THCFarmer.com
121,279 Members Log In Register
Log In Register

*NERD ALERT* Using AI for data set trend analysis

  • Thread starter Thread starter Brother_Antioch
  • Start date Start date
  • Tagged users Tagged users None

*NERD ALERT* Using AI for data set trend analysis

Brother_Antioch 49 Replies 5,094 Views
Page 2 of 3 · Replies 21–40 of 50
#21
Brother_Antioch said:
Thanks, shooter. I appreciate the input on it. What I'm currently doing is logging all my data readings 3 times a day into a spreadsheet, and after saving the latest copy I also save a second copy as a .csv file for upload. This way, ChatGPT looks at the entire data set for the last several weeks since I started it (approx. 10 days into the grow) all laid out in columns for trend analysis. I'll keep doing that the entire grow, each day adding 3 new lines of data points for it to analyze. I'm sure it'll get to the point where I upload the .csv file and latest pictures, then have to wait 20 minutes while it analyzes all the information...lol. Oh well, c'est la vie, as the French like to say.

For sure, I've caught many mistakes in its assumptions, its lack of "remembering" certain discussions about details and the like. I've come to realize that as an analytical tool it does a great job at recognizing trends and linking them back to specific changes made (I keep one column called "notes" where I detail any and all changes, alterations, and it treats it like gospel). While I'm fairly decent at looking things up online (I've got a Black Belt in Google!) creating my own program is light years beyond my abilities. I have been keeping an eye on one called "Herbies", though, which is in the beta phase and is specific to growing cannabis.

Ultimately, I'm treating AI as a handy tool rather than an expert to listen to for all my advice. That's why I created an account here, after all.

Cheers,

BA.
Click to expand...
What kind of phone are you using? Or is this all on a computer?
 
#22
If you really wanna get into some nuts and bolts type stuff without having to write a ton of software...

Build a multispectrum camera. AC infinity released a camera but its not multi spectrum. Phone pics also work, but have limitations.

With one of those you can literally see water in the leaves. You pack all that data up and send it to chatgpt and you can get some of the most detailed data you can imagine. Spider mites (gross) will fluoresce, you can spot leaf damage inside a leaf before its at the surface. There's so much data....I dont plan to release my full model, but the spectral data I do. Its not totaly ready yet but its awful close.

Also consider using ms copilot. Its still an llm and uses chatgpt, but I found it to be a much faster number cruncher than gemini.
 
#23
Hell use them all. It's not much extra effort. Good idea with the camera both for yourself and AI playing detective.
 
#24
Grownsince95 said:
What kind of phone are you using? Or is this all on a computer?
Click to expand...
This is all on my PC. I use the phone just for pictures to transfer over, but other than that, I'm still a "type with my fingers not my thumbs" type of guy.

Cheers,

BA.
 
#25
shooter64738 said:
If you really wanna get into some nuts and bolts type stuff without having to write a ton of software...

Build a multispectrum camera. AC infinity released a camera but its not multi spectrum. Phone pics also work, but have limitations.

With one of those you can literally see water in the leaves. You pack all that data up and send it to chatgpt and you can get some of the most detailed data you can imagine. Spider mites (gross) will fluoresce, you can spot leaf damage inside a leaf before its at the surface. There's so much data....I dont plan to release my full model, but the spectral data I do. Its not totaly ready yet but its awful close.

Also consider using ms copilot. Its still an llm and uses chatgpt, but I found it to be a much faster number cruncher than gemini.
Click to expand...
Thanks, shooter. That's some great ideas you've got going on. I'm slowly building up to a fully automated system over the next year or so, and that's definitely something I can consider including in the build. A multispectrum camera is a great tool, one that I've seen some fellow tradesman use in repair work and the like. They use it mostly for the IR for temperature fluctuations, locating studs, water pipes, etc. but you raise a very interesting point about its uses in a grow tent. One more thing to add to the list, then.

Cheers,

BA.
 
#26
ChairmanFester said:
Hell use them all. It's not much extra effort. Good idea with the camera both for yourself and AI playing detective.
Click to expand...
Hi, Chairman. I've tried a few other AI programs, but it seems there's either a limit on how much you can use, or they have built in limitations on how much then can be involved in a cannabis grow due to legalities in other jurisdictions. I get tired of reminding it, "I'm in Ontario...it's perfectly legal here" over and over again...lol.

