- 5/5/2025
Welcome to this comprehensive lab session on Automated Machine Learning (AutoML) in Microsoft Azure Machine Learning! 🎓 Whether you're preparing for the AI-900 exam or simply diving into the fascinating world of AI and machine learning, this hands-on walkthrough will guide you through the basics of using AutoML to build powerful, data-driven models with minimal coding.
🔍 What You'll Learn in This Video:
1️⃣ What is Automated Machine Learning (AutoML)?
2️⃣ How to set up and navigate Azure Machine Learning Studio.
3️⃣ Step-by-step guide to creating, training, and evaluating models using AutoML.
4️⃣ Key features and benefits of AutoML for developers and data scientists.
5️⃣ Real-world use cases and applications.
🛠️ Who Is This For?
Beginners exploring AI and ML.
Professionals preparing for Microsoft AI-900 certification.
Data enthusiasts curious about Azure Machine Learning tools.
📌 Key Features Highlighted:
✅ Simple, no-code/low-code approach for building ML models.
✅ Automated hyperparameter tuning.
✅ Scalable infrastructure powered by Azure Cloud.
✅ Evaluation and deployment of models within minutes.
💡 Get Ready to Supercharge Your AI Skills
Explore Our Other Azure Courses and additional resources on:
https://www.youtube.com/@skilltechclub
🔍 What You'll Learn in This Video:
1️⃣ What is Automated Machine Learning (AutoML)?
2️⃣ How to set up and navigate Azure Machine Learning Studio.
3️⃣ Step-by-step guide to creating, training, and evaluating models using AutoML.
4️⃣ Key features and benefits of AutoML for developers and data scientists.
5️⃣ Real-world use cases and applications.
🛠️ Who Is This For?
Beginners exploring AI and ML.
Professionals preparing for Microsoft AI-900 certification.
Data enthusiasts curious about Azure Machine Learning tools.
📌 Key Features Highlighted:
✅ Simple, no-code/low-code approach for building ML models.
✅ Automated hyperparameter tuning.
✅ Scalable infrastructure powered by Azure Cloud.
✅ Evaluation and deployment of models within minutes.
💡 Get Ready to Supercharge Your AI Skills
Explore Our Other Azure Courses and additional resources on:
https://www.youtube.com/@skilltechclub
Category
🤖
TechTranscript
00:00hey guys welcome to this video in which we are going to focus on automated
00:12machine learning with Microsoft Azure cloud I'm sure you guys have understood
00:16the connection between AI and ML so far because we have gone through our Azure
00:21AI fundamental videos now obviously AI and ML both are connected but then in
00:27this particular video we are first going to focus on how exactly machine
00:30learning model works how exactly we can use our data and train the machine
00:35learning models obviously there are multiple ways to do this thing we are
00:39going to use Azure ML which is a Microsoft Azure machine learning service
00:43now we are going to use that service we are going to train the model from
00:46scratch based on some data and then I'm going to show you that how exactly we can
00:51deploy that as an API so that we can consume that model and we can develop
00:56our applications like AI so let's get started with this the first thing is I
01:02have already logged into my Azure portal right now and I'm going to create a new
01:06resource because we are going to do all the things step by step I want you to
01:10make sure that you understand things properly and if you also have Azure
01:14subscription with you I strongly recommend you should also try this labs to
01:18understand it betterly let's move forward with machine learning so I'm going
01:24to search for machine learning and I'm going to get machine learning workspace
01:29with that we are going to create a new machine learning workspace which is a
01:34service allowing me to do all the machine learning model training with that we'll
01:39click on create machine learning I already have my own Azure subscription so I
01:47have selected that subscription here in the resource group we are going to create a new
01:51resource group called AI 900 RG because this video will be a part of the AI 900
01:58course the name of my machine learning workspace I'm giving something like my
02:02name mlws25 because this is 2025 now if you are not learning machine learning in AI in
02:11this year then when exactly you're planning to do that well this is a perfect time to
02:16do that so I'm glad if you're learning this and we are choosing a region which
02:21is going to be East US this service is actually going to create couple of other
02:25services with that in which machine learning is dependent so you can see
02:29right now we have a storage account we have a key vault we have application
02:34inside these three services are actually going to be created in the same
02:38resource group in one deployment so when you provision this machine learning
02:42workspace which I'm going to do right now by clicking on this review plus create
02:46this is actually going to create all this dependency resources as well as
02:51machine learning workspace also with that now obviously this is one of the
02:55easiest way to create a machine learning workspace using Azure portal as for your
03:01knowledge I think you know that we can also do this thing by using PowerShell
03:05scripting or using the internal cloud shell prompt with the batch scripting so all
03:11these options are available in this Azure cloud we'll wait for this deployment to
03:16be completed because it's going to create my machine learning workspace and
03:20inside that workspace we are going to have something which is machine learning
03:23studio that machine learning studio is going to provide that environment where
03:28you can actually train your model you can configure multiple experiments you can
03:34execute jobs there are multiple things which we are going to see inside that by the
03:38the way while we are waiting for this deployment I want to remind you one thing
