Follow Mark on LinkedIn Follow @marknca on Twitter Follow marknca on YouTube
marknca

AWS Machine Learning

Road to AWS re:Invent 2019


Join the discussion on LinkedIn

Tweet about this episode

Full machine generated transcript follows

Here is the slide that I used during the live stream.

All right everybody. There we go. How's it going today? Thank you for joining. My name is Mark. Dekoven. You can hit me up on social at Market in the ca obviously you've already figured that out cuz you're here on the stream on LinkedIn or on Twitter. This is the final one in a series of streams that I've been doing leading up to AWS reinvent a final one simply because AWS reinvent is 4 days away.

Or that's crazy schedule has been all over the map as it gets busier and busier and busier busy or leading up to reinvent. I'm So today we're going to cover the basics of AWS machine learning. We're going to look at a few of the services. Are we going to see how the service are structured and then we are going to see a couple examples and I'll give you a few tips for reinvent if you're coming in a few tips, if you're not coming actually am because there's a lot you can do even if you aren't in Vegas next week for a divorce rate.

So let's dive in right away. Let me share a out Chrome this one Ark so we've done a bunch of these already. If you want a pop the link already in the LinkedIn life comments, basically on my sightmark and. CA the main pinned post is this one on reinvent and you can see all the previous song.

If you scroll down are you going to see my guide to reinvent? You see the Practical security session got it is also the toxin giving at reinvent and then I'm all the different stream so you can see the last time we did that was on computer then go there.

You can also see the transcript and actually the transcript the reason why I want to show this the transcript to something that you were going to see the service that we can use to do that with audio or video today. So I think that makes a lot of sense and hopefully it does so bottom line check out that link if you want to see the previous today with these this slide.

So if you have any questions as we're going as we're going through hit me up here on LinkedIn or on Twitter and I will respond in kind of find his best to answer questions as we're going as opposed to killing them up at the end. So I'm just just picking in the chat there and and we'll dive in you know, I've already seen some some great questions.

Some folks already this morning. I'm trying to answer as we go or as we're leaving out. But here we are we're looking at at all of the services that AWS labels as machine learning Services. Now, there's one that's not on here that actually been retired unless you already have an account in are using it you can no longer use it and that was actually called wait for it AWS machine learning horrible name.

I don't know why they picked it in the beginning. I mean I know why they picked it because it made perfect sense. But that service is actually being retired eye. If you have used it on an account already, you can still access it but new account can't so I left it off there because as you can see there is already a crazy amount of machine learning services.

So we're going to start from the transactional services. Now, there's the most of these Services know these are just my categories as I find. This is the best way to think about machine learning on AWS. Now if you want to know my credentials on I am a part of the IEEE International Electronics Electrical engineers Association as part of that.

I've been a part of the community for computational intelligence the last 12 or 13 years or so. I've really enjoyed our machine learning and AI subsets, you know, I'm all these different areas. It's been a passionate and a hobby of mine for a while. So I love that it's being really big ass or mainstream attention now so I know how what's going on in do a lot of this stuff and today we're just going to have to go over and show you how you can start to get started.

So you're most the time we're going to get started with these transactional services in the reason why I called them transactions because you don't need to build up any models. You don't need to have any of the underlying stuff that makes any of this work. They are simply you give the service something the service returns results to you.

So we've got transcribe and translate transcriber Duffy battles go one-by-one. It's easier. So transcribe takes audio gives you back text translate takes one language gives you back. Another comprehend a does natural language processing which is a subset of machine learning where it looks at how language is structured. So you give Amazon comprehend set of text and it's going to give you back the structure of that text is going to identify verbs nouns and it's going to give you the semantics of that text is going to give you topic modeling and I'm actually did a course on this service for plural sight and it's really really interesting.

But if you're not a language nerd if you're not trying to break down what text means it's not super useful, but it's still very very interesting Amazon textract. Is actually a good name. It's a it's a portmanteau of text and extract. So you give this an image of a PDF and it's going to actually pull out text very very cool.

I'm inly I see your comment there on pen testing machine learning for Automation and there is actually some really cool stuff. You can do I'm tied to machine learning and somebody Services May pay off for you and but specifically for pentesting want to be absolute best areas of study.

Lee is fuzzy logic so fuzzy logic you're going to see that in fuzzers which basically goes. Hey algorithm figure out a whole bunch of weird inputs and throw them against something and see what happens and that's actually a subset of machine learning and which is under the bigger umbrella of AI so text back.

Like I said give you an image to PDF and give you back some some some text information. Am I did a video in that launched actually a link that in the chat cuz that was super useful and actually very very Cool at extract I cannot spell text drive to save my life, which is ridiculous.

And but here you go. So this is that video Fort extract Amazon extract while we're doing that. I just wanted to call out something that was really interesting and I miss belt extract in the thing, but you get the deal if you're wondering why some of these are Amazon some of the rate of US Corey Quinn actually called us out the other day cuz a lot of people know it Amazon Services tend to be able to be used on the Rhone AWS prefix services mean that you need that they're building block, right so you can see all these transactions are Amazon.

So textract give it an image to PDF get back text forecast to give it a bunch of data and it's going to do predictions as soon as you extrapolations out. And so this is great. If you have a bunch of like sales information, you can do it and say hey, here's all the sales transactions for 2019.

What is 20/20 look like same with actually interesting a back to Lee's Comment around pen testing in machine learning forecast is really cool. If you shove a whole bunch of network transactional data into it as far as source and destination to see where it's going to forecast Network patterns in the future.

Very cool could probably do a whole nother stream on that Lex is actually the basis for Alexa. It is a chatbot service recognition is the one that you've seen and unfortunately in the news quite a bit, it's the facial-recognition that's not just facial recognition. It's actually an image processing and image recognition.

We're going to dive into that a little bit. Amazon personalized is a personalization recommendation engine so you can give it a whole bunch of buying habits and it's going to say people who bought this also bought that so it powers on amazon.com recommendation system and which is kind of cool and then Amazon Polly is the last one.

I think Polly want a cracker you give it a text it gives you voice and you can pick which invoice you want which is kind of cool. So those are the transactional services. We're going to dive into a couple of those in a second and because they're really great for demos and we'll we'll take up most of time today I think of doing those demos, but then I wanted to call out some of the ones that you're probably not ready for yet.