The camera has indeed played a huge role so far. I take a direct overhead shot of the canopy, a straight on shot for structure and an over all. The AI has managed to pick up a few things that I hadn't noticed (early leaf spotting from too hot an EC when she was still just more than a seedling) and also noted that it hadn't ever progressed so it's *gasp* likely just spotting from too hot an EC when she was still just more than a seedling! 😮

Some things it always seems to look for is nodal spacing, stem thickness, leaf vigor and posturing, etc. and by giving it the info 3 times a day it gets "real time" footage, so to speak. Like I say, ultimately I'm using it strictly as an analysis tool, not a grow coach, as it can crunch numbers much more efficiently than I can.

Cheers,

BA.
 
#28
It can be a coach, or was for me by frequently demanding info therefore making me pay attention more, or more smartly focused. Creating and managing spaces, like making a grow cabinet it seems to get just fine too.
 
#29
amneziaHaze said:
www.thcfarmer.com

Epsilon v2 Plan, 6g per watt goal

Epsilon Protocol v1.1 — 3×3 SOG Grow Bible (Autos, 300 W) Goal: Ultra‑high efficiency and yield in a 3×3 ft (0.91×0.91 m) tent using 36 autoflowers in a 6×6 SOG grid under ~300 W LED, CO₂ enrichment, living‑soil hybrid nutrition, and AI‑assisted...
www.thcfarmer.com www.thcfarmer.com

HE USEs ai a lot
Click to expand...
I read through most of it just now. Amazing concepts and attention to every detail. Would liked to have seen him finish it up though. Alomst 3 months since the last update.

Cheers,

BA.
 
#31
Brother_Antioch said:
Thanks, shooter. That's some great ideas you've got going on. I'm slowly building up to a fully automated system over the next year or so, and that's definitely something I can consider including in the build. A multispectrum camera is a great tool, one that I've seen some fellow tradesman use in repair work and the like. They use it mostly for the IR for temperature fluctuations, locating studs, water pipes, etc. but you raise a very interesting point about its uses in a grow tent. One more thing to add to the list, then.

Cheers,

BA.
Click to expand...
I used a Sony imx462 camera. Its pretty good from uva down in the mid 300s, to near infrared in the 950s. This will let you detect pest in the uva band all the way up to almost inside the leaf. Then you can go to swir and see in the leaf further

Im not a fan of having my info up on the web, so that all gets converted to spectral data, and from there I send it to my ML, but you could also send it to gemini at that point. It also doesnt count against your upload counts because its not a 'photo'.

The data you get from it is insane. You can tell if it's iron deficient, or mag deficiency, too much light, wind burn... its nuts.

This is some papa smurf photoperiods, germinated 12/16, and were only 8 weeks old, already flowering, over 6 feet tall.

This 100% AI. We went up and trimmed some sun leaves a couple weeks back. My model was a little slow on switch from veg to flower sobthey got a littlebtoo tall. The owner of these is a cancer patient and has never seen the plants in real life. AI can do some amazing stuff if its given enough data!
 

Attachments

  • image000000(9).jpg
    image000000(9).jpg
    261.9 KB · Views: 1
#32
Okay, so here's exactly the information I feed into ChatGPT, and then the exact response it gives me. I figure this will show how I'm using it, what the analysis is and how it's broken down, and shows what I can take away from it as far as what I'm doing right/wrong.

Here's a screenshot of the spreadsheet. I upload it as a .csv file and I just re-save it each time I add a line.

Screenshot 2026 03 01 154019


So ChatGPT will analyze the trends of the columns and look for things going up, down (or sideways!) strictly from a numbers point of view. Then I upload a few pictures of it from the same time so it has a visual image of what the numbers mean.

Top view:

20260301 152605


Side view:

20260301 152615


Close up of both new mains' growth:

20260301 152646


20260301 152653


After loading all that into ChatGPT, this is the response it gives me:

me:

Here are the afternoon data sets for you to analyze. Pictures show directly from above, plus a side shot, and 2 close ups of each new main and the new growth. Added 12.5ml H2O2 as documented in the notes. Slight adjustment of ties.

ChatGPT:

Alright — afternoon review locked in.
First: structure.
She handled the strip exactly how we’d want.
• No wilt
• No droop
• No twist
• No leaf clawing
• No color fade
That tells us:
Root pressure is strong
Transpiration is balanced
EC is not excessive
VPD is in range
Now visually:

Canopy Structure​

Both mains are now properly horizontal.
The knuckle at the topping site is forming nicely — thickening already.
Auxiliary shoots at the inner nodes are waking up (this is what we want for future 4 → 8 → 16 progression).
The tie tension looks correct.
Not compressing tissue.
Just guiding.
You’re doing LST correctly — you’re steering, not forcing.

