Okay, now it's time to create our first machine learning model. In this case, we will create an image recognition model that is able to identify different class of flowers. We will use AutoML in this case, AutoML exist under Vertex AI. So, first of all, we need to be sure that we have the Vertex AI API enable. Under the dashboard, you will need to check that you have the API enabled. In this case, I have it enabled, if it was not enabled you need to enable first the API. Once the API is enabled, next step will be create our first managed data set. In this case, we will create a managed data set that will contain a lot of images of flowers because we want to create an image recognition model that will be an image classification. Single label model that will predict the class that belongs to a specific flower based on one picture that we pass into the model. You can see here that we can create manage data sets of different kinds of problems that we want to solve for image models, text models, tabular data or video models. In this case, I select image classification, single label. And I will create this managed data set under the region in US central one, I click Create, and now we need to upload the images into this managed data set. The images are already uploaded into a Google Cloud storage bucket. So, for that, in this case, in the select and import method I will select import files from Google Cloud Storage, and here we need to specify the path to the bucket that contains these images. So, here we have the path to this Google Cloud Storage bucket and we click in continue. So, now the important work starts, we need to wait around 10 minutes until this work finished, and then once we have them managed data set already created, we can train our first machine learning model. The manage data set is already created. You see here that we find an error. If we click under details it's not a big problem, we are having a problem uploading one image. So, this is why we can dismiss this error, and here we have more than 3600 images of different kinds of flowers, in this case, five different classes of flowers. We have the manage data set already created. The next step is to train the model, so we click in this button train a new model and we will specify, we want to use a custom model or an AutoML model. We want Google make all the hard work for us. Yes, this is the case that we want to use now. So, this is why we select AutoML and click under continue. We assign a name of this model. I will call this model, flowers consider, if I click here in advanced options, we can see that AutoML, by default, from all of our image datasets will use 80% of these images for training, 10% for validation, and 10% for testing. We will keep these numbers and we will click continue, and in the budget I will specify eight hours but don't worry because I have enabled earlier stopping. That means that if AutoML finished to train a good model in less time, we will only pay for the amount of time that AutoML was used. So, I click start training and now the magic happens in the background for us. Now AutoML is start to making his jobs, and we need to wait around one hour to have this job finished. So, we will wait until the job is already finished, and once we have our model created we will evaluate the different performance metrics. And if we are happy with the results, we want to deploy our model into an endpoint and test if our model is good and it's working. When the training job finish, now if we click under models we will see our model here, Flowers Consider is ready. If we click under this model we can check the different performance metrics that we obtain training this model. We can check the confusion metrics and if we are happy with these results we can now deploy our model. If we click under deploy and test, I now want to deploy our model, so I click to deploy to end point. I will specify a name for our end point. So, its flowers consider. And if I click under here, continue, you see that under traffic split I have 100% to this end point. Why? Because in this case will be the first version of this model that I will deploy but imagine that I have more than one version, and I want to create some kind of canary or AB testing deployment. I can specify please send 80% of your traffic to this end point with this model and 20% of the traffic to this other model. So it's super, super good and super, super easy to do all these jobs. So, here I specify 100% and number of compute nodes that I want to use for this, example to this end point, I will select one and I click deploy. So, again we need to wait, we need to wait a few minutes until our model is already deployed to an endpoint, and in this moment we will be able to test how our model is working. The deployment has finished, let's test our model. So, I will go to Google and I will download an image of my favorite flower, Sunflower, images, and for example, this test one and fingers crossed. Let's see if our model is working fine, upload image, this one. Now our model is trying to predict the class of this image and our model is working fine. The model predicts that it's a sunflower picture. Good job, you finished your first image recognition model. Congratulations.