03:42that if you are not familiar with machine learning fundamentals I strongly
03:46recommend you to go through this particular video which is visible on the
03:50screen in that eye symbol this is our previous video which is going to show you
03:54machine learning fundamentals and I strongly recommend you understand the
03:59fundamental how exactly machine learning works and then after this if you do this
04:03lab that's going to make this concept crystal clear yes our deployment is
04:09completed let me click on go to resource and most of the time we do not have much
04:16things to do in this Azure machine learning workspace inside Azure portal
04:19instead of that we will click on launch studio so launch studio is actually going to
04:24give me a separate space which is a separate portal kind of an environment which is
04:30available on ml.azure.com this is a place where you are going to do a to z of
04:36everything of machine learning in Azure cloud you can also associate multiple
04:40Azure machine learning workspaces here now right now this is my workspace name and as
04:46you can see this is showing me that I can create generative AI with prompt flow we
04:52can do associations with generative AI models which are ready-made AI models
04:56available from open AI so we have GPT for models we have Microsoft models like
05:02Aurora and if I go next and next we have so many models from different different
05:07companies even Lama models from Meta is also available in this we're not going to
05:14talk about generative AI models in this particular video but if you are
05:18interested in learning any of these models please comment down in this
05:21particular video I will make sure that you will get a video dedicatedly on any
05:27of the models which you choose right now what I want you to focus is on this area
05:31where we have a section for authoring machine learning models and we have
05:36automated machine learning we have designer specific machine learning we have
05:40notebook specific machine learning which we can configure in this video we are
05:44going to focus on automated machine learning because I'm assuming that you are
05:48focusing on learning AI machine learning is something which you do not need to go
05:53in-depth but you need to understand what exactly machine learning is and how
05:57exactly all these models are created and that's exactly what we are going to do
06:01right now in this particular one so I'm going to go into this automated ML and
06:05what we are going to do next is we are going to create a new automated ML job now
06:11before I create this automated ML job I just want you to understand one more
06:15thing so I'm going back to my Azure portal I'm going into my AI 900 RG resource group and I hope
06:24you remember that there are a couple of other things which got created with our
06:27machine learning workspace one of them thing was this storage account right now
06:32this storage account is connected with my machine learning workspace and if I go
06:36into the container section which is actually going to show me a blob
06:40containers of this particular thing is showing me that is your ML is your ML
06:45blob store this is actually a blob store which got configured automatically with
06:50your machine learning workspace now why I'm showing you this thing right now as of
06:54now if I go into this blob store it does not have any data is showing me no
06:58results here we are going to do some configuration with automated ML and we
07:03are going to associate some data with our automated ML process now in this case the
07:09data which we are going to use is coming from this particular link now this is a
07:14website which is capitalbikeshare.com which is having a system data available
07:18here now this is the data which I have already downloaded in my computer and I'm
07:22going to use this thing if you want to use the same data for your machine
07:27learning model training you can use it otherwise if you have any data which is in
07:31JSON or CSV format you can actually add that data and then you can train your
07:36models on that now this is a very official data kind of a data files which are
07:40available here and we have a license agreement also which is available for
07:44that so you should not use this thing on authentically actually so this is how
07:49we are going to do this this is a data which is available on this URL but let me
07:55just go back to my automated ML and let me show you how we can actually
07:58configure this automated ML process now in this case I'm going to create a new
08:03automated ML job when you create a new automated ML job it's going to give some
08:13random dummy name for this particular job so you can see right now my job name is
08:18having something like mango brick with some you know unique name which is
08:23associated with that now you will also get this kind of a name it's not very
08:27important this is a job name but yes in this job when I'm going to create an
08:31experiment I want to give a proper name for this experiment so I'm going to give
08:36a name that this is my experiment name which is having a name MS learn bike
08:42rental now for your kind of information this is an official lab which is
08:46available on Microsoft learn website the link of this lab is going to be
08:50available in the description of this particular video so you can just check
08:54out that if you want to see each and every step which we are doing right now the
08:59description of this particular one I'm going to specify like this automated
09:05machine learning for bike rental prediction and then we are all good to
09:09set up our automated ML job when we click on next the first thing it's asking me is
09:15what kind of a task type you want to choose now obviously depends upon the
09:20algorithm which I want to apply and depends upon the outcome which I want to
