Machine Learning for dummies with Python
Formal Metadata
Title 
Machine Learning for dummies with Python

Title of Series  
Part Number 
3

Number of Parts 
169

Author 

License 
CC Attribution  NonCommercial  ShareAlike 3.0 Unported:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal and noncommercial purpose as long as the work is attributed to the author in the manner specified by the author or licensor and the work or content is shared also in adapted form only under the conditions of this license. 
Identifiers 

Publisher 

Release Date 
2016

Language 
English

Content Metadata
Subject Area  
Abstract 
Javier Arias Losada  Machine Learning for dummies with Python Machine Learning is the next big thing. If you are a dummy in terms of Machine Learning, but want to get started with it... there are options. Still, thanks to the Web, Python and OpenSource libraries, we can overcome this situation and do some interesting stuff with Machine Learning.  Have you heard that Machine Learning is the next big thing? Are you a dummy in terms of Machine Learning, and think that is a topic for mathematicians with blackmagic skills? If your response to both questions is 'Yes', we are in the same position. Still, thanks to the Web, Python and OpenSource libraries, we can overcome this situation and do some interesting stuff with Machine Learning.

00:00
Digital photography
Goodness of fit
Spring (hydrology)
Film editing
Multiplication sign
Virtual machine
Video game
Office suite
01:13
Uniformer Raum
Multiplication sign
View (database)
Video game
Generic programming
Ext functor
Office suite
01:50
Casting (performing arts)
Video game
Musical ensemble
02:32
Machine learning
Video game
03:16
Area
Presentation of a group
Arithmetic mean
03:37
Presentation of a group
Open source
Source code
Expert system
Video game
Theory
Element (mathematics)
Library (computing)
04:19
Presentation of a group
Machine learning
Dependent and independent variables
Goodness of fit
Algorithm
Sample (statistics)
Summierbarkeit
Virtual machine
04:53
Medical imaging
Pattern recognition
Similarity (geometry)
05:24
Atomic number
Multiplication sign
Software testing
Game theory
Game theory
Number
06:09
Predictability
Complex (psychology)
Inclusion map
Spring (hydrology)
Googol
Moment (mathematics)
Virtual machine
Mathematical singularity
Game theory
Alpha (investment)
06:50
Multiplication sign
Rhombus
07:15
Machine learning
Pattern recognition
1 (number)
Interactive television
Mass
Open set
Line (geometry)
Content (media)
Mathematics
Arithmetic mean
Internet service provider
Scalar field
Universe (mathematics)
Videoconferencing
Row (database)
Elearning
08:30
Machine learning
Multiplication sign
Electronic mailing list
Reading (process)
09:14
Predictability
Classical physics
Natural number
output
Computer programming
09:44
Revision control
Pattern recognition
Complex number
Personal digital assistant
Image resolution
Insertion loss
Computer programming
10:20
Machine learning
Algorithm
Multiplication sign
Computer
Energy level
Computer
10:46
Point (geometry)
Machine learning
Mathematics
View (database)
Execution unit
Price index
Computer
Wave packet
Computer
Spacetime
11:10
Point (geometry)
Optical character recognition
View (database)
Energy level
Measurement
11:43
Algorithm
Sample (statistics)
Algorithm
Correspondence (mathematics)
Order (biology)
12:09
Algorithm
Algorithm
Decision theory
Multiplication sign
Projective plane
Combinational logic
Water vapor
Mereology
Event horizon
Wave packet
Wave packet
Medical imaging
Process (computing)
Vector space
Network topology
Modem
13:15
Word
Prediction
Different (Kate Ryan album)
Video game
Endliche Modelltheorie
Theory
Formal language
13:48
Open source
Algorithm
Scalar field
Open source
Different (Kate Ryan album)
Coordinate system
Video game
Cycle (graph theory)
Event horizon
Library (computing)
14:35
Predictability
Randomization
Functional (mathematics)
Algorithm
Linear regression
State of matter
Logistic distribution
Fitness function
Set (mathematics)
Measurement
Wave packet
Wave packet
Data model
Medical imaging
Prediction
Configuration space
Speech synthesis
Energy level
Software testing
Selectivity (electronic)
Process (computing)
Endliche Modelltheorie
Clef
Local ring
16:38
Predictability
Functional (mathematics)
Prediction
Software testing
Clef
Measurement
17:26
Medical imaging
Pattern recognition
Algorithm
Prediction
Algorithm
Scalar field
Configuration space
Maize
Line (geometry)
Machine code
Wave packet
18:33
Classical physics
Metropolitan area network
Dataflow
Matching (graph theory)
Programmer (hardware)
Moment (mathematics)
State of matter
Mereology
System call
Demoscene
Alpha (investment)
19:46
Functional (mathematics)
Implementation
Artificial neural network
Set (mathematics)
Water vapor
Semantics (computer science)
Tensor
Different (Kate Ryan album)
Operator (mathematics)
Scalar field
Graph (mathematics)
Vertex (graph theory)
Predictability
Addition
Multiplication
Matching (graph theory)
Dataflow
Artificial neural network
Weight
Open source
Operator (mathematics)
Mathematics
Array data structure
Right angle
Metric system
Resultant
21:45
Multiplication
Weight
Function (mathematics)
Vector potential
