Python3 Migration Replicability

[Anonymous]

Hi, 

If I need to migrate a model that's deployed on MLOps to python3, will I be able to ensure the same results? I will try to replicate the model in a new project using the same dataset, samples and features and then replace the model.

Thanks!

[dustin.burke]

@Bruce If its alright, let's follow up this discussion within the thread you initiated here: https://community.datarobot.com/t5/platform/what-does-python-2-disabling-mean-to-me/m-p/14654#M2207

[Bruce]

@dustin.burke Perhaps you can set me straight here. I am using the Python API for Datarobot. I am writing in Python-3. On the older project I got a warning that Datarobot was switching to Python-3 and that "this python-2 based project is depricated". Fine, as far as it goes. AFAIK this is really an internal Datarobot corp issue. I never made any decision about using Python-2 in the project, and I am personally using Python-3. But, I just now created a new project and gave me the exact same warning. So, how do I tell Datarobot to use Python-3. What exactly is it complaining about?

[dustin.burke · Accepted solution]

Hi @Hannah - thanks for your question.

As an MLOps user, you could take advantage of automatic retraining which might make it easier for you to recreate the project and model with the same settings and compare the models before making the replacement.  That will likely save you some time and button clicks from trying to recreate the project from the original dataset and ensuring the same settings are used.

In terms of "will I be able to ensure the same results", it depends what you mean by same results.

• Yes, you should be able to create a model that will meet or exceed the performance of the original model in terms of accuracy.
• However, you may or may not get the same exact blueprint.  It's possible in many cases to create the same exact blueprint, but it depends on whether the original blueprint is still supported (e.g. Scaleout and OSS models were deprecated some years ago).  It's also possible that another blueprint might come up as the top model on the leaderboard since over time we make improvements to the platform.
• Although we have high confidence you can find a model that will perform as well or better, we can not guarantee row-by-row prediction consistency between the models.

Hope that helps!  Let me know if any other questions related to the migration.

Thanks!

Dustin

---

@Anonymous wrote:

Hi, 

If I need to migrate a model that's deployed on MLOps to python3, will I be able to ensure the same results? I will try to replicate the model in a new project using the same dataset, samples and features and then replace the model.

Thanks!

---