And I think it's good just to know that they exist and most of time if you're just started with machine learning you are going to be diving into a lease transaction services, but there are these building blocks services. So there is an Ami for easy to that's an Amazon machine image.

And that is preloaded with a bunch of software that helps you run machine learning models. So if you're going to try to run your own models and filled out your own machine Learning System start with the AWS deep learning. Am I that is managed by AWS? They keep the versions of current and basically wants you to play that server.

It's pre-installed with a whole bunch of stuff, which is great. One of those things is pre-installed with which is Apache mxnet, which is a machine learning library that lets you do a whole bunch of cool stuff on Amazon or AWS actually maintains M.. They're the leaders on that project and then they also keep a forca tensorflow as well these two things.

If you don't know what they are. Don't worry about it. They are the libraries that power some of those transactional services. But if you want to go down a level and get into the nuts-and-bolts, like I said, these are building blocks Amazon elastic inference is a GPU acceleration service.

And then what that means a GP was a graphics Processing Unit your computer that you were Turn on right now has a CPU and a GPU CPU is the central processing unit. It's a generic thinking machine. Right? It's a generic processor and where it'll just go through functions in process computations GPU is designed around Graphics, but it turns out why building all these Graphics processors over the years.

We actually optimize them for a different type of math that math wind up very very nicely with machine learning and Amazon elastic inference lets you take a graphics processing unit and attach it to your normal server's not to run Graphics not to run games are things like happened to do the type of mask that we need to do in machine learning.

So it'll basically it lowers the cost of your machine learning and makes it faster so building blocks really cool if you need to wear if you want to go deeper than the transactional you go to these partially managed, which is essentially one service at Amazon sagemaker. Very very cool.

It's a level of Set of containers that you set up a jupyter notebook, which is a data science. Notebook is basically a mix of python code and data and some other code can be in there too. But normally developed Python and then sagemaker takes care of all the infrastructure to spin up a container Fleet to run through all of this stuff and give you these model.

So if you have something where you're doing design work and go like I don't know what the best shape of a phone is this by the way, though the new folding phone super cool. Have you can run it through sagemaker and run like a thousand concurrent models and it will come back and say hey, here's the crazy new shape that the computer came up with.

So very cool. If your kind of getting past the transaction services in want to do more sagemaker ground truth is actually a labeling service where he uses both machine learning and humans to label data. So it's setting a ground truth. And the interesting thing is anytime you're using machine learning if you're in the building blocks are partially managed.

You have to create A betasept to train the machine to learn on basically, right you're creating a curriculum for the learning. So you need a whole bunch of images that are labeled in order to make an image processing service are transactional image service, right? So sagemaker ground-truth helps you and farm that work out and say hey, what are all these pictures of or can you make sure that all these numbers are credit card numbers are format of the crate with don't use credit card numbers, but that was the first thing security guy first thing that came to my mind.

I want all these phone numbers format of the same way or I want all these whatever cleaned up and smooth ground troops the way to go and then finally have to tutorial Services. Now, these are really really clever Services AWS built these to teach you different machine learning techniques and the the way they did that was attached it to Hardware.

So this is actually a deep lens camera so you can take check that out essentially a normal webcam with a Little computer attached to it. Right? So it's super clunky, but the interesting thing is that this actually runs machine learning model on it. So we'll see that a second when we do a demo recognition, but this is a way where you can sit this up and it will actually recognize things and label parts of the images.

They like, this is Mark. This is Mark to visit humans had right? This is a webcam that kind of stuff so they did this and they also did deep race which is a car the remote control car where you don't remotely control it controls itself now and this was a simple a training model that they built to show you how to build models for image recognition and deep racer is all about Oh, geez, I can't unblock you on the term, but it's another way of teaching you a different type of.

Machine learning course. I'm going to I got a guy to Google that. It is come on. Yes reinforcement learning. Sorry reinforcement learning so that basically tells you you give the Deep racer model. I said a parameters and say I want you to Value turning you don't being faster at 2 the next Waypoint versus being more accurate.

And again, it's a fun way to do it. In fact, it's so fun that there is the AWS deep racer League finals coming up next week in Las Vegas very very cool. So that's an overview of how the machine learning Services breakdown for the vast majority of us. We are going to stick in the transactional services.

And the reason for that is that we don't don't worry about the model. We don't have to worry about the training data and we are just getting results. So let's flip over and see how that works right now. So we are in Amazon rekognition. Now, this is the one like I said where this is actually capable of doing facial recognition.

So it's capable of saying I see Mark in this photo write I see a lie in this photo I see you know mean in this photo is capable of doing that if you give it the appropriate photos with the labels, but what you can see right out of the gate is with their sample image and you can see the results on the right hand side here.

It's 98% at 8% Confident that is found in automobile that there's Transportation here. That's a car if you actually go on the little boxes. It will tell you where the a I thought I found something. So here it thinks it's found a skateboard. And of course it did but that's just a default training image or default demo image.

Where is she going to give it to me images from the from unsplash. So I found a couple of these images. So we're going to use this one. I was a great photo from Andrew. I'm petrich Ave. And where it's got a singer and out into an audience and a few other things.

We're going to take this photo and drop it into the demo. So if we drag this in here, it's going to take a second. And let's see what it comes up with. So now it's process these results and it gives us our results on the side here as well as a confidence rating right to know the confidence rating.

The 99.6% is how confident the service is that this is what actually in the image you'll notice. It's not a hundred percent because it's not sure but based on the modeling. This is a really high confidence level and that's actually part of the challenges around the facial recognition and why the service there was a lot of an interview me wrong with basil recognition is an absolutely critical thing that we need to discuss about its use-by government by law enforcement.

I'm actually cover that if you go to my website Mark end. CA my regular Tech call him where I do here in Canada for the IBEW Local radio program that I work with as well as the local news outlets. It's absolutely important. One of the critical thing is this confidence level you absolutely need to understand your sort of risk tolerance or your air.

Tolerance here if we just want to figure out like a what kind of in this image. This is great. This is fine anything over, you know, 90 something or 85. Something is normally good enough. If you're trying to actually recognize somebody and then, you know, take legal action. There should be 99 or 98 or higher this actually guidelines published by recognition on it and put back to the back to the issue at hand here.

So this is about a person great and this is found a crowd. So if we look it's identified at didn't box it but it's identified the background as a crowd then all these little people here here here here here and this is pretty accurate, right? It's thinks there's a building in here which we know there is on the edge and in the background and skin is kind of creepy, but it's found skin.

