Recall "Dark Horse" back in a post here November 16, 2017 - https://www.aapg.org/care...net/activity/aft/105
here is update on interesting actionable twist, i.e., no-code ML model training to deployment. Simply required time-series like data (think production history, wireline log) prepared as comma-separated values (csv file). https://medium.com/data-s...-automl-e00a9f878d1c
Note - 1) AutoML is not just training a ML model, but let you select among many the best or optimal one for your application (e.g. decline curve analysis). 2) Worth noting is that average ranked at the bottom of the heap (a worthy reminder that using average often yields an overly optimistic picture on Shale well economics). 3) through the experiment (link above), you can opt to kick off deep learning (more sophisticated model like neural network etc.). Yet in many of geoscience time-series like data use cases, my observation is that effort on data prep and feature engineering has a greater impact vs sophistication of the ML model (e.g., multi-variate regression vs reinforcement learning).
Hypothesis testing: Geo-Deep Learning = Machine Learning + Geology DNA (do no average / simply human)
Perspective - AutoML made life 10x easier to experiment with different ML models, and bodes well with putting more effort into data quality and integrity (no small challenges). Feel free to share your experience and help AAPG develop the collective wisdom on using deep learning to expand our energy supply faster, cleaner and more sustainable than how it was done before.