.Net Foundation ML.NET, 1 Jun 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: ML.NET End-to-End 2: Building a Web API

Description

Continuing the end-to-end ML.NET project by creating a Web API that loads the ML.NET model and gives a prediction.

First video in series - https://www.youtube.com/watch?v=VVekyUMEKsI

Code - https://github.com/jwood803/MLNet-EndToEnd/tree/master/WineAPI

Contact:
Twitter: https://twitter.com/JWood/
Blog: https://jonwood.co/

A

Hey everyone, if you remember from our first video in this in Medinah indian series, we created our resources and they recreated this console project where we used a Madonna to get data from the database to create our pipeline and our Madonna model, and then we also saved that model to Azure blob storage. So now what I want to do in this video is I want to actually utilize the the model that we created here and create a Web API.

A

So we can predict from our model and to do that I'm, a still in the same solution here, I'm going to create another project, there's going to be a web project here, let's call it one API.

A

Here, I'll just do an API project: let's do an API project like a mess with HTTP. Here, no authentication.

A

And now we have our project here: remove this values, controller and I'm a created, nearly controller I'll, do an empty API controller and a call clicked it. This is the control we're going to use to call the predict method of our mo data model.

A

Before we continue I'm gonna, add some new get product packages here and I'll do mo net, so we can load the model and and what an AdMob so load. The in my data model I'm using version 1 for this and I'm gonna get a couple of Microsoft extensions for configuration, because I'm still going to use that that JSON file from my configurations here so I'll get the main configuration 1 and then I'm get to jason 1.

A

Very good all right! So let's go back to this controller here controller here, and this just like I said before this is going to be used to load in the model and it predicts, and since we are going to be sending in data and to this API, to use that data to predict from it. We can't use an HTTP GET call because read to send us some data, so standard we're doing HTTP, POST method here and I'm.

A

Gonna put it as a sink because we're gonna mix an async all within it and I'm a return to float and that's going to return the divine quality. A number that will give that from our prediction from the model opposed and to send in the data that we need to use to predict predict from I'm gonna use the from body, because it's going to get it from the body of our request and it's going to be the wand data call my data yeah.

A

So we don't have reference to this, because this I have reference to this, because this is in this of this console project. So what I can do instead? Since that references this this model from from both these projects, I can actually create.

A

Another project we're kind of hold all of our our common code, and that is going to be do it done a core class library I just in wand, common good. What I can do here, I'm good? What I can do here is I, can just move this up in here or remove this class that I created and go and to use this project within these others, too. I have to add a reference to it.

A

And I'll do the same for previous one I need to come back here and fix this one go.

A

So using one common.

A

I'll delete this one too, because it was conflicting with it. So nice should.

A

A

And then in our control right here, I think I forgot to do, but I'll move this up here.

A

This still has the main space of lawn regression, so I'm going to change that the lawn common to match the name of the project now I move this in use, wand data and now I need to update this using one common there we go so that fixes that and now there's only one place where I have this I can reference this wand data first thing: I need to do I need of since I'm going to get it from the.

A

If your blob storage I need a fix out my configuration kind of similar to what we did up here before we built a configuration from a JSON file. I'm gonna do the same thing in here, but this is a little bit different since it's a Web API project, if your project so I'm startup here I have this configuration and then this configure services is where I can I can set this so I'm actually going to take some code. Here we have app settings, there's our app settings and I'm going to replace this.

A

With our blob connection string bring your namespace there you go now. I just need to set that configuration equal to the build.

A

And made this ascetical need to singleton, so it loads only once when the project starts.

A

And yeah, that's just said for that, and what we can do to have access it within our controller method is I can create a constructor here. I can tell it to pass in I configuration the configuration, bring it and I can make a private field.

A

And they just said it didn't a constructor.

A

I'm gonna create a couple of other now class level fields here and create a read-only string. That's gonna hold our path to our model when we downloaded from Missouri blob, storage and I'm going to create an email context.

A

We're good anyone can do here. Cuz I can set the context they go to a new in the context and for the model path. I'm gonna get that from.

A

This new system, I have combined and I'm gonna do is I'm gonna, set it into my My Documents folder, and to do that, I can do it, an environment that get folder path and then near environment dad and then near environment. Dad special folder. My documents never give it a name of one ship.

A

It's gonna be just the same name as what we have in our blob storage, I'm back to our post method. Here, what I want to do first is check if we already have the file downloaded. If you already have it downloaded, there's no need to download it again. So I can do this. If it's, the file does not exist and I'll just pass in the model path from here since it doesn't exist, I can access it from blob storage. You do is kind of do similar to what we did before.

A

When we access to our blob storage, you create a storage account and that new get package as well, and that is knows your storage install that.

A

A

From here the storage account equals cloud storage counts, bringing the namespace and the parse out the connection string, which is figuration Bob next rain. I. Think that's what I named it check here.

A

Yeah block connection string, I get a client from its.

A

Kraid cloud dog client from there I can get access to the container container reference attend your name's gonna be model.

A

And the blog equals container that give block block references. That's going to be our name of the model file, which is one the name of the model file, which is when Dad zip.

A

And what that does I have that the block reference from the bulb account I can await blog download to file I'm just going to use that model path and I'm the tailor to always create a new one when it downloads.

A

So if we don't have the model.

A

Downloaded yet that's what we're going to be here, just download it from Azure storage and what we're just downloaded and get access to that file and create a file stream here. So now, I would open read with that model path.

A

Actually, let me create an instance of this model here, it's going to be in it's just a transformer set the model, and so that model equals to our ml context, model dot load and it takes in an out variable of our.

A

Data view schema so I'm just do get up here. Out of you schema schema just past that in that's now a row now we have to model load it locally. We can start and by creating the prediction engine, and that comes in from the context that model that create prediction engine- and this is just like we've seen several times before and m/l done- that we have to give at the info schema and the prediction schema which we don't have created yet and then we'll just pass in the model.

A

Name and what we can do, my one comment: I'll just add a new class come on prediction.

A

One item here about predicted wine quality and we give it the call name of score and, as you need to install and when I met in here too so I can use. That attribute.

A

Good bring that in all right and then in the controller now to seize up, because we ought to have that namespace brought in and now we have our prediction. Engine I can create our prediction function or create our prediction and do prediction. Engine I have to predict and we'll just pass in the data we get here.

A

And then we just return prediction: I forget the quality.

A

Go now: let's, then we can try to run this and make sure API is set of startup project and get.

A

You guys start so it tries to go to that values. Controller.

A

Before- and this is a common mistake- I make once again I need to tell this app settings to copy if newer.

A

Okay, like I, was saying before it tries to go to this values controller when you start up, but that's okay and it can be found because we deleted that what we can do to test this out. Is you slip to my postman.

A

Yeah we got all right. We can use this URL here.

A

Yeah so not predict yeah, it's so not predict controller name and C headers. You want to give the content type of application jason, because we're gonna send him some jason and give it a raw body of adjacent object here, and these are just the different values that we can.

A

They were sending it so by the type is quiet and the different attributes of the wand to determine the quality and hits in and see if it works.

A

Now and I didn't because I'm told you a gate requesting we don't have a get request there we go this and we got a predicted quality of of none so seems like this is a pretty good one. That's that's it kind of a basic way to create an API the loads in a mode on that model and gives us a prediction, and so our next video is we're.

A

Gonna, create kind of a a small web application that will use this API and it will just kind of give it a customer-facing way to use the model that we built so I'll see you all, then thanks thanks.