And if we go down you'll see actually this is where it gets really interesting 81% confident that is has a festival right? So it's got the attributes of a festival and it looks very festive Lee the tents in the background is it Are people outside there's the inevitable smoke and fog and it's 75.7% confident that this is actually a concert again.

I believe it's a concert right? It's 67.1% confident that it's a rock concert that that's really really interesting because it's making a I mean, I don't want to say judgment but the easiest way to do it to invade is that it's a judgment based on other images that it's scanned and it says this one is very similar to ones that knows our rock concerts, but it's not super sure right to 67% You know, it's it's a but this is recognition.

So we take another photo. Let's take another photo here and show you the difference of this one's a little more interesting. This is from clerical canola or kucova again from unsplash and this is you know a person in silhouette using a smartphone to take a picture of blinds with this really cool gradient, but there's also a building in the screen on the On the smartphone.

So if we drag this image over, let's see what it comes up with now. Analyzing analyzing and here we go. So it says there's a person. Yes. Okay human great. I hope if there's a person there's a human photographer. So it assumes because this person is taking a photo it's made that connection that it's a photographer photography as a subject but it's less confident Photo & Electronics if you scroll down there's actually a camera at 55.1% so far less results because there's less going on and also lower confidence ratings here because it's not quite sure right in this was a hard hard photo know the last one I want to do is this one so very similar again somebody taking a photo but it should be a higher confidence level because we've got straight onto the person they're not in silhouette.

We've got a actual traditional camera as opposed to a smartphone with those drag this one over and see what's going on image can't be larger than come on. Let me squash this image down real quick and but you see as we're doing this you see the difference, right is that Giving it to a transactional service.

We're giving it a photo. It's giving us back results. We're doing it through the web browser, but we could very easily do this through at Michigan have to shrink this and we could very easily do this through the API and write the API is a very simple call you simply send the photo up and it sends you back a block of results.

So here we've done that with that photo and you'll see we have a person human helmet so we know this is wrong. It's 98.2% confident, but this is actually incorrect. That's not a helmet. That's a hat sewing machine learning is not always perfect. Even if it does have this really really high confidence rating, right? This is not correct, but it's kind of clothing apparel Electronics camera photographer photography, right? So it's found very similar things finger face in portrait.

So it's very spot on except for the helmet because the kids these days you don't The brims anymore, right? So it's it to think this is a helmet instead of a hat. So even with high confidence levels, you need to take that into account if you're using this in a project because if you said at 98% that's all and I'm done.

It's still not a guarantee but this is recognition spelled horribly fantastically useful. This is just the straight-up use of saying what's in an image and you can use it to moderate images to make sure there's nothing sensitive being shown and sensitive as Define but you so you can say you know what? I don't want any humans in my images.

So if you're running a service for Landscapes, you could run every image through recognition and say I don't want any people you could do facial analysis you could do celebrity recognition, which is really interesting have face comparison and then it ties into into text. It also does video which is essentially just a series of images.

So it should be shocking to anybody that video is there so that's recognition. That's one of our transactional level services and that's where we give it an image and it gives us back results. So take a look at something else something different. So this is Amazon poly Amazon poly very very simple and straightforward.

You give it text it gives you back sound now, I forgot to pre-install the system audio driver for my streaming software. So you guys won't be able to hear this but you can see you can actually download the MP3. So if you just hit this listen to speech hi, my name is Joey.

I will read any text you type here. Just the shotgun like And what's the weather in Old School? My name is Joey. I will read any text you type here. She might have her that I'm into an MP3. The interesting thing here is a poly and that was very robotic but Polly really has some good human utterances in to make it seem like a better than just a robot if you guys are old enough, you might remember Dr.

Sbaitso from way back in the day when soundblaster first came out very very crazy. But text to speech is Handy if you're doing audio stuff. So this is part of the Alexa back and stuff, but you can use it in your own. So if you're creating like a kiosk display, if you have a project where you want to provide accessibility AWS themselves actually use this on their block if you go to the AWS blog At which is insanely busy this week, by the way, because there's just so much going on.

But if you click here, this is a fantastic one by Jeff Barr. He did 15 years of blogging for AWS, which is awesome. Just such a great guy. Every blog post. They have has this little playlist which is actually Voice by Polly. So this will read the blog post to you and they use this through poly as a service very simple, very straightforward.

You might find uses for it on your own at which is great to the Pali think Polly want a cracker and and you can pick your voice is right there. But a number of female number of male very very cool stuff and again not in the transactional set of services and let's look here.

We got a couple more minutes before we get our thing. Let's look at transcribe. So transcribe now is the opposite of poly you give it a voice it's going to give you text. So I use this for I'm actually is Google's version of for the most part because it had delivered.

Sometiming things that I needed but I have a script that uses this as well to build out my blog. So when you saw this summary of when I did a terrorist compute machine-generated transcription false, this is all generated by all the texts and highlighting was done by web service in that case.

It was Google's a text or speech-to-text and I gave it the video just like today after I finish this I'll give it the recording of This stream and it will spit out this text. So that's very very useful and transcribe actually takes legitimate a screaming. So if I start clicking on this now, you should start seeing this pop through if it will share my microphone now, you'll see there's some errors in here at but it is very very good for sweet by find its run 95% effective.

So if you're looking to create content, this can be really really good if people aren't happy that we're comfortable writing you can just get them to hold their smartphone record themselves either on a video or a I was just straight voice send it to transcribe and you are going to be able to get the text back and then you can lightly edit the text and you're Off to the Races.

I just stopped. It was an Amazon transcribe very very simple. You can queue up jobs and all these Services as well and if you have more If you have more volume ASL for transcribe, this is just the demo. You're just seeing the Demos in the browser. They all have an API behind them.

Transcribe can do real-time like you just saw or you can just send it a file and which is great. It can be a little beyond what it wants is far as format, but insanely have how useful very very useful in a corporate setting to try to get more content out great white people are normally weigh more comfortable talking and then they are writing stuff out and then you can just use an Editor to to clean this up last one in the transactional service.

I want to show you I'm not going to go through this but I just wanted to give you an idea of what Lex is capable of and this is just a straight-up intro sort of demo for Lex. So are you get to create your own bot and it sent you what you do is you map out the flow and Lex can do it in text is that they respond to other entities based on an intense.

So in this case, you can see there's a workflow for booking a hotel. So the user says, you know, I'd like to book a hotel and then Do that like typing it or speaking it because we know if they speak it we can try and we can transcribe it into text automatically.