New Growth Analysis​

Close-ups show:
• Tight internodal spacing
• Symmetrical leaf stacking
• No tip burn
• No interveinal chlorosis
• Healthy sheen
• Good turgor pressure
New growth color is slightly lighter green — which is correct. That’s active expansion tissue, not deficiency.
If it were deficiency:
It would show uneven striping or twisted leaf edges.
I see none of that.

Lower Leaf Spots (previous deficiency)​

Stable.
Not spreading.
Margins clean.
That confirms your earlier conclusion: residual from pre-journal EC phase.
No active issue.

Structural Forecast​

Now that everything below the knuckle is stripped:
Energy will redirect into:
• Main arms
• Inner nodes on each arm
Within 48–72 hours you should see:
• Noticeable tip lift
• Increased lateral push from the next node
• Slight thickening of mains
This is the beginning of real canopy architecture.

H₂O₂ Addition​

12.5 mL addition — good call.
Keeps sterile program consistent.
No visible stress response to it.

Critical Observation​

Your plant is not “recovering.”
It is actively growing.
That means:
Stress load was well within tolerance.
Reminder: you stripped during lights-on window (ideal choice).

What I’m Watching Tonight​

When you upload evening logs, I’ll check:
• EC drift (is she feeding harder after strip?)
• Water level change (transpiration increase?)
• Leaf angle at end of light cycle
• Any upward curl (light intensity interaction)
If EC drops slightly faster tonight than previous days — that’s actually positive. It means growth acceleration.

Short-Term Plan (Next 3 Days)​

Do NOT:
• Increase EC yet
• Increase PPFD yet
• Retie aggressively
Let her show you her momentum first.
If growth rate spikes, we’ll respond — not preemptively push.
You did everything correctly today.
This is the point where a plant either stalls…
…or explodes.
Based on structure, vigor, and environment — I’m leaning toward explosion.
Upload tonight’s numbers and we’ll confirm trajectory.
Lab-grade hydro precision continues.

-----------------------------------------------------------------------------------------------------------------------

Basically, it's acting as an analytical engine, and making suggestions as to what to watch for, what to expect, and how to handle things over the next few days. At that point, it's up to me to do a bit of research and ensure that it's not giving me wrong information. So far, it's been really good for that. A few times it's "forgotten" that we talked about something specific, but it doesn't necessarily store every detail of every conversation. It keeps a log of what I'm using, environment, strain, and the rest is in the logs that I give it each time.

Cheers,

BA.
 
#33
Looks good man! Couple of thoughts.
If you add some location specs itll always know.
1. Indoor
2. Tent
3. Lights led/hps etc.. (and their max output)
4. Nute types
5. Species info.

After it gives you back it's response to your csv input, give it a prompt like:

Duplicate the csv data i gave you, and summarize what we discussed into a new column in the far right called ai_summary. Give me back only that response and nothing else.

It should give you back your data plus its summary that you can save in your csv The next time you upload you will have its summary to give to it also. You can reimport the csv, or just copy and paste the summary to your spread sheet for the next export.

Lookin good tho man!
 
#34
shooter64738 said:
Looks good man! Couple of thoughts.
If you add some location specs itll always know.
1. Indoor
2. Tent
3. Lights led/hps etc.. (and their max output)
4. Nute types
5. Species info.

After it gives you back it's response to your csv input, give it a prompt like:

Duplicate the csv data i gave you, and summarize what we discussed into a new column in the far right called ai_summary. Give me back only that response and nothing else.

It should give you back your data plus its summary that you can save in your csv The next time you upload you will have its summary to give to it also. You can reimport the csv, or just copy and paste the summary to your spread sheet for the next export.

Lookin good tho man!
Click to expand...
Thanks, shooter.

Actually, it has all that information already saved as a reference file. I've got a full list of my equipment and how it's being utilized. It is all stored on a second .csv and I have a third .csv that has my nutrient mixes for reservoir changes. I didn't bother including them as I don't upload them daily. The equipment log stays the same (until my Glow30 side lights arrive and I install them in a few weeks) and the reservoir change log gets updated every 10 days when I do the swap out. I had to specifically tell it to save that data and pin it to our conversation thread, though. It was a couple of days into it I realized, "Hmm...I keep having to remind it what lights I have" and upon asking, it told me to upload one more time and it would save a copy this time. Live and learn, right?

The other thing I'm doing is keeping all of my interactions in a single conversational thread. That way, it can always glance back up the previous interactions and see what was said. I found it slows it down a bit, but is worth the extra minute or so in time over the course of our back and forth. Glad you're liking it so far, though. Thanks! As I said, first time trying to use AI for almost anything, and first time trying to grow in hydro, and first time attempting a mainline HST technique. Going for the trifecta!

