Deployment: Free Version

[saatcsi]

Hello All,

I was trying to run the MLops tutorial but I am on the free version. So, the workflow for deployment isn't the same. I wanted to first know if it is possible to deploy using the free version. If so, what can we deploy and what are the limitations? 

Is there a resource that explains the process?

I am currently running tutorials for my class and through the Dataiku knowledge center. 

Thanks in advance,

Saa

---

Operating system used: Mac OS M1

[Turribeach]

The following table shows all the different features and their availability on the different Dstaiku versions:

https://www.dataiku.com/product/plans-and-features/

Neither batch scoring (Automation node) nor Realtime prediction API (API node) are supported in the Free version. 

[saatcsi]

Thank you for your response. I've seen this table. 

For an individual who is new to the platform and is super curious it is still a little ambiguous. There were several layers to my query, one of them being is it possible to deploy a model with the free version?

If there is no way to deploy using the free version, therefore you are just able to experiment in the dev mode then this will push me back towards RapidMiner. I've tried to reach out to sales to receive a quote but never received a response. 

Is it possible to deploy in anyway possible using the free version? If so, how? If not, how far are we able to push the free version? Please don't explain by sharing the link to the table. It is a shorthand for the information and I would like to understand on a deeper level.

Thank you. 

[Turribeach]

Deploy a model can mean many things. Is it a Visual ML model? Is it a custom coding model? Is it a pickle model? Is it an ML flow model? Each of those will have different deployment options. 

Then you need to think what does your model do. Does it do batch scoring in which case it could deploy to the Automation node. Is it real time scoring in which case it could need a new API service deployed in the API node. Or do you want to deploy it to a Kubernetes cluster or an external service like Sagemaker?

So I’m afraid it’s impossible to answer your question without understanding your requirements properly. So I think you could start by saying what exactly you are trying to achieve and why. And then we can try to guide you. 

[saatcsi]

I am glad you asked all of those questions. Much more helpful than pointing me to the feature list for each plan.

As stated in the original post, I am going throught the ML Ops quick tutorial. During one of the steps, you are to create an API from the Design Module using the provided Model in the flow. But, you cannot access the Deployer with the free version. So, this learning module is ineffective for beginners learning the deployment process with the free version. 

Searching through the knowledge center I could not find tutorials that explain the deployment options for the free version and how to understand them. 

Right now, I would simply like to walk through the process from building recipes in the flow (comfortable with this), selecting an auto ml option and adjusting the parameters (comfortable with this), refining and training a model (still foggy on this. Improving), then deploying the model (this process is pure darkness for me. I have no clue.) 

I feel really comfortable preparing and performing EDAs with the data. I also feel okay with training the model (although, I have some doubts with my current workflow to get the right results), I really need to understand refining and then deploying. 

I hoped by going through the ML Ops Quick Lesson I could get a top down view of the entire process from building recipes all the way to deployment and accessing the model (e.g. a python script, etc.) but the resources are not there. 

I have trained a few decision systems and as I've written in different ways would like to know how to get from beginning to end as well as what my limitations are with the free version. 

I am new but a quick learner. I just feel like I am jumping all over the site to find what should be a straight forward process. 

Thank you for your time and attention. 

[Turribeach]

As I said on my previous post neither Automation Node nor the API node are available on the free version so you will not be able to do the most common deployments without a license that enables them. Having said I wouldn't put too much weight in not knowing those steps as they are relatevely simple steps, assuming you are deploying to "simple" Automation and API Nodes without any Kubernetes etc. This video shows the GUI and how project gets deployed. The Automation Node is nothing else than a rebranded Designer node with some very small GUI differences, so you will very familiar with how the project runs as it will be very similar to the Designer Node. 

With regards to deploying an API service this documentation page covers the steps. Again there isn't much complexity here and deploying to the API node is very striaght forward. The option is actually shown in the video above too. I don't have the free Dataiku version but if you are able to design an API in the Designer Node you will be able to also test it externally as the Designer Node comes with an embedded API Node for development purposes. And provided you know what port it is running on you can call the new API even without doing an actual deployment. 

So all in all I don't think you are missing much.