That's what Lex doesn't the background and Lex then replies in the light blue here shirt which city right and then the user response to that other and saying, okay. Will New York. The New York is now an input for lacks to do some searching at but Lex wants more info.

So it's as well. What date do you want to check in? And that's the second piece they need to do that search and then they fulfil that search by going. Okay. We are going to look for New York City November 30th, and then we'll go through you can do the same thing that you see there with you.

If you order flowers, if you schedule an appointment, then this can work on purely as a chatbot online so you can hook it up to like Facebook Messenger. You could have it running on Twitter. You can do it on a number of these services are just wrong your website or you can have it running on a phone.

You can have an answer and use it purely in audio or a mix thereof. That's Lex it manages those transactions and it's actually think like Amazon Alexa Lex lets you build it in your own stuff. And which is very very cool. Even though it's not Alexis the tech that powers Alexa don't know where we're running a little long hair, but that's okay.

I just want to show you this last thing here before I give you some tips and tricks for reinvent next week 8 of us deep racer. You do not actually need the car to do the bracer. You can get a car and I think they're two or three hundred bucks on amazon.com us to get that didn't get one shipped in a very very cool, but you can actually do it online and essentially what it is is this flow so it's designed to help teach you reinforcement learning and you create a model they have a default set model for you you train that model you evaluate it and then you kind of adjust and model is a really fancy term in this case for essentially giving it a score.

So you say try to change these variables like turn the wheels a little bit turn the wheels a little bit or like accelerator break whatever the case maybe I am, but you don't tell it those commands. What you do is you give it a goal as so if you say if you are 5° off from your next Target on the track, that's a plus 5 or + 10 cuz that's really good if you're 25 degrees away from your goal.

That's a zero, if you're 40°, that's a -5 so your model gives it a score for where it's sitting on the track right now and its speed and its time over the course right to give an idea of saying hey, you are this close to your goal or overall and to the next goal.

Here's your score based on that reinforcement learning then decides to turn the wheels or accelerate or brake to get another score in the next slice. So reinforce this model deepracer basically is a set of constant evaluations where you set up the scoring system and then the model makes choices based on that score to go tape.

I did worse. I'm going to turn the wheels more I did better. I'm going to make it less of a change next time to see if that continues to increase my score the goal of this system is to increase the score and finish the track and it's a really clever way of getting you to do this.

So essentially when you click Start you have to create a bunch of our resources. These do cost money. You're going to go through some of their tutorials at and then you create a model and you can see if you can take about an hour and 15 to get a up and running.

But the great news is there's a kind of great information in here to kind of walk you along how to do this. I'm making me very very cool. And if you have built model, you can actually race them at the end of your summit still course. We're now done for Summit season 2019.

There is one more chance at reinvent and then the finals are reinvent and but it's a really fascinating way so you can see here actually the people eraser league and so you can join the warm-up leave the championship cop is what's coming up. So you get one more chance on Monday and then it's the big Championship through during the week a very very cool.

If you look at the standings last time, I checked a few of my friends one some of these which was very very cool, but that you can go through and see all these results and I thought this was a really fantastic way to teach you about that cause Set the reinforcement learning because it's kind of weird to not give instructions to something.

You just have to say do something try something and I'll score you in based on those scores try something different and that's essentially reinforcement learning. So we covered a ton of crazy stuff today. I'm going to review that in 2 seconds, but I just wanted to say if you are interested this on the technical side and here is a great book.

I'm going to drop the link. I dropped a link in the in a LinkedIn chat computational intelligence by Everhart and she is very much a textbook. Don't get me wrong. It is a textbook if you are in a machine-learning know, it's VTEC. Is it very very cool. It covers more than just machine learning.

It's basically the introductory Concepts. So this would be an intro Class A to computational intelligence or a I definitely worth reading on it's one of my favourites. I say that as a nerd and I realize I say that out just how nerdy that sounds bad. But today we covered a lot.

We looked at some of these services in depth but essentially of transactional services you give it something you get results back. You got a bunch of building blocks. If you want to dive deeper most of you if you do want to dive deeper you just going to go to the partially manage stuff with sagemaker, but if not, if you want to play around with some real-world stuff, that's at Oriole pieces are cool.

So deep cleanse, that's what this is or deep racer, which is the car. They think you could be a lot of fun to to work within a great way to learn different types of machine learning. I'm in a practical way, right? So that's an overview of what Course, this is all going to be there going to be a ton more stuff out of next week at reinvent.

So if you are at reinvent, I would strongly recommend that if you look on my main reinvent page here there is this Ultimate Guide to let me open that out as I write this every year it gets kind of lengthy but in a good way as to the kind folks over a cloud Guru where I'm also an instructor or are they are nice enough to Hostess and but if you'll see this is the table of contents alone, and I'm just all the different things to reinvent to get you on the ground.

The biggest thing to know at this point is if you're going to survive bring sneakers bring sneakers in your backpack everyday have water a battery for your phone lip balm, maybe eye drops and some protein or granola bars and their long days they're fun days, but that's absolutely critical.

And if you don't have good socks and shoes forget fashion your folks just go straight up for the sneakers because you're just going to eat. So two years ago. I didn't leave the Venetian and I walked 15 to 12 to 15 km of day and I never left the one property and remain goes across seven different properties, which is absolutely bananas.

So I'll check that out again. Here's the the link and let me drop that in the chat all things reinvent. This is been a lot of fun. I really like doing this series. Thank you guys for jumping in and hanging out with me while we do these I think it's it's been beneficial hopefully for you guys.

I'm going to do more of these in 2020 ad there going to be a little more structure. They won't be leading up to an event as if you have suggestions drop them in the chat or hit me up online at Mark and CIA or by email me at Marquette.

CA happy to talk to you guys were through the see what you want. Cuz it obviously the audience really makes us work the Community Drive this but in the meantime, has it been a lot of fun. Thank you for joining me. If you have any questions, like I said, hit me up and if you're in Vegas next week to me on Twitter and we'll see where around me will catch a drink or beverage and but enjoy the show if you're not coming to Vegas.

So check it online. You can register for the live streams for online. You can watch the Keynotes online as well as are all the talks get pushed or almost all the talk to get pushed onto a tab uses YouTube channel live in about a week or two after the If you're not missing out on the course, I'll have a big summary of that on my site.