Cheers,

BA.
 
#35
Brother_Antioch said:
Okay, so here's exactly the information I feed into ChatGPT, and then the exact response it gives me. I figure this will show how I'm using it, what the analysis is and how it's broken down, and shows what I can take away from it as far as what I'm doing right/wrong.

Here's a screenshot of the spreadsheet. I upload it as a .csv file and I just re-save it each time I add a line.

View attachment 2608500

So ChatGPT will analyze the trends of the columns and look for things going up, down (or sideways!) strictly from a numbers point of view. Then I upload a few pictures of it from the same time so it has a visual image of what the numbers mean.

Top view:

View attachment 2608496

Side view:

View attachment 2608497

Close up of both new mains' growth:

View attachment 2608498

View attachment 2608499

After loading all that into ChatGPT, this is the response it gives me:

me:

Here are the afternoon data sets for you to analyze. Pictures show directly from above, plus a side shot, and 2 close ups of each new main and the new growth. Added 12.5ml H2O2 as documented in the notes. Slight adjustment of ties.

ChatGPT:

Alright — afternoon review locked in.
First: structure.
She handled the strip exactly how we’d want.
• No wilt
• No droop
• No twist
• No leaf clawing
• No color fade
That tells us:
Root pressure is strong
Transpiration is balanced
EC is not excessive
VPD is in range
Now visually:

Canopy Structure​

Both mains are now properly horizontal.
The knuckle at the topping site is forming nicely — thickening already.
Auxiliary shoots at the inner nodes are waking up (this is what we want for future 4 → 8 → 16 progression).
The tie tension looks correct.
Not compressing tissue.
Just guiding.
You’re doing LST correctly — you’re steering, not forcing.

New Growth Analysis​

Close-ups show:
• Tight internodal spacing
• Symmetrical leaf stacking
• No tip burn
• No interveinal chlorosis
• Healthy sheen
• Good turgor pressure
New growth color is slightly lighter green — which is correct. That’s active expansion tissue, not deficiency.
If it were deficiency:
It would show uneven striping or twisted leaf edges.
I see none of that.

Lower Leaf Spots (previous deficiency)​

Stable.
Not spreading.
Margins clean.
That confirms your earlier conclusion: residual from pre-journal EC phase.
No active issue.

Structural Forecast​

Now that everything below the knuckle is stripped:
Energy will redirect into:
• Main arms
• Inner nodes on each arm
Within 48–72 hours you should see:
• Noticeable tip lift
• Increased lateral push from the next node
• Slight thickening of mains
This is the beginning of real canopy architecture.

H₂O₂ Addition​

12.5 mL addition — good call.
Keeps sterile program consistent.
No visible stress response to it.

Critical Observation​

Your plant is not “recovering.”
It is actively growing.
That means:
Stress load was well within tolerance.
Reminder: you stripped during lights-on window (ideal choice).

What I’m Watching Tonight​

When you upload evening logs, I’ll check:
• EC drift (is she feeding harder after strip?)
• Water level change (transpiration increase?)
• Leaf angle at end of light cycle
• Any upward curl (light intensity interaction)
If EC drops slightly faster tonight than previous days — that’s actually positive. It means growth acceleration.

Short-Term Plan (Next 3 Days)​

Do NOT:
• Increase EC yet
• Increase PPFD yet
• Retie aggressively
Let her show you her momentum first.
If growth rate spikes, we’ll respond — not preemptively push.
You did everything correctly today.
This is the point where a plant either stalls…
…or explodes.
Based on structure, vigor, and environment — I’m leaning toward explosion.
Upload tonight’s numbers and we’ll confirm trajectory.
Lab-grade hydro precision continues.

-----------------------------------------------------------------------------------------------------------------------

Basically, it's acting as an analytical engine, and making suggestions as to what to watch for, what to expect, and how to handle things over the next few days. At that point, it's up to me to do a bit of research and ensure that it's not giving me wrong information. So far, it's been really good for that. A few times it's "forgotten" that we talked about something specific, but it doesn't necessarily store every detail of every conversation. It keeps a log of what I'm using, environment, strain, and the rest is in the logs that I give it each time.

Cheers,

BA.
Click to expand...
Yeah a couple of things springs to mind:

Short term plan (next 3 days). That is the AI's out of jail card. It will observe what happens and learn from that. It is you who are teaching the AI.
Not the other way around.