22:25
Revision control
Function (mathematics)
Sinc function
23:03
Goodness of fit
Open source
Freeware
23:32
Functional (mathematics)
Multiplication sign
Virtual machine
Mereology
Graph coloring
Formal language
Revision control
4 (number)
Frequency
Different (Kate Ryan album)
Core dump
Energy level
Endliche Modelltheorie
Physical system
Dependent and independent variables
Demon
Moment (mathematics)
Sampling (statistics)
Multiple Regression
Computer graphics (computer science)
Category of being
Type theory
Loop (music)
Process (computing)
Vector space
Software
Personal digital assistant
Bellman equation
Right angle
Active contour model
Library (computing)
00:01
good afternoon everyone it's my pleasure to introduce cut yeah yeah who is a senior developer at different again Anderson and here in spring and he's going to be talking about machine learning for them the meat using by with and you hear me well there in the yeah something they cannot please let me know close I lower my voice you need to be made in the back of convinced of the photos course we so that would also put my my OK that I was going to the this I could be added in your life this could be to several times to leave the office
01:15
and you font this view the reference to walk home it's fun because generic tall uniform where phonemes or what's the appropriate time to the office in fact on fridays the following day you there is no Europe to unified and so on because some fighting usa defiance so you will get back
01:38
you and you happen to be a copy disallowing not that everybody of us can be but let's imagine for a minute in the in the light that's another of thing was kind of there is hope pilot
01:53
features we felt the pilot because they want to keep the lane by moving the feeding will keep this speed and that dispense with the rest of the cast and not only that because the ceremony to learn not only from its own experiences but the mysterious system because last from around the world so you at home and you want to play some music but do not feel like
02:21
to see many music so the last thing the fun to play some music for you that you don't know but you actually know like this when
02:32
listening going missing you take your phone which have very well organized and definitely think of what is like here with architectural another but you know about them it was speaker that is able to look inside the follows and see what's inside and thank them accordingly these
02:54
things are happening they lead to thousands of even millions of people from around the world and all of them have something in common you know what it is it's matching learning matching learning is already realized he said when US and it will be very important and then as you know so
03:18
what's this presentation about and tried to explain area earlier the Weimar to learning the so important for us have you assuming and for companies before the rest of the world 4 so
03:34
as to explain all of my limiting meanings that
03:39
last and then kind of back and continue 6 months so I didn't know anything elements in London and I'm not an expert so it will be it would be new yes to study and practice to yourself but you can get started pretty quick and do
04:01
very interesting and finding things because there are many technologies and many other sources had found that the theory that open source and of course using Python and libraries with life on and off I tried to explain some very basic matching learning concepts and
04:20
will see a couple of gold so as
04:27
many of you know how much a living will be very important in the next years but 1 of the 1st
04:32
questions that came to my mind with that is is really Maxine learning uh Indonesian and the sum of all of those already better than and the response is that in some conquered questions is something good aspects Maxine's and algorithms are already better than it's simple that is the
04:56
much recognition is whole thing talent for much learning 1 a matter recognition and the performance of the we know about it and have been improving a lot during the last years thanks to the adoption
05:10
of the learning and this kind of in 2015 there we know about all of them the following data about human I think organizing things inside images something happened something similar
05:25
happened tests about 20 years old when they in the blue defeated the world champion Garry Kasparov the I remember vividly reading about these the news and reading it them the newspaper was 20 years so maybe that's right somebody extending out by not Monsignor develop bad but but then there was quite amazing at time but there all the games
05:54
that are much more complex and text for example the game of go big game of always so complex that the number of movements is bigger than the number of atoms in the universe for such a complex game it was called for such a
06:12
complex came into being important to played with inclusion of further research many people thought that's what it means to never win and they champions and did this spring when of all right well defeated the world champion use here this thing I wouldn't be even better than us and conquer the best of
06:36