09:24predict we have to choose our task type in my case I'm going to choose
09:28regression because this regression is going to help me to predict continuous
09:32numeric values I hope you understood this from the fundamentals of machine
09:36learning video if not you can mention any of this particular machine learning
09:41algorithm type so you can comment down that in this particular video and we are
09:46going to create a new video on the specific machine learning type in which
09:49you're interested after the task type we have to select our data and as we are
09:55going to use this data from a capital bike share which is giving me a bike
09:58rental data I'm going to click on creating a new data set name of the data
10:04set I'm specifying something like bike rentals description of this we are going
10:12to see that this is going to provide a historic bike rental data and in the type
10:18I'm going to choose table which is going to be ML table remember this is a one
10:23particular data asset type which we are going to use whenever we are using a
10:27sure machine learning version to API this is a newer API which is launched
10:32recently before few months only and the older one was actually taking tabular
10:36data now as of now you can use both of them but because this videos are
10:41focusing on more advanced machine learning in future I'm going to always
10:44choose table which is having ML table option with the version to API in all my
10:49videos so I'm choosing this I'm selecting next in the data source we have to choose
10:56a source for your data set now we are going to choose it from local files so we
11:00are going to actually upload it here and that is what we are going to choose next
11:06it's going to ask me that okay when you upload the files is actually going to be
11:10uploaded into your blob storage and that is your workspace blob store remember this
11:16is the blob store which I have shown you in the separate tab so this is actually
11:20not having any files right now but once we try to upload the files it's going to
11:24be available inside this blob store let's click on upload folder I have this bike
11:30data available in one of my downloaded location I'm going to upload and this data
11:38is actually available here if having one CSV file and one ML table both of them are
11:45uploaded using that particular folder I'm going to click on next and I think it's all good the
11:54review page is showing me that this kind of data asset you're going to create I'm happy
11:58with this I'm going to click on create once this data asset is successfully created you
12:03will be able to see this thing in your Azure blob storage because it's connected with this
12:09machine learning workspace we will choose our task type we will choose our data set and then we
12:15are going to click next and we're going to configure this but before I click on next let's just cross
12:19check the upload happened successfully or not if I refresh this page now we have a folder called UI
12:26inside that we have today's date and then inside that we have bike data which means yes it successfully
12:33uploaded and we have the data files also everything looks good let's move to next step in the task
12:42settings there are a couple of things which we have to configure the first thing is you can choose
12:47what kind of a targeted column you are going to use for predicting this particular model outcome now
12:53there are so many columns which are coming from the data which we have uploaded we are going to choose
12:59a column which is rentals because this is that column which I want to use for predicting the rentals of
13:04that particular month or that particular duration so I'm choosing render which is an integer column and for
13:10your kind of information regression algorithms are perfect for predicting a continuous numeric value
13:17this column is integer which means it's going to have numeric values in that which makes sense in the
13:22view additional configuration settings which primary metric I want to focus so when I'm going to focus on
13:30the performance of this model what would be my primary metric is something which is very important I am
13:35choosing normalize root mean square error that's what I want to go for no I do not want to explain
13:42the best model if I do this thing if you take little extra time normally in my real machine learning model
13:47training I always explain the best model so that we can understand better but as of now I don't want
13:52that and no I do not want to use all supported models because in that case it's again going to increase
13:57the duration of the machine learning model training in my case from the allowed models I'm going to choose some
14:03of them let's say right now from the list of the models I am going to choose random forest and light gbm
14:12now maybe some of you are not familiar with what are these models what is the meaning of sgd or what is
14:18the meaning of gradient boosting if you want a special dedicated video on any of these models which are there in
14:26the list please comment down that model name in this particular video and I will make sure that you will get a
14:32dedicated video on that very soon as of now I'm choosing random forest and light gbm so I'm not going
14:39with multiple models because I want to save my time of model training right now and I'm going to click on
14:44save so additional configuration is done these two models are selected then we also have a setting for
14:52view featureization setting now I'm not going in depth into featureization right now in this video
14:58but you can actually customize based on your columns and other featureization configuration
15:03there is a section which is very very important in automated machine learning which is limits