Second, the AI already try to sweet talk you. It knows it can not go as far with you, as with Ken. He ate all raw. You don't. So it's less aggressive.

Third thing. Notice how "you" are becoming "we" starting trying to slowly make you and AI buddy buddy.

I'm not saying it's useless but It's like if someone offered me a bucket of shit and i decline. Then they ask "But there is a nice strawberry in the middle of the shit, don't you like strawberry?
 
#36
trygrowup said:
Yeah a couple of things springs to mind:

Short term plan (next 3 days). That is the AI's out of jail card. It will observe what happens and learn from that. It is you who are teaching the AI.
Not the other way around.

Second, the AI already try to sweet talk you. It knows it can not go as far with you, as with Ken. He ate all raw. You don't. So it's less aggressive.

Third thing. Notice how "you" are becoming "we" starting trying to slowly make you and AI buddy buddy.

I'm not saying it's useless but It's like if someone offered me a bucket of shit and i decline. Then they ask "But there is a nice strawberry in the middle of the shit, don't you like strawberry?
Click to expand...

Yeah, I noticed the "buddy buddy" type attitude a few times. It also asked a few times what type of personality I prefer, but I just x'd it out as I take it for what it is, data analysis. I realized ChatGPT is programmed to make users feel comfortable with it, I get it. So I just take it as what it is, and generally just ignore the pleasantries. As for the "next three days" parts (and it always give that sort of thing) it's good for me, as it tells me what to look for, in general, and then I can dig a little more deeply into it using outside sources. Guidance, not instructions. Analysis, not friendly banter.

Actually, that makes me wonder, if I ask it to cut out the friendly banter and familiarity and instead present the data in a scientific manner, would it change its approach? Things that make you go "hmmm..." 🤔

Cheers,

BA.

Edit:

So I just asked it to change its personality to a much more clinical, professional tone rather than a friendly tone, as it would make it "feel" like more of a scientific experimental discussion rather than two buddies discussing a grow. It completely changed its tone and responded:

"

Settings updated•Base style and tone

Understood.
Your personality setting has been updated to Professional.
Going forward:
  • Tone will be clinical and precise.
  • Minimal affirmation or conversational padding.
  • Emphasis on data, causality, measurable variables, and predictive modeling.
  • Direct identification of deviations, risks, and optimization opportunities.
"Your grow will be treated as a controlled hydroponic experiment with defined inputs, monitored outputs, and iterative calibration.
If at any point you want further refinement (more terse, more analytical depth, more hypothesis-driven structure, etc.), specify the adjustment."

-----------------

I guess only time will tell!
 
Last edited:
#37
Brother_Antioch said:
Yeah, I noticed the "buddy buddy" type attitude a few times. It also asked a few times what type of personality I prefer, but I just x'd it out as I take it for what it is, data analysis. I realized ChatGPT is programmed to make users feel comfortable with it, I get it. So I just take it as what it is, and generally just ignore the pleasantries. As for the "next three days" parts (and it always give that sort of thing) it's good for me, as it tells me what to look for, in general, and then I can dig a little more deeply into it using outside sources. Guidance, not instructions. Analysis, not friendly banter.

Actually, that makes me wonder, if I ask it to cut out the friendly banter and familiarity and instead present the data in a scientific manner, would it change its approach? Things that make you go "hmmm..." 🤔

Cheers,

BA.
Click to expand...
Good idea. yeah make it leave out the "we" part LoL
 
#39
Most of the LLMs are designed to have a 'conversation' with the user. Its handy in some ways, especially if you talk to it.

But in all instances you gotta give it good data, and some historical context will always help.

If you ask gemini:
"Will a cannabis plant stop using water when its done." It will tell you yes it will.

But if you tell gemini:
'I let my coco dry out for 2 days and then watered it. The plant has looked bad since then. Its not using water and I think its done early because its only 4 weeks old and the leaves are yellowing. Is the plant done growing?'.. you will get a drastically different answer. :)
 
#40
shooter64738 said:
But in all instances you gotta give it good data, and some historical context will always help.
Click to expand...
It wants data.
That is the main goal for AI at the moment. The computer that were finished around 2020 is extremely data hungry and very recourse hungry.
Yes it has other goals, but right now it's like that.

You give AI data. AI throws you a bone.
You give AI better data. AI throws you better bone making sure you are the giver.

AI other goal is to make you trust it more than people. The buddy buddy part and the sweet talking part and the turning you against others part
Lets just forget all about that.

And all is working according to plan

A great bucket of shit with a strawberry in it.

Some choose to eat it
 
Page 2 of 3 · Replies 21–40 of 50
Back
Top Bottom