Codes

[Doctor Youness]

Hi, everybody. I'm trying out DataRobot (traial version for 14 days). However, I have not seen the option, where I can download the code of each algorithm within the DataRobot platform?

Thank you very much

[Anonymous · Accepted solution]

Hi Doctor Youness - we're working to get the right answer for you. Hoping to get back to you asap.

[Doctor Youness]

Thank you, I'm waiting to discuss with you about machine learning algorithms for prediction and classification

[nenshad · Accepted solution]

Hello @Doctor Youness, it sounds like you want to learn more about the algorithms we use in DataRobot.  If you follow the trial tutorials, you will upload a dataset and then start the automated modeling process. You will then notice models being created by DataRobot on the models tab. If you click on any specific model under the modeling tab, you will see options to evaluate, understand, describe, predict, and add comments. If you click describe, you will then be brought to a a view with our model blueprints, which show the automated preprocessing and algorithmic tasks that the blueprint contains. Within the blueprint, if you click on a given node in the blueprint, you will see DataRobot Model Docs hyperlink, which will explain the algorithms in great detail. Please see the attached screenshot for an example. I hope it helps!

[Doctor Youness]

Thank you very much for your reply. However, what I need, for example, let's talk about this example, which has been resolved by DataRobot is about Predictive Maintenance (https://app2.datarobot.com/new/3), using diferents kinds of machine learning algorithms as:

1. Elastic-Net Classifier
2. AVG Blender
3. LightGradient boosting on ElasticNet Predictions
4. eXtrem Gradient Boosted Trees Classifer
5. RuleFit Classifier
6. Elastic-Net Classifier (mixing alpha=0.5/Binomial Deviance)
7. Generalized Additive2 Model
8. Light Gradient Boosted Trees Classifier with Early Stopping
9. Keras Slim Residual Neural Network Classifier using training nschedule (1 Layer:64Units)
10. Gradient Boosted Trees Classifier
11. RandomForest Classifier

The question is how can I download or consult the codes that have been developed to solve this particular problem problem i.e Predictive Maintenance (https://app2.datarobot.com/new/3).

Once again, Thank you very much for your time and support

[nenshad]

Thank you for clarifying what you are looking for @Doctor Youness . We have recently launched DataRobot Pathfinder, which has a rich library of use cases with detailed descriptions of the preprocessing, feature engineering, and algorithmic techniques used to solve many of the most challenging business use cases. Some use cases are described in much more detail than others, so please make sure to read a few of them to get a sense of how one might solve your specific use case.

We are looking into enhancing DataRobot to be able to launch many example use cases described by Pathfinder directly in the application with sample datasets, features, and recommended model blueprints, and your feedback helps to confirm that this is the direction we should be pursuing! I hope this information helps you.

[Doctor Youness]

Thank you for your reply. However, Matlab provides us some example in machine learning using Graphical User Interface (GUI) and in the same time the codes behind graphical user interface.

I'm trying to download the codes of some examples in DataRobot Platform. Unfortunatly, just GUI without codes!!!

So. How could I do to download the codes about the examples solved by DataRobot?

Thank you very much for your time

Have a great day!

[duncanrenfrow]

Hi @Doctor Youness ,

To directly answer your question, DataRobot does not directly allow you to export the python code behind the blueprint. There are a few reasons for that:

1. We rigorously test each blueprint across a large number of real world datasets every time to guarantee performance, accuracy, avoid leakage across partitions, etc. By not letting users download / change these blueprints we ensure that all DataRobot users have similar experiences / quality of models 
2. One of our goals frankly is to abstract away worrying about low level details. We build models using a variety of techniques and provide the visualizations / tools to interpret & evaluate our models. In many ways this is similar to how Microsoft Excel or other, older, analytic tools have always functioned. To generate a random uniform number between 0 and 1, Excel uses the Mersenne Twister Algorithm. But Microsoft doesn't provide the specific source code that implements that algorithm, and in my mind they really shouldn't. 
3. The code we write at DataRobot is designed to function in a complex, enterprise grade product with a massive codebase. This is different, and likely far more complex, than the typical ML pipelines that a data scientist would write for a one-off project. 
4. Last, but obviously not least, all software vendors have to balance transparency vs. protecting their intellectual property. Google famously obscures the details of their search ranking algorithm(s), Salesforce guards their CRM source code, etc. We provide a variety of ways to export our models (including JAR files, linear approximations, direct formulas, API endpoints, etc.) but as a company may have to protect certain tips / tricks that we developed and we believe give us a competitive advantage

It is worth noting that we're committed to open source. Many of our models utilize open source packages and we've often contributed bug fixes / other enhancements back to these open source projects. 

