by default, be 82 00:07:30,656 --> 00:07:37,455 displayed to a lot more decimal places, and format short [inaudible] default of 83 00:07:37,455 --> 00:07:44,200 just printing a small number of digits. >> Okay. That's how you work with variables 84 00:07:44,200 --> 00:07:50,529 now lets look at vectors and [inaudible] let's say you want to assign A to matrix 85 00:07:50,529 --> 00:07:56,081 let me show you an example, twelve:34: . Five, six. This generates a three by three 86 00:07:56,081 --> 00:08:01,366 matrix A who's first row is 1,2, second row three, four, third row is five, six. 87 00:08:01,366 --> 00:08:06,721 What the semi colon does is essentially say go to the next row of the matrix. 88 00:08:06,721 --> 00:08:12,297 There are other ways to type this in. Type angles one, two semi colon. Three, four; 89 00:08:12,297 --> 00:08:18,149 five, six, like so. And that's another equivalent way of assigning A to be the 90 00:08:18,149 --> 00:08:24,232 values of, this, three by two matrix. Some [inaudible] assign factors. So V=1, two, 91 00:08:24,232 --> 00:08:30,392 three. This is actually a [inaudible] vector. Or this is a three by one vector, 92 00:08:30,392 --> 00:08:36,321 right? This is a [inaudible] Y vector, excuse me. Not, this is, a one by three 93 00:08:36,321 --> 00:08:44,273 matrix. Right not N by one if I want to assign this to a. Harden factor. What we 94 00:08:44,273 --> 00:08:50,290 do instead is B equals one semicolon, two semicolon, three and this will give me a. 95 00:08:50,290 --> 00:08:57,162 Three by one is one by three vector so this is a column vector. Here some more 96 00:08:57,162 --> 00:09:04,299 useful notations b equals one colon 0.1 colon two. What this does is it sets b to 97 00:09:04,299 --> 00:09:12,236 the bunch of elements that start from one. And increments and, and steps of 0.1 until 98 00:09:12,236 --> 00:09:20,508 you get up to two. So if I do this, V is going to be this you know, low vector. Or 99 00:09:20,508 --> 00:09:28,679 this is what, one by eleven matrix really. That's 1.1, one, 1.1, 1.2, 1.3 and so on 100 00:09:28,679 --> 00:09:37,429 until we get up to two. Now and I can also say B = one:6 and that sets B to be these 101 00:09:37,429 --> 00:09:44,835 numbers one through six. Okay now here's some other way to generating matrices, 1's 102 00:09:45,093 --> 00:09:52,328 2x3 is a command that generates a matrix that is a 2x3 matrix that is the matrix of 103 00:09:52,328 --> 00:09:59,649 our 1's so if I set C = two x 1's 2x3 this generates a 2x3 matrix that is all 2's and 104 00:09:59,649 --> 00:10:07,228 this is you can think of this as a shorter way of writing this NC = 222 semicolon 222 105 00:10:07,228 --> 00:10:16,100 which will also give you the same result. Lets say W = 1's, 1x3 so this is going to 106 00:10:16,100 --> 00:10:27,316 be a row vector or a you know row of three 1's and some of you, you can also say W = 107 00:10:27,316 --> 00:10:36,121 0's 1x3 this generates a matrix, 1x3 matrix, of all 0's. Just a couple more 108 00:10:36,121 --> 00:10:43,205 ways to generate matrixes. There I do W equals rand one by three. This gives me a 109 00:10:43,205 --> 00:10:49,628 one by three matrix of all random numbers. If I do rand. Three by three excuse me a 110 00:10:49,628 --> 00:10:53,918 three by three matrix of all random numbers drawn from the uniform 111 00:10:53,918 --> 00:10:59,105 distribution which is 01. So every time I do this I get a different set of random 112 00:10:59,105 --> 00:11:04,163 numbers drawn uniformly between the zero and the one. For those of you who know 113 00:11:04,163 --> 00:11:09,158 what a Gaussian random variable is or for those who know what a Normal random 114 00:11:09,158 --> 00:11:14,792 variable is you can also say that W called rant N one-third. So these are going to be 115 00:11:14,792 --> 00:11:19,787 three values drawn from a Gaussian distribution which means zero and variance 116 00:11:19,787 --> 00:11:25,654 distribution is equal to one. And. Can set more complex things like W = -six plus 117 00:11:25,654 --> 00:11:31,475 square root ten. Times. Let say rand [inaudible] one by 10,000. [inaudible] 118 00:11:31,475 --> 00:11:37,928 semicolon at the end because I don't really want to [inaudible] it out. This is 119 00:11:37,928 --> 00:11:44,695 going to be a what. Well this is going to be a vector of 100,000 [inaudible] 10,000 120 00:11:44,695 --> 00:11:50,352 elements so. Well actually you know what let's print it up so this would generate a 121 00:11:50,352 --> 00:11:56,445 matrix like this Right? With 10,000 elements so that's what W is. And if I now 122 00:11:56,445 --> 00:12:02,958 plot a histogram, W, with the hiss command I can now enter [inaudible]. Hiss command 123 00:12:02,958 --> 00:12:08,915 takes a couple seconds to bring these up but this is a histogram, my random 124 00:12:08,915 --> 00:12:14,871 variable for W that was minus six plus zero ten times this Gaussian random 125 00:12:14,871 --> 00:12:21,305 variable. And I can plot a histogram with more buckets, with more bins, with say 50 126 00:12:21,305 --> 00:12:27,737 bins and. This is my histogram of [inaudible] with mean - six because I have 127 00:12:27,737 --> 00:12:34,531 a -six t here plus [inaudible] ten times so the, the variants or this calcium is 128 00:12:34,531 --> 00:12:41,408 therefore ten on the standard deviation the square root of ten which is about what 129 00:12:41,408 --> 00:12:46,664 3.1. Finally one special command to generate a matrix which is the I command 130 00:12:46,664 --> 00:12:52,878 so. I stands for, this is maybe pun on the word identity. But so they set i4, this is 131 00:12:52,878 --> 00:12:58,813 the four by four identity matrix. And I, of course i4, this gives me a four by four 132 00:12:58,813 --> 00:13:04,602 identity matrix and I [inaudible] i5, i6, that gives me a six by six identity 133 00:13:04,602 --> 00:13:09,878 matrix, as i3 gives the three by three identity matrix. Lots of these are 134 00:13:09,878 --> 00:13:15,887 [inaudible] has one more useful command which is the help command. So you can type 135 00:13:15,887 --> 00:13:21,761 help I, and this brings up the help function for the identity matrix. Hit Q to 136 00:13:21,761 --> 00:13:27,702 quit and you can also type help rand brings up documentation for the rand or 137 00:13:27,702 --> 00:13:33,643 the random number generation function or even help, help which shows you help on 138 00:13:33,643 --> 00:13:40,568 the help function. So, those are the basic operations in octave and with this you 139 00:13:40,568 --> 00:13:46,486 should be able to generate a few matrices, multiply, add things, and, and use the 140 00:13:46,486 --> 00:13:52,111 basic operations in octave. In the next video, I'd like to start talking about 141 00:13:52,111 --> 00:13:58,321 more sophisticated commands and how to move data around and start to process data 142 00:13:58,321 --> 00:13:59,052 in octave.