happening more and more frequently for that there's so many people are making about the apocalyptic predictions from the moment that machines will be more intelligent and that's it would be a and it's the singularity or not OK I don't want to be
06:52
so apocalyptic we want to talk
06:54
to you about major success 6 months ago I didn't know anything of much of learning and I thought that it will would take weeks or months status just to get started but that's not true if you have a diamond like me you don't know anything about much learning you can get stuck and I did time
07:16
you can get very useful and very surprising things and this for me when that I decided to study and not by using courses so bold but by using Massive Open Online Courses and the schools
07:32
have an videos and exercises and that different provided therefore the what you can discuss with other people of the things you're doing and there are many different providers so you have to use your research here where change was the universities interaction too much means that it could be any other ones we'd be fit for you it's very well organized and he's using phone and a scalar and this was the last thing and also for me was very important that you can read at Europe their own place because I don't know I can't stick tool to the lines so this was a blasphemy and 1 thing that all my attention on the substrate which is but the recognition for fun simple things I can imagine all the profit but what about the final and because that's like these copper friends sharing of of 1 note by hand and and it ends
08:31
like this with the same friends and more when so it seems that much learning can also be an of so what is
08:42
meant learning I want to explain have been basic
08:47
concepts and some insights that many times and not well explained or just a little when you're stuck with matching learning and I think that there are things that are important and have a list for me and I want to share them with you on this is not the actual costs and you should let read books and of courses and the things that when we work in matching learning we want to solve complex problems so we have something
09:14
here in the middle and you shall you have some inputs which is the data naturalness features and we have to go up prediction which is dealt with of this thing here in the middle the 1st
09:30
approach classical approaches going broadening all of us here now how to program so you have to understand your problem and you have just plain it step by step in baby steps to a computer so that it can follow your
09:46
steps to what's the the solution but the problem is that for many complex domains churches the matter
09:56
recognition and may be seen these kind of things this this in this case because all let's see let's imagine that we have to solve something for medicine maybe you have to call those sensor and felt sense of of of loss and the loss of not excited because of medicine is not an exact science so maybe you have conflicts and you have to solve them and it's very
10:21
difficult to solve that if you use meta learning the approach is totally different instead of explaining the computer whatever the steps to solve the problem you show the computer some level have some examples of the time and you let the you let reality of love and take its
10:43
own conclusions from the data
10:46
this has huge indications and it's that we can teach computers to do things that we don't know how to do and they show you that in some units
11:00
so I would to explain from a user point of view I'm not going to enter into mathematical space because they study of mathematics
11:11
something like 23 years old and I couldn't remember about I actually just plain in from the point of view of the user How and let me know what's what's at a higher level and we will solve an example
11:26
for character recognition so we will have some sense of the measures like these into 1 these containing a character later on and we we're have delighted labels then announced that the 1st let the use and they have this economy and so on so the 1st step in
11:45
much of learning is getting our data that we have that we might say that corresponds when they have no the generally met which use gene and so on the 2nd
11:57
step is to choose an algorithm 98 before this was talking about naive Bayes that there many many things so maybe candidates of order and some of
12:09
them there is support vector readiness for or kmeans decision trees
12:14
neural networks many of them and you have to tools and the training I wouldn't that can be at which the 4 year project and also the use of events that can be computed so you have to choose some combination of modems and cooperation so this is part of the lecture learning and then to train job water for training and I want them to start by showing some images and you say OK this is usually much semantic and these generally Matt said and so on using this stick is like exact showing a baby to read it's like eating real 1st there just to to get really changed so you show me all time to the algorithm and will tell you I think the society but the ticket question
13:16