One final point is that we are also trying to make DataRobot as open a platform as possible. Our MLOps capability allows users to upload custom models developed outside of DataRobot in R, Python, etc. These custom models can then compete on the leaderboard and take advantage of DataRobot's model monitoring / management, interpretability, and other capabilities. 

Cheers,

Duncan 

[Doctor Youness]

Thank you very much for your message. In matlab for example, I've used in many cases of research project graphical user interface (GUI) and in the same time matlab gives me the option to convert everything I have done in "GUI" into codes into Application Programmer Interface (API) as M-files.MATLAB.

What I would like, from you please as a help that datarobot does the same as Matlab with the option to export the codes of the best algorithm that has been found by datarobot about my database. I am doing a research project without profit. Therefore, I am going to cite DaraRobot Inc in references section.

Once again, thank you very much for your time

Best regards

Youness

[duncanrenfrow · Accepted solution]

Hi @Doctor Youness ,

Unfortunately we don't offer that capability and likely will not in the short term for the above reasons I described. However, if you click on the Describe -> Blueprint tab you'll see our blueprints outlined as a series of rectangles. If you click on any of those rectangles, you'll link to our documentation which cites academic papers that describe the techniques would use. 

It's worth noting that our blueprints implement an entire ML pipeline not just a single algorithm. As a simple example, one of our blueprints uses unsupervised learning techniques, a supplementary linear classifier, and then feeds it all into a gradient boosted tree. 

We have had multiple researchers publish papers using DataRobot, and they have either cited our platform directly or gone further and cited the references we provide in our documentation. 

Regards,

Duncan 

[Doctor Youness · Accepted solution]

Thank you very much duncanrenfrow, I need to ask you, if you can share with me please, the papers that datarobot has used to carry out scientific research?

Thank you very for your time

Regards

Youness

[duncanrenfrow]

Hi Dr. Youness,

No problem! We're sending a list of papers in a private message in the next few hours.

Cheers,

Duncan 

[Doctor Youness]

Thank you very much duncanrenfrow

Have a great day!

[Doctor Youness]

Hi!

They said:

"DataRobot Prime allows you to download scoring code (Python, Java) for any model on your leaderboard and use it directly in your application. When run, it creates an approximation for that model".

but it is not true, I already have DataRobot Prime, I have tried to generate the code and downloaded it, but I have not been able to open it, so that I can start editing it, like what I am doing with Matlab!

I need a help about this issue, please!

Awaiting your reply. Please!

Best regards

Youness

[RobinJohnson]

Hello @Doctor Youness !

DataRobot Prime is intended to be scoring code that you use to get predictions on new data. So while it is transparent, it's not designed for development work. 

But maybe I can help - can you tell me the big picture of what you're aiming to do?   

Thanks!

[Doctor Youness]

Thank yo so much for your time in replying my questions

All that I want is that generate python code from DataRobot Prime to get prediction on new data. For example, I need DataRobot application in real estate valuation, taking into account attributes such as location, urban proximity reference, number of levels, remaining useful life in months, general profitable units, land area, number of bedrooms, number of bathrooms, number of half bathrooms , number of parking spaces, elevator, appraisal date, construction class. It's done. However, I need to generate python code from DataRobot Prime to get prediction on new data about real estate valuation but using python code.

How could I generate python code from DataRobot Prime to get prediction on new data, please?

Could you give me an example with technical details, please?

Once again, thank you very much for your time

Best regards

Dr. Youness

[doyouevendata · Accepted solution]

Dr. Youness - if your account is licenses to use DataRobot Prime, the GUI workflow to generate and download the code can be found in this article: Exporting models with DataRobot Prime