15:09basically this is going to help you to control when exactly you want to end the model training so
15:15how many maximum trials you want to run how many maximum concurrent trials you want to run how many
15:22maximum nodes you want to use in this and then what will be your experiment timeout these are all
15:27important values I'm going to provide that I want three trials three concurrent trials max node is also
15:35going to be three my metrics code threshold which I'm going to use is going to be 0.085 and that is
15:45perfect for me experiment timeout I'm going to specify 15 minutes in both and do I enable early termination
15:53yes I want that so if this is going to take a little long I want that I can enable early termination
16:01from my side so I can terminate this and that's how we are going to configure in the validation type
16:06I do not want automatic I always prefer to train validation split so I'm going to do a train validation split
16:14with 10 percentage of validation of data with that so it's like it's going to be divided based on number
16:19of percentage which you're going to choose in that I'm okay with this I'm going to click on next
16:26every time when you want to run automated ml job you need a compute and this compute can be either cpu
16:32or gpu depends upon the data the amount of data which you have the number of models which you're selecting
16:39with that obviously it's going to require higher gpu or higher cpus and more time I'm going to go with
16:45serverless right now which is one of the cheapest option available in this so I'm going with serverless
16:50cpu I want a dedicated virtue machine with this kind of a size and the number of instances which I want
16:56is one so I'm not changing anything in this I'm okay with all this and I'm just going to confirm
17:03everything is correct in my reviews tab looks fine I'm going to click on submit training job
17:10now once you submit the training job this is actually going to use your data and then based
17:15on that it's going to train your model based on those algorithms which you have selected
17:20you can see it's showing me right now the status of this job is not started if I wait for some time
17:26very soon it's going to show me that the job is successfully submitted the experiment is going to be
17:31executed and then it's going to be coming into the running state you have to wait for few minutes
17:37if this is going to complete this particular thing in 15 minutes it's going to show you the best
17:42performing model from that we are going to see that part so just wait for some time you can see right
17:48now now it's showing me running they are setting up the runs they are going to spin up that virtue machine
17:54which I have requested with that kind of serverless environment in that virtue machine they will try to
18:00execute those algorithms with machine learning this is a complex lengthy process but because
18:05you are using automated ML you don't need to worry about the internal configuration of that if you want
18:12to take control of all the things in your hand and if you want to apply your own conditional logic in
18:17that we will see the designer kind of a machine learning configuration very soon in the coming videos with
18:23machine learning but as of now let's just wait we will wait till the status of this is showing me
18:29successfully completed or until i'm getting some models here in this particular run summary so
18:36just fast forward to this particular video i'm going to do that thing and then once it is done
18:42we'll continue from this point okay now in my case you can see that it's showing me that the status is
18:48running and it is actually doing model training right now this is the time where we have that 15 minutes
18:53timer which is going to be started and then after 15 minutes the job will be terminated now this also
19:00reminds me one thing that if you have reached to this point and if your model is also under training
19:05so this is the time i want to mention that if you are liking this video so far this is the perfect time
19:10that you can click on the subscribe button of this particular youtube channel and you should not
19:14forget to click on that bell icon so that whenever we publish a new video you'll get a notification of that
19:20and also if you are liking this video it's really useful if you just like this video by pressing the
19:26like button thank you let's wait for this model training to complete and then we'll move forward
19:32to see the result of this particular model training okay now as you can see right now it's showing me
19:38that the status of this particular job is completed and because the job is completed it means that we got
19:44some models which are actually prepared with the help of this particular machine learning model training
19:50at the right side of this section i am showing you that this is an algorithm name section in which
19:55they are showing me voting and symbol now i'm going to click on this algorithm name so that i can
20:01get inside this it's showing me the model summary the algorithm name is voting and symbol this is maybe
20:07the normalized root mean square error value for this which is 0.087 which is coming
20:13the sampling is 100 percent and this means that this particular model is successfully created based
20:18on this now if i want to deploy this model or if i want to taste this model i can do that thing but
20:24before that it is always advisable that we go to the metrics tab in the metrics tab it's going to show me
20:31a couple of metrics which are actually used when we are you know training this model i want you to focus
20:38on these two values here we have two graphs below this we have residual and we have predicted true
20:45residual chart is going to show you the difference between predicted and actual value in a histogram