theories is that letter really I mean these are overloaded and predicting correctly the letters and sounds we will answer this question later from we have a lot
13:35
of words and we have different languages such as model and I have here on the talk about life and of course and I don't have to so here's
13:51
the beauty of life from but it's embedded with fear and it's following itself philosophy of batteries included the many many libraries with very good quality that we could use to solve problems where we see and we will solve the darker the coordination problems using I think and this could lead to 1
14:13
of the most popular library so this open source with iPhone phone and the documentation is wonderful not just for the library but I guess that's a reference to followed events and technologies and everything and as a scalar and give support to the full life cycle much of solar cells
14:36
so this problem we will do that in the following steps we'll get the features with labels we already have the data which is Europe and I would then we will be using logistic regression with no cooperation with the same then we do predictions and then we will buy them so let us that with the
15:02
example we have these big set here and thousand of with the labels we want to seperate them in 2 different datasets when the this is still training the other is for testing for validating our throats and we will use training this speech function from a skill level and we will speak of a person labels we will give a science for the whole training data set size matters but we will not displaying right here and there is some random state that happens to be a constant and the this shopping in the the very beginning when the hell are we giving a random state and happens to be a constant and the answer is that there test the speed we set by our big data datasets in a random set of images for the training and the test but bypassing discussed and the selection of the measures for the 2 because it would be the same always so that we can compare different configurations of the algorithms for different I wouldn't then we initialize our model and logistic regression local pollution and even if the fit is the training and we passed the training data set and the labels this is the other way when we beat the maybe 2 weeks that it's very soon we don't have to be very kind of thing and a skill and that's it for us Jupiter
16:39
chance we just call function on our classifier and that's all but the question is if our predictions where would survive to
16:52
solve this question we use accuracy there are many different ways in which the learning but we will use the communities is the simply 1 and we don't have to and from uh so what we always do predictions on our test dataset and we have some test predictions what we is compare the present labels which is begun to the things that we know what to to the test set predictions and the way we know what the percentage of the measures that we have been predicting correctly With this
17:27
example I just gave you we what 89 % of accuracy this means that of 500 5 thousand images were predicting where 90 per cent almost in just 5 lines of code and this is what I mentioned before I I I never did Matt's recognition or let's processing it would take many weeks or months to implement such a thing without much in learning and we it here in 20 minutes of and of course we can
18:01
improve difficult we can play with training that there less going out from the preview and told mentioned before we can change the LOD them and the configurations and this is very using a scalar because their API for for the different algorithms is almost the same and you can give a scalar different configurations for 1 over them and even with the best all the different configurations and give you the best 1 I'm so where the
18:38
scene in this very simple example with as classical matching learning Baron and doing that and of course the diversity of the learning I just at the very 1st lesson and with the idea the 1st exercise of a couple of exercises and I wanted to send them to you because I think that it's a part of all that you marvelous can understand I'm not going to explain noted that never wants and please don't ask it because I don't know anything about it but I think that you can get timedependent we should not how we are going to structure the call so this
19:21
is the course and still doing so I'm going to try and we will use a pencil flow which is highlighted antibiogram and queries the alpha of the course as well and it seems that uh the blurring is not so fine because there is no wine and the cost of the moment but I their 1st listens I will begin mathematical so we will be
19:47
using this fluoride i down there really political difference between a scalar and the potential flow is that sk Lebanese using an imperative API and this is from what you do is described by a set of mathematical operations in that kind of and then you execute that could have from with your written with your data so you have to know what are the mathematical operations to implement your and neural network so please don't make questions of of these but this is
20:26