Thanks a lot. Have a great day. For those of you in the states. Enjoy your long weekend and Happy Thanksgiving and will talk to you all soon. Take care. All right everybody. There we go. How's it going today? Thank you for joining. My name is Mark. Dekoven. You can hit me up on social at Market in the ca obviously you've already figured that out cuz you're here on the stream on LinkedIn or on Twitter.

This is the final one in a series of streams that I've been doing leading up to AWS reinvent a final one simply because AWS reinvent is 4 days away. Or that's crazy schedule has been all over the map as it gets busier and busier and busier busy or leading up to reinvent.

I'm So today we're going to cover the basics of AWS machine learning. We're going to look at a few of the services. Are we going to see how the service are structured and then we are going to see a couple examples and I'll give you a few tips for reinvent if you're coming in a few tips, if you're not coming actually am because there's a lot you can do even if you aren't in Vegas next week for a divorce rate.

So let's dive in right away. Let me share a out Chrome this one Ark so we've done a bunch of these already. If you want a pop the link already in the LinkedIn life comments, basically on my sightmark and. CA the main pinned post is this one on reinvent and you can see all the previous song.

If you scroll down are you going to see my guide to reinvent? You see the Practical security session got it is also the toxin giving at reinvent and then I'm all the different stream so you can see the last time we did that was on computer then go there.

You can also see the transcript and actually the transcript the reason why I want to show this the transcript to something that you were going to see the service that we can use to do that with audio or video today. So I think that makes a lot of sense and hopefully it does so bottom line check out that link if you want to see the previous today with these this slide.

So if you have any questions as we're going as we're going through hit me up here on LinkedIn or on Twitter and I will respond in kind of find his best to answer questions as we're going as opposed to killing them up at the end. So I'm just just picking in the chat there and and we'll dive in you know, I've already seen some some great questions.

Some folks already this morning. I'm trying to answer as we go or as we're leaving out. But here we are we're looking at at all of the services that AWS labels as machine learning Services. Now, there's one that's not on here that actually been retired unless you already have an account in are using it you can no longer use it and that was actually called wait for it AWS machine learning horrible name.

I don't know why they picked it in the beginning. I mean I know why they picked it because it made perfect sense. But that service is actually being retired eye. If you have used it on an account already, you can still access it but new account can't so I left it off there because as you can see there is already a crazy amount of machine learning services.

So we're going to start from the transactional services. Now, there's the most of these Services know these are just my categories as I find. This is the best way to think about machine learning on AWS. Now if you want to know my credentials on I am a part of the IEEE International Electronics Electrical engineers Association as part of that.

I've been a part of the community for computational intelligence the last 12 or 13 years or so. I've really enjoyed our machine learning and AI subsets, you know, I'm all these different areas. It's been a passionate and a hobby of mine for a while. So I love that it's being really big ass or mainstream attention now so I know how what's going on in do a lot of this stuff and today we're just going to have to go over and show you how you can start to get started.

So you're most the time we're going to get started with these transactional services in the reason why I called them transactions because you don't need to build up any models. You don't need to have any of the underlying stuff that makes any of this work. They are simply you give the service something the service returns results to you.

So we've got transcribe and translate transcriber Duffy battles go one-by-one. It's easier. So transcribe takes audio gives you back text translate takes one language gives you back. Another comprehend a does natural language processing which is a subset of machine learning where it looks at how language is structured. So you give Amazon comprehend set of text and it's going to give you back the structure of that text is going to identify verbs nouns and it's going to give you the semantics of that text is going to give you topic modeling and I'm actually did a course on this service for plural sight and it's really really interesting.

But if you're not a language nerd if you're not trying to break down what text means it's not super useful, but it's still very very interesting Amazon textract. Is actually a good name. It's a it's a portmanteau of text and extract. So you give this an image of a PDF and it's going to actually pull out text very very cool.

I'm inly I see your comment there on pen testing machine learning for Automation and there is actually some really cool stuff. You can do I'm tied to machine learning and somebody Services May pay off for you and but specifically for pentesting want to be absolute best areas of study.

Lee is fuzzy logic so fuzzy logic you're going to see that in fuzzers which basically goes. Hey algorithm figure out a whole bunch of weird inputs and throw them against something and see what happens and that's actually a subset of machine learning and which is under the bigger umbrella of AI so text back.

Like I said give you an image to PDF and give you back some some some text information. Am I did a video in that launched actually a link that in the chat cuz that was super useful and actually very very Cool at extract I cannot spell text drive to save my life, which is ridiculous.

And but here you go. So this is that video Fort extract Amazon extract while we're doing that. I just wanted to call out something that was really interesting and I miss belt extract in the thing, but you get the deal if you're wondering why some of these are Amazon some of the rate of US Corey Quinn actually called us out the other day cuz a lot of people know it Amazon Services tend to be able to be used on the Rhone AWS prefix services mean that you need that they're building block, right so you can see all these transactions are Amazon.

So textract give it an image to PDF get back text forecast to give it a bunch of data and it's going to do predictions as soon as you extrapolations out. And so this is great. If you have a bunch of like sales information, you can do it and say hey, here's all the sales transactions for 2019.

What is 20/20 look like same with actually interesting a back to Lee's Comment around pen testing in machine learning forecast is really cool. If you shove a whole bunch of network transactional data into it as far as source and destination to see where it's going to forecast Network patterns in the future.

Very cool could probably do a whole nother stream on that Lex is actually the basis for Alexa. It is a chatbot service recognition is the one that you've seen and unfortunately in the news quite a bit, it's the facial-recognition that's not just facial recognition. It's actually an image processing and image recognition.

We're going to dive into that a little bit. Amazon personalized is a personalization recommendation engine so you can give it a whole bunch of buying habits and it's going to say people who bought this also bought that so it powers on amazon.com recommendation system and which is kind of cool and then Amazon Polly is the last one.

I think Polly want a cracker you give it a text it gives you voice and you can pick which invoice you want which is kind of cool. So those are the transactional services. We're going to dive into a couple of those in a second and because they're really great for demos and we'll we'll take up most of time today I think of doing those demos, but then I wanted to call out some of the ones that you're probably not ready for yet.

And I think it's good just to know that they exist and most of time if you're just started with machine learning you are going to be diving into a lease transaction services, but there are these building blocks services. So there is an Ami for easy to that's an Amazon machine image.

