Building recommendation systems using Amazon Personalize | AWS Machine Learning

[Music] hello everyone in this video we are going to look at amazon personalize which is a machine learning service which is uh used for creating recommendation systems without writing any code so as you can see i have created some dummy recommendation systems so for this video we are going to create everything from scratch now so when you go to your amazon personalized service create a click on create data set group and you can enter the name of your data set group so i will put recommend movies click on next it says recommend movies for successfully created now when you go back to your amazon personalize you have one data data set group all right so when the status changes to active you can actually do and do uh other things on this particular data set group so we we will be importing some data and we will be creating some models data set group is active you can see it here we will now be importing the data sets all right so i've taken a data set which has uh the ratings the movies and the users all right so i've opened them these are the csv files and now we will be importing them the first one is the user item interaction data so click on import it says data set name so i'll set up right and it says use existing schema or create a new schema now here you need to enter what you have in your csv file the column headers all right so uh so this is what i already had so in my ratings user so the ratings.csv file is actually the user interaction data set all right so i have user id item id rating and time stamp right so that is what i need to uh need to be need to show on my schema so if i create a new schema so by default you have user id item id and time time stamp all right so in the existing schema i i added a new column which is the rating all right so uh if you don't uh have this you have something else then make sure you add that and then then only you can import the data set all right so what i'll do is i'll use the existing schema which is the movie ratings well let's do one thing we'll create a new schema copy this create new schema enter the name which will be the user interaction like this and then you can click on next so it says user item interaction data set was successfully created import job name give an im service rule which i already have and the location of your data so you have to give here your s3 bucket location all right so what we'll do is we'll go to our s3 service and we are going to create a new bucket now click on create bucket and say go next keep default uncheck block all public access click on next and say create bucket all right so i have this bucket which is empty now what i'll do is now as you can see i have the user id item id rating and time stamp in my csv file so what i'll do now is i will upload this particular csv file on this bucket so i'll drag this and upload so i have this csv file here now but this s3 bucket and does not have access to the amazon personalize all right so what we need to do is we have to go to the permissions tab and then to bucket policy and we have to add a policy to your policy here all right so for that i'll copy this part this is in json format i'll paste it here and then i have to give the name of the bucket so the name of the bucket is you can save so now your amazon personalized also has access to this particular bucket all right go back to amazon personalize and you have to uh give the link to the ratings.csv file all right so now you click on readings.csv it says copy path click on this go back to our personalize paste that link here and you can click on start import [Music] so it will say create printing and it will be in progress right now so the same thing that you need to you have to do it for you have to do for user data and item data all right so click on import and give some name and user data i have a schema which is user id movie id and timestamp all right so i've i have to make sure that this is what i have in the uh schema definition all right so what i'll do is click on new schema and you can paste it here or i have the previous one so i already have an existing schema which says user id movie id and timestamp alright so i'll use this and i'll click on next name will be user data and again i need to give the s3 location so i'll upload my data usb file to get to that s3 bucket along with that i also add the items the next data set that we are going to import [Music] okay so go on users click on copy path go back to personalize and paste it here click on start import do the same for item data as well and the definition schema will be item id title and genre all right so i think i have that particular schema as well yeah so i'll use this copy the path so i have all the data sets in the uh progress mode so when i click on this so it's still processing so these are my data set all right so on the left as you can see you we have some options which is the event tracker solution and reps recipes filters campaigns and batch interface inference so uh the data set that we have imported will take some time so what i'll do is i already have another dataset group which where i've imported everything and they are active all right so as you can see i have these particular data set that had imported earlier the same data sets right so as you can see i have something called a solutions here now solutions is what uh you're going to do for to train your model on the dataset that you have imported and recipes recipes is the algorithm that you're going to choose so for recommendation system there different algorithm that you can choose and uh with that with that model you're going to train the data set that you have uploaded so solution is the treating and recipe is the algorithm choosing all right so i've created four solutions so how can you create a solution is simply click click on create solution give it a name then you choose your algorithm algorithm which is the recipe so as you can see there are multiple options the aws personalized ranking popularity count and you have others also so as per your application you can choose your algorithm so click on one of them and you have the advanced configuration of perform hpo the hyper parameter optimization all right there are other options as well click on next so it has created your solution so this is the solution that we created so it will take this will also take time for the status to change to active so i already have i have created four different solutions with four different models four different algorithms and we'll see the results all right so the next thing that we have is filters so filters are something if you want to remove certain items from your recommendation based on rules that you define you can you can use a filter so in this video we are not going to look at filters all right the next thing that we have is campaign so once you have uh imported your data set you have chosen your algorithm and then you have created a solution as well solution is a training now a campaign is what you do is what is going to deploy your solution all right so for the four models that we have trained and created uh the models we have created four different campaigns right so to create a campaign what you do is click on create campaign give it a name and you can choose solution so right now we created a solution test solution right and this is the minimum provision transaction personnel let's keep it one so it says selected solution must be an active solution version so as i said it takes some time around 10 to 15 minutes minutes based on your data uh to change the status to active all right so you can so i already have four different solutions and i can choose one of them and create a solution all right so as you can see so it has accepted one version which is the active version for this particular solution okay it says more than five resources in active state please delete some and try again all right so we will be looking at the four campus that i've ordered that i already created before all right so if i go on campaign two it's telling me so this is where you're going to get the recommendations all right so i've used let's look at the details you here you'll find everything the solution version id and everything else so i need to give a user id so let's go to our users so i'll say 18 and now i have the item ids so this particular algorithm is the personalized ranking algorithm that i've chosen all right so i have to give a list of my item ids as well so let's go to our data set get some of the ids one more okay for the user id 18 i have given some item ids all right so i'll just click on get personalized item rankings when i click on this it's given give is giving me a ranking all right with some score so it says for one two two one id with a score of 67 percent so this is the personalized ranking algorithm algorithm that i've chosen all right so let's go back to our campaigns and look at something else as well let's go to campaign 3 and you can give the user id let's say 62 and add the context as well then click on get recommendations so you get a list of item ids with some score all right so this is how you can create different solutions with different algorithms which are the recipes and then you can create your own campaigns uh to deploy your solution versions and this is how you use amazon personalize all right so there are different other things also you have something called as batch inference jobs so you can create a batch you have you also have a event tracker so i have not covered that part in this video but you can do so you can look at the documentation the documentation is quite great it's easy to understand all right so if you like our content please like share and subscribe thank you [Music] you

Channel: HackerShrine

Views: 4,258

Rating: 4.9578948 out of 5

Keywords: aws, aws machine learning, aws machine learning tutorials, aws personalize tutorials, amazon personalize tutorials, aws recommendation systems, aws personalize recommendation system, recommendation system with aws, aws tutorials, education, technology, cloud computing, #softwaredeveloper, #hackershrine

Id: Qz9w-DdIJRg

Length: 15min 59sec (959 seconds)

Published: Mon Sep 07 2020