the simplest possible neural network in the water this is set to like there's no redundant water and here in the left we have found match X and then on the right the right and we have a prediction of the prediction he sad Maddox with a set of probabilities for the 2 so we would would be the letter with more 1 and implementation is very they match the schematics and we will go the methods will be guided by might be so weights and we we will at the another metrics which is if I asked and we will apply to real function which is the same kind of feature of this is what I understood from differences and you have been blocked making up by their their their results from each 1 is going to be like this in next and then we will to the 2nd layer which is again a Matej multiplication and addition of of ideas and we will get our semantics of found weights and then we will do so max which is a function that transforms a set of weights in a simple probabilities so the implementation it's very simple it's uh if you
21:47
remember that you have to remember is
21:50
that we're what we're going to see how we do our a multiplication of meticulous of for this dataset with their 1st and my potential of weights for the 1st light and we have the biases the other 1 lighter 1 knows its and then we have like that available on the lighter 1 and we will have the output for the 1st layer then would repeat we had matter to multiplication and then shown and then work and then
22:28
we applied a softmax over the output of the previous version and this is the implementation of fast simple and there are no and it work this is the collection node and what you have to to train so that you can have these different and metrics since we've coded values for your problem and it's much more complex and maybe next year I was playing so this is
22:57
what I wanted to to share with you the idea is that unless there is quick summary matching learning these
23:06
hearings already in our lives and it would be gone
23:10
very important for everybody
23:13
so you view and I mean like any metalearning don't be afraid to come to worry finding the things with the
23:21
resources and tools that we have
23:23
the outside and they were very good quality free open source to use and you can do very funny things so you
23:32
want to use them and all interesting things just do it for your property have a
23:38
loop and bringing
23:41
in the thank you for the search sample questions and you can get that you know I think you could the vector regression analysis and those are known to to be a the individual of your immune to this is part of now you know this this this this this is 1 of those and think you for the question the soul and makes the in my experience and the question is that offenders can depression and got a new and from and the the questions is that have come 1st and then their performance from the 1st example of and from the 2nd and also what color compares with the performance with other languages such as C + + and the response is very easy for me because I'm not sure what to and so that we may experience and they did that performance of duty that at least for me my thought of the snow it works the smaller than is more than enough and there have been many advances in those years with the adoption of of 4th and graphic graphic process do we use and and and before and I evidence for example of the value function we're using here instead of single machine those functions that have made it possible to train very complex and networks in any case I don't know how it compares with see that that's where all this but want to show that if you want to train very complex models you will need to worry very expensive work but you can play with the with the just that smaller might thank you it has to have a small question like there are a lot of tools with different level of support for example which provide basically the iPhone emotionally the period of the what's the from idea something from will make which is have you use them and is it worth to play around with them for example 4 text revision did a lot of interesting stuff as well as agreeing to order it is better to just develop something going on thank you for I don't have a lot of experience experience with that system so just mentioned weeks avoid the types a small debate will 5 of tackled using and that's what I language processing libraries from Microsoft in that case and my experience is that these models are already very well trained for various he proposes and if you have the time and money and your is your core business but it's better for you to train neural models and have people specialized if it said that some sightseeing reason of main being near the bulk of the money provided this already viewed and moments son more than enough for you but for a specific things do we don't have an already trained things that is my business leaders of thinking questions greater than that by 16 hour we get them I told otherwise tomorrow at 9 AM there will be a fantastic he by putting them in the right will be talking about the use of Python and let the snake and I'm told it's a great told so I encourage you all to them that thank the speaker is