And that is preloaded with a bunch of software that helps you run machine learning models. So if you're going to try to run your own models and filled out your own machine Learning System start with the AWS deep learning. Am I that is managed by AWS? They keep the versions of current and basically wants you to play that server.

It's pre-installed with a whole bunch of stuff, which is great. One of those things is pre-installed with which is Apache mxnet, which is a machine learning library that lets you do a whole bunch of cool stuff on Amazon or AWS actually maintains M.. They're the leaders on that project and then they also keep a forca tensorflow as well these two things.

If you don't know what they are. Don't worry about it. They are the libraries that power some of those transactional services. But if you want to go down a level and get into the nuts-and-bolts, like I said, these are building blocks Amazon elastic inference is a GPU acceleration service.

And then what that means a GP was a graphics Processing Unit your computer that you were Turn on right now has a CPU and a GPU CPU is the central processing unit. It's a generic thinking machine. Right? It's a generic processor and where it'll just go through functions in process computations GPU is designed around Graphics, but it turns out why building all these Graphics processors over the years.

We actually optimize them for a different type of math that math wind up very very nicely with machine learning and Amazon elastic inference lets you take a graphics processing unit and attach it to your normal server's not to run Graphics not to run games are things like happened to do the type of mask that we need to do in machine learning.

So it'll basically it lowers the cost of your machine learning and makes it faster so building blocks really cool if you need to wear if you want to go deeper than the transactional you go to these partially managed, which is essentially one service at Amazon sagemaker. Very very cool.

It's a level of Set of containers that you set up a jupyter notebook, which is a data science. Notebook is basically a mix of python code and data and some other code can be in there too. But normally developed Python and then sagemaker takes care of all the infrastructure to spin up a container Fleet to run through all of this stuff and give you these model.

So if you have something where you're doing design work and go like I don't know what the best shape of a phone is this by the way, though the new folding phone super cool. Have you can run it through sagemaker and run like a thousand concurrent models and it will come back and say hey, here's the crazy new shape that the computer came up with.

So very cool. If your kind of getting past the transaction services in want to do more sagemaker ground truth is actually a labeling service where he uses both machine learning and humans to label data. So it's setting a ground truth. And the interesting thing is anytime you're using machine learning if you're in the building blocks are partially managed.

You have to create A betasept to train the machine to learn on basically, right you're creating a curriculum for the learning. So you need a whole bunch of images that are labeled in order to make an image processing service are transactional image service, right? So sagemaker ground-truth helps you and farm that work out and say hey, what are all these pictures of or can you make sure that all these numbers are credit card numbers are format of the crate with don't use credit card numbers, but that was the first thing security guy first thing that came to my mind.

I want all these phone numbers format of the same way or I want all these whatever cleaned up and smooth ground troops the way to go and then finally have to tutorial Services. Now, these are really really clever Services AWS built these to teach you different machine learning techniques and the the way they did that was attached it to Hardware.

So this is actually a deep lens camera so you can take check that out essentially a normal webcam with a Little computer attached to it. Right? So it's super clunky, but the interesting thing is that this actually runs machine learning model on it. So we'll see that a second when we do a demo recognition, but this is a way where you can sit this up and it will actually recognize things and label parts of the images.

They like, this is Mark. This is Mark to visit humans had right? This is a webcam that kind of stuff so they did this and they also did deep race which is a car the remote control car where you don't remotely control it controls itself now and this was a simple a training model that they built to show you how to build models for image recognition and deep racer is all about Oh, geez, I can't unblock you on the term, but it's another way of teaching you a different type of.

Machine learning course. I'm going to I got a guy to Google that. It is come on. Yes reinforcement learning. Sorry reinforcement learning so that basically tells you you give the Deep racer model. I said a parameters and say I want you to Value turning you don't being faster at 2 the next Waypoint versus being more accurate.

And again, it's a fun way to do it. In fact, it's so fun that there is the AWS deep racer League finals coming up next week in Las Vegas very very cool. So that's an overview of how the machine learning Services breakdown for the vast majority of us. We are going to stick in the transactional services.

And the reason for that is that we don't don't worry about the model. We don't have to worry about the training data and we are just getting results. So let's flip over and see how that works right now. So we are in Amazon rekognition. Now, this is the one like I said where this is actually capable of doing facial recognition.

So it's capable of saying I see Mark in this photo write I see a lie in this photo I see you know mean in this photo is capable of doing that if you give it the appropriate photos with the labels, but what you can see right out of the gate is with their sample image and you can see the results on the right hand side here.

It's 98% at 8% Confident that is found in automobile that there's Transportation here. That's a car if you actually go on the little boxes. It will tell you where the a I thought I found something. So here it thinks it's found a skateboard. And of course it did but that's just a default training image or default demo image.

Where is she going to give it to me images from the from unsplash. So I found a couple of these images. So we're going to use this one. I was a great photo from Andrew. I'm petrich Ave. And where it's got a singer and out into an audience and a few other things.

We're going to take this photo and drop it into the demo. So if we drag this in here, it's going to take a second. And let's see what it comes up with. So now it's process these results and it gives us our results on the side here as well as a confidence rating right to know the confidence rating.

The 99.6% is how confident the service is that this is what actually in the image you'll notice. It's not a hundred percent because it's not sure but based on the modeling. This is a really high confidence level and that's actually part of the challenges around the facial recognition and why the service there was a lot of an interview me wrong with basil recognition is an absolutely critical thing that we need to discuss about its use-by government by law enforcement.

I'm actually cover that if you go to my website Mark end. CA my regular Tech call him where I do here in Canada for the IBEW Local radio program that I work with as well as the local news outlets. It's absolutely important. One of the critical thing is this confidence level you absolutely need to understand your sort of risk tolerance or your air.

Tolerance here if we just want to figure out like a what kind of in this image. This is great. This is fine anything over, you know, 90 something or 85. Something is normally good enough. If you're trying to actually recognize somebody and then, you know, take legal action. There should be 99 or 98 or higher this actually guidelines published by recognition on it and put back to the back to the issue at hand here.

So this is about a person great and this is found a crowd. So if we look it's identified at didn't box it but it's identified the background as a crowd then all these little people here here here here here and this is pretty accurate, right? It's thinks there's a building in here which we know there is on the edge and in the background and skin is kind of creepy, but it's found skin.

And if we go down you'll see actually this is where it gets really interesting 81% confident that is has a festival right? So it's got the attributes of a festival and it looks very festive Lee the tents in the background is it Are people outside there's the inevitable smoke and fog and it's 75.7% confident that this is actually a concert again.

