Three years ago or so, I wanted to update my workflow to model crop nutrition, identify plant needs and provide an idea of fertilizer dosage. I wrote a blog post to help me put my ideas into place, and why not help others by doing so. I forgot about that post and left it unpublished. I found it back by chance while scanning my profile on NextJournal. Machine learning have evolved tremendously since then, but I think it is still valuable.
In the notebook, I explain how to obtain a plant nutrition recommendation system based on local soil and weather conditions.
The computing platform Noteable recently created a plugin that allows interaction with ChatGPT. I opened a notebook, then dropped a dataset. I asked ChatGPT to generate the code in #Rstats to create an AI model on this data set. 30 seconds later, I had an optimized code. It's sick. It completely (and positively) upsets the way of teaching numerical computations. Rather than focusing on writing code, teaching should now focus more on the importance
to ask the right questions and
to understand the meaning of the answers
In a survey of 791 articles in 5 scientific journals, about half of the articles misinterpreted the notion of p-value, yet a gold standard in science (for the wrong reasons). We will be able to really focus on training critical thinking and put aside the ability to type code!
When a colleague told me he assigned an undergrad student full-time to copy-paste historical weather data from Environment Canada, I searched into my files and found some code which I could format and share. The weathercan package, a gem written by Steffi LaZerte, allowed me to automate fetching tasks and save countless hours. All I had to provide was the spatial coordinates of my sites and, since my work was about agricultural data, time periods expressed as beginning of the season to harvest.