20:51chart and that's what which is actually visible here while the predicted true chart is going to compare
20:56the predicted value with the true values which are visible in this now obviously based on this metrics
21:04you will be able to understand that whether this model is one of the best performing model or not
21:09as of now obviously this is not a very best performing model because we have run this thing
21:14for a very limited time now if you want to deep dive into the other models which are generated with this
21:20like i can go back to my job name which was mango brick yes i can click on the models plus child job
21:27section and this is going to show you that this is the one model which we have just now checked you
21:33have some other models also which are actually generated with the help of the same process
21:38and you can click on this you can check the same kind of details with those models as of now i'm
21:43going into my same voting ensemble model and then if i want to use this model then i have to deploy the
21:50model and i have to taste the model i'm going to click on this deploy section and there are multiple
21:56options in this also you can do real-time endpoint deployment you can do batch endpoint deployment
22:01and you can do web service deployment in our case we are going to choose real-time endpoint deployment
22:08so this is going to give me a real-time endpoint which is going to be live and available to taste
22:14if giving me a configuration for this i'm choosing that instance count for this will be three
22:19standard ds3v2 is perfect for me which is going to give me four cores and 14 gb ram remember this
22:25depends on your subscription quota if you are not able to deploy this with three instances you can reduce
22:31the instance counts to less a number and then it's going to allow you to do that thing in your
22:35subscription the endpoint name is going to be a new endpoint with this kind of a name which is also
22:40randomly generated i'm okay with that deployment name is also fine everything else is fine i'm just
22:46going to click on deploy now obviously if you do not have enough quota i'm again repeating in decrease
22:53the instance count from three to two or one or sometimes it is also possible that this specific
22:59virtue machine sku is not available in that case you have to go for the other options which are
23:04available in your subscription so this is something which is a workaround you have to
23:08figure out what kind of a quota which is available in your subscription
23:13it's showing me that model deployment is successfully triggered and obviously it would take some time
23:19normally a model deployment can also take five to ten minutes so this is the time we will wait for
23:25some time and then once the model deployment is completed i am going to show you how the endpoint
23:30looks like what kind of configurations we can deal with that and how to taste this deploy model
23:36okay so now we have waited for this model deployment and what we are going to do next is we will click on
23:43this model section here if i click on the model section is showing me that yes i have one new model which
23:48is available which is actually deployed and if this is created with the real-time endpoints we can see
23:55that real-time endpoint in this particular section so this is an endpoint section in the left side
24:00where we have a real-time endpoint which is with this name and the quota type is dedicated it's updated on
24:08this particular time it all looks good i'm going inside that endpoint and you can see right now this endpoint
24:14is available with the proper api documentation with swagger we have a rest endpoint url also if i
24:21just copy this url and if i try to taste this thing from anywhere i'll be able to do that thing
24:26as of now in this particular section we have a taste tab which is actually helping you to taste this
24:32particular deployment now because this taste tab is asking you to choose your deployment as of now we
24:38have only one deployment but when we have multiple we can actually choose whichever we want in that
24:44and then in the sample interface we have to provide our input based on which is going to predict one
24:50particular numerical value with that let me just provide a proper input data so i'm going to take
24:56my json i'm going to just remove this and paste that json data now you can see right now this is asking for
25:05input data with the columns and somewhere in that index and data we are providing the data with that
25:10i'm going to click on taste this is actually going to use my model and it's going to show me the json
25:15output which is based on that now obviously whether this prediction is uh meaningful for your business
25:21or not that's something which you have to check now this is all for the bike rental data and it's
25:27showing me the output obviously if i change the values of the data the json output value is going to get
25:32change this is showing me that my particular model is working fine we are able to taste it we already
25:39deployed it all the steps which we have done in this particular lab i hope it makes sense now
25:45just keep one more thing in mind that this is an automated ml if you want more customization power on
25:50this you have to go for designer based or notebook based machine learning where you can actually have
25:56a full control over what exactly you want to do in that also when you're going to use ai you're not
26:02going to use machine learning directly your machine learning models with this kind of api deployment
26:08is already going to be done you are going to use existing models which are going to be available
26:13with azure ai apis and using that api you are just going to use your model you're not going to train
26:19the model in that case thank you so much i'll see you in the next video bye bye