I believe it's a concert right? It's 67.1% confident that it's a rock concert that that's really really interesting because it's making a I mean, I don't want to say judgment but the easiest way to do it to invade is that it's a judgment based on other images that it's scanned and it says this one is very similar to ones that knows our rock concerts, but it's not super sure right to 67% You know, it's it's a but this is recognition.

So we take another photo. Let's take another photo here and show you the difference of this one's a little more interesting. This is from clerical canola or kucova again from unsplash and this is you know a person in silhouette using a smartphone to take a picture of blinds with this really cool gradient, but there's also a building in the screen on the On the smartphone.

So if we drag this image over, let's see what it comes up with now. Analyzing analyzing and here we go. So it says there's a person. Yes. Okay human great. I hope if there's a person there's a human photographer. So it assumes because this person is taking a photo it's made that connection that it's a photographer photography as a subject but it's less confident Photo & Electronics if you scroll down there's actually a camera at 55.1% so far less results because there's less going on and also lower confidence ratings here because it's not quite sure right in this was a hard hard photo know the last one I want to do is this one so very similar again somebody taking a photo but it should be a higher confidence level because we've got straight onto the person they're not in silhouette.

We've got a actual traditional camera as opposed to a smartphone with those drag this one over and see what's going on image can't be larger than come on. Let me squash this image down real quick and but you see as we're doing this you see the difference, right is that Giving it to a transactional service.

We're giving it a photo. It's giving us back results. We're doing it through the web browser, but we could very easily do this through at Michigan have to shrink this and we could very easily do this through the API and write the API is a very simple call you simply send the photo up and it sends you back a block of results.

So here we've done that with that photo and you'll see we have a person human helmet so we know this is wrong. It's 98.2% confident, but this is actually incorrect. That's not a helmet. That's a hat sewing machine learning is not always perfect. Even if it does have this really really high confidence rating, right? This is not correct, but it's kind of clothing apparel Electronics camera photographer photography, right? So it's found very similar things finger face in portrait.

So it's very spot on except for the helmet because the kids these days you don't The brims anymore, right? So it's it to think this is a helmet instead of a hat. So even with high confidence levels, you need to take that into account if you're using this in a project because if you said at 98% that's all and I'm done.

It's still not a guarantee but this is recognition spelled horribly fantastically useful. This is just the straight-up use of saying what's in an image and you can use it to moderate images to make sure there's nothing sensitive being shown and sensitive as Define but you so you can say you know what? I don't want any humans in my images.

So if you're running a service for Landscapes, you could run every image through recognition and say I don't want any people you could do facial analysis you could do celebrity recognition, which is really interesting have face comparison and then it ties into into text. It also does video which is essentially just a series of images.

So it should be shocking to anybody that video is there so that's recognition. That's one of our transactional level services and that's where we give it an image and it gives us back results. So take a look at something else something different. So this is Amazon poly Amazon poly very very simple and straightforward.

You give it text it gives you back sound now, I forgot to pre-install the system audio driver for my streaming software. So you guys won't be able to hear this but you can see you can actually download the MP3. So if you just hit this listen to speech hi, my name is Joey.

I will read any text you type here. Just the shotgun like And what's the weather in Old School? My name is Joey. I will read any text you type here. She might have her that I'm into an MP3. The interesting thing here is a poly and that was very robotic but Polly really has some good human utterances in to make it seem like a better than just a robot if you guys are old enough, you might remember Dr.

Sbaitso from way back in the day when soundblaster first came out very very crazy. But text to speech is Handy if you're doing audio stuff. So this is part of the Alexa back and stuff, but you can use it in your own. So if you're creating like a kiosk display, if you have a project where you want to provide accessibility AWS themselves actually use this on their block if you go to the AWS blog At which is insanely busy this week, by the way, because there's just so much going on.

But if you click here, this is a fantastic one by Jeff Barr. He did 15 years of blogging for AWS, which is awesome. Just such a great guy. Every blog post. They have has this little playlist which is actually Voice by Polly. So this will read the blog post to you and they use this through poly as a service very simple, very straightforward.

You might find uses for it on your own at which is great to the Pali think Polly want a cracker and and you can pick your voice is right there. But a number of female number of male very very cool stuff and again not in the transactional set of services and let's look here.

We got a couple more minutes before we get our thing. Let's look at transcribe. So transcribe now is the opposite of poly you give it a voice it's going to give you text. So I use this for I'm actually is Google's version of for the most part because it had delivered.

Sometiming things that I needed but I have a script that uses this as well to build out my blog. So when you saw this summary of when I did a terrorist compute machine-generated transcription false, this is all generated by all the texts and highlighting was done by web service in that case.

It was Google's a text or speech-to-text and I gave it the video just like today after I finish this I'll give it the recording of This stream and it will spit out this text. So that's very very useful and transcribe actually takes legitimate a screaming. So if I start clicking on this now, you should start seeing this pop through if it will share my microphone now, you'll see there's some errors in here at but it is very very good for sweet by find its run 95% effective.

So if you're looking to create content, this can be really really good if people aren't happy that we're comfortable writing you can just get them to hold their smartphone record themselves either on a video or a I was just straight voice send it to transcribe and you are going to be able to get the text back and then you can lightly edit the text and you're Off to the Races.

I just stopped. It was an Amazon transcribe very very simple. You can queue up jobs and all these Services as well and if you have more If you have more volume ASL for transcribe, this is just the demo. You're just seeing the Demos in the browser. They all have an API behind them.

Transcribe can do real-time like you just saw or you can just send it a file and which is great. It can be a little beyond what it wants is far as format, but insanely have how useful very very useful in a corporate setting to try to get more content out great white people are normally weigh more comfortable talking and then they are writing stuff out and then you can just use an Editor to to clean this up last one in the transactional service.

I want to show you I'm not going to go through this but I just wanted to give you an idea of what Lex is capable of and this is just a straight-up intro sort of demo for Lex. So are you get to create your own bot and it sent you what you do is you map out the flow and Lex can do it in text is that they respond to other entities based on an intense.

So in this case, you can see there's a workflow for booking a hotel. So the user says, you know, I'd like to book a hotel and then Do that like typing it or speaking it because we know if they speak it we can try and we can transcribe it into text automatically.

