I conduct quantitative evaluation on numerous topics. And, I run into an inherent constraint in my lack of proficiency in extracting complicated knowledge.
With the arrival of AI coding help, I explored whether or not it might help me in overcoming my lack of knowledge extraction proficiency.
Earlier I handed AIs by a number of coding challenges within the R language. See linked articles under.
Out of all of the AIs I examined, I discovered Claude to be the extra user-friendly for a starting coder.
My knowledge extraction problem: Marin County Crime Charge Information
I wished to research crime price tendencies in Marin County (Northern California) utilizing the County Sheriff Reported Crimes knowledge set (hyperlink under to this knowledge).
This knowledge set is genuinely means out of my league absent any AI help.
This knowledge set has:
- Over 44,000 rows. Every row captures a criminal offense occasion at a really particular time. The rows’ order is random (not chronological).
- 101 distinctive crime varieties.
- 21 completely different crime lessons.
- 141 completely different cities (the bulk outdoors Marin County).
- 109 completely different cities (the bulk outdoors Marin County).
- It covers the interval from 2013 to the current day. However, bear in mind there isn’t a chronological order to the info.
The completed product: Marin County Crime Charge Developments article
With out AI coding help, I’d not have been in a position to produce this analytical article centered on 9 particular crime varieties for 8 cities inside Marin County over the 2013 to 2023 interval.
Claude is so a lot better than any human help I ever had
Within the latest previous I availed myself of two most important forms of human help:
- Stackoverflow
- Colleagues at work
Stackoverflow by no means offered me with precise coding help. I usually would ask “How can I code this …” (offering a nicely specified instance in plain English). Invariably, the members of this group would request I submit my codes earlier than offering help. Do you get the irony?!
My former colleagues at work have been proficient, however hardly ever had the time to supply me help.
Now comes Claude…
- It’s a far sooner coder than any human being I interacted with.
- It clearly paperwork and explains its codes.
- It’s at my service 24/7.
- It offers fast coding that I can paste inside my R script.
- It has 100% recall of what we final labored on collectively. You possibly can go away a coding venture for months, and later reactivate it. Claude remembers precisely the place to choose up from.
Claude is a lot sooner than the Web
No matter coding I realized earlier than the AIs, I primarily realized it from the Web by Google search, GitHub, and Yelp. However, coding help with these sources typically was prolonged analysis tasks.
Is Claude excellent? No, it may possibly make a variety of errors
However, it makes “good” errors primarily based on arcane coding technicalities. It doesn’t “hallucinate” as Gemini does when it codes in Chinese language (take a look at my linked article “Can AI(s) code Deep Neural Community fashions?…”).
You possibly can examine my associated thread at Claude when engaged on this venture.
Collectively, Claude and I labored out by many errors. When Claude’s codes generated errors in my R script, I copied and pasted the error into Claude’s chat field, and it most frequently mounted such errors.
One AI is sweet. Two AIs are higher
In a few situations, Claude couldn’t resolve an error. You already know when Claude is caught when it makes an attempt to resolve Error A with a solution that generates Error B, then Error C… and finally makes an attempt to repair Error C by re-generating Error A. At that time, Claude is caught.
In my venture, this occurred a few instances associated to knowledge visualization of the crime knowledge.
Inside the graph matrix under, I couldn’t get Claude to offer me working codes to generate one single widespread legend on the backside (as a substitute of beneath every of the 9 graphs).
To resolve this problem I went to Chat GPT. See its reply on the hyperlink under.
https://chatgpt.com/c/48c25ab3-388e-40f7-8667-e314d2d8a381
The opposite occasion when Claude obtained caught was when coping with the time sequence scatter plot proven under. I wished the LOESS confidence intervals to point out up in the identical colour because the LOESS regression traces (crimson for San Rafael and inexperienced for Sausalito). Claude couldn’t present me working codes for that. As a substitute, it generated grey confidence intervals that didn’t clearly differentiate between the San Rafael and Sausalito auto theft tendencies.
So, once more I went to Chat GPT to repair this problem. And, you may see the way it did it on the following hyperlink:
https://chatgpt.com/c/8ec40433-453f-4d57-97dd-1e3c31f736ac
Is Chat GPT higher than Claude?
I don’t assume so.
My earlier AI coding challenges (talked about and linked articles) indicated that Claude carried out as nicely or higher than Chat GPT on these coding challenges.
Claude is my first-go-to coding assistant. When it will get caught I am going to Chat GPT subsequent.
I’m fairly certain that if I reversed the order and used Chat GPT as my first-go-to coding assistant, Claude would return the favor and repair Chat GPT’s errors.
THE END