That's what Lex doesn't the background and Lex then replies in the light blue here shirt which city right and then the user response to that other and saying, okay. Will New York. The New York is now an input for lacks to do some searching at but Lex wants more info.

So it's as well. What date do you want to check in? And that's the second piece they need to do that search and then they fulfil that search by going. Okay. We are going to look for New York City November 30th, and then we'll go through you can do the same thing that you see there with you.

If you order flowers, if you schedule an appointment, then this can work on purely as a chatbot online so you can hook it up to like Facebook Messenger. You could have it running on Twitter. You can do it on a number of these services are just wrong your website or you can have it running on a phone.

You can have an answer and use it purely in audio or a mix thereof. That's Lex it manages those transactions and it's actually think like Amazon Alexa Lex lets you build it in your own stuff. And which is very very cool. Even though it's not Alexis the tech that powers Alexa don't know where we're running a little long hair, but that's okay.

I just want to show you this last thing here before I give you some tips and tricks for reinvent next week 8 of us deep racer. You do not actually need the car to do the bracer. You can get a car and I think they're two or three hundred bucks on amazon.com us to get that didn't get one shipped in a very very cool, but you can actually do it online and essentially what it is is this flow so it's designed to help teach you reinforcement learning and you create a model they have a default set model for you you train that model you evaluate it and then you kind of adjust and model is a really fancy term in this case for essentially giving it a score.

So you say try to change these variables like turn the wheels a little bit turn the wheels a little bit or like accelerator break whatever the case maybe I am, but you don't tell it those commands. What you do is you give it a goal as so if you say if you are 5° off from your next Target on the track, that's a plus 5 or + 10 cuz that's really good if you're 25 degrees away from your goal.

That's a zero, if you're 40°, that's a -5 so your model gives it a score for where it's sitting on the track right now and its speed and its time over the course right to give an idea of saying hey, you are this close to your goal or overall and to the next goal.

Here's your score based on that reinforcement learning then decides to turn the wheels or accelerate or brake to get another score in the next slice. So reinforce this model deepracer basically is a set of constant evaluations where you set up the scoring system and then the model makes choices based on that score to go tape.

I did worse. I'm going to turn the wheels more I did better. I'm going to make it less of a change next time to see if that continues to increase my score the goal of this system is to increase the score and finish the track and it's a really clever way of getting you to do this.

So essentially when you click Start you have to create a bunch of our resources. These do cost money. You're going to go through some of their tutorials at and then you create a model and you can see if you can take about an hour and 15 to get a up and running.

But the great news is there's a kind of great information in here to kind of walk you along how to do this. I'm making me very very cool. And if you have built model, you can actually race them at the end of your summit still course. We're now done for Summit season 2019.

There is one more chance at reinvent and then the finals are reinvent and but it's a really fascinating way so you can see here actually the people eraser league and so you can join the warm-up leave the championship cop is what's coming up. So you get one more chance on Monday and then it's the big Championship through during the week a very very cool.

If you look at the standings last time, I checked a few of my friends one some of these which was very very cool, but that you can go through and see all these results and I thought this was a really fantastic way to teach you about that cause Set the reinforcement learning because it's kind of weird to not give instructions to something.

You just have to say do something try something and I'll score you in based on those scores try something different and that's essentially reinforcement learning. So we covered a ton of crazy stuff today. I'm going to review that in 2 seconds, but I just wanted to say if you are interested this on the technical side and here is a great book.

I'm going to drop the link. I dropped a link in the in a LinkedIn chat computational intelligence by Everhart and she is very much a textbook. Don't get me wrong. It is a textbook if you are in a machine-learning know, it's VTEC. Is it very very cool. It covers more than just machine learning.

It's basically the introductory Concepts. So this would be an intro Class A to computational intelligence or a I definitely worth reading on it's one of my favourites. I say that as a nerd and I realize I say that out just how nerdy that sounds bad. But today we covered a lot.

We looked at some of these services in depth but essentially of transactional services you give it something you get results back. You got a bunch of building blocks. If you want to dive deeper most of you if you do want to dive deeper you just going to go to the partially manage stuff with sagemaker, but if not, if you want to play around with some real-world stuff, that's at Oriole pieces are cool.

So deep cleanse, that's what this is or deep racer, which is the car. They think you could be a lot of fun to to work within a great way to learn different types of machine learning. I'm in a practical way, right? So that's an overview of what Course, this is all going to be there going to be a ton more stuff out of next week at reinvent.

So if you are at reinvent, I would strongly recommend that if you look on my main reinvent page here there is this Ultimate Guide to let me open that out as I write this every year it gets kind of lengthy but in a good way as to the kind folks over a cloud Guru where I'm also an instructor or are they are nice enough to Hostess and but if you'll see this is the table of contents alone, and I'm just all the different things to reinvent to get you on the ground.

The biggest thing to know at this point is if you're going to survive bring sneakers bring sneakers in your backpack everyday have water a battery for your phone lip balm, maybe eye drops and some protein or granola bars and their long days they're fun days, but that's absolutely critical.

And if you don't have good socks and shoes forget fashion your folks just go straight up for the sneakers because you're just going to eat. So two years ago. I didn't leave the Venetian and I walked 15 to 12 to 15 km of day and I never left the one property and remain goes across seven different properties, which is absolutely bananas.

So I'll check that out again. Here's the the link and let me drop that in the chat all things reinvent. This is been a lot of fun. I really like doing this series. Thank you guys for jumping in and hanging out with me while we do these I think it's it's been beneficial hopefully for you guys.

I'm going to do more of these in 2020 ad there going to be a little more structure. They won't be leading up to an event as if you have suggestions drop them in the chat or hit me up online at Mark and CIA or by email me at Marquette.

CA happy to talk to you guys were through the see what you want. Cuz it obviously the audience really makes us work the Community Drive this but in the meantime, has it been a lot of fun. Thank you for joining me. If you have any questions, like I said, hit me up and if you're in Vegas next week to me on Twitter and we'll see where around me will catch a drink or beverage and but enjoy the show if you're not coming to Vegas.

So check it online. You can register for the live streams for online. You can watch the Keynotes online as well as are all the talks get pushed or almost all the talk to get pushed onto a tab uses YouTube channel live in about a week or two after the If you're not missing out on the course, I'll have a big summary of that on my site.

Thanks a lot. Have a great day. For those of you in the states. Enjoy your long weekend and Happy Thanksgiving and will talk to you all soon. Take care.