DAY 1 [11th November 2024]
- So to be honest I am not sure how this should be documented, more on the professional side or causal? But who cares as long as the point gets across eventually (LOL)
- So I officially declare AI to be my best friend, in fact AI is every aspiring coder’s best friend !
- Perplexity AI gave step by step guide on what to do to get this AI model created within the span of 5 days with only 2hrs everyday (based on my prompt)
- this first day main focus was on getting my Google Colab set up and getting my dataset all prepared and well organised
- I already had my whole dataset prepared from scratch but then realised I needed images of crops as well for my AI if not it could very easily mistake crops for weeds obviously I couldn’t start collecting all crops in the world so I decided to quickly collect images for wheat which is very common and also of grass as one of the weeds could very easily be mistaken for grass(literally called black grass..)
- after getting that fixed i finally got onto the actual setting up of Google Colab
- It was so easy to start on Google Colab! Just click new notebook - like…after all those videos taking good 10-15mins to set up and download multiple files I was so happy I did not have to go through the stress like..-not today sir.
- Everything went smoothly mostly copy and pasting code given by perplexity AI - obviously I still tried to get a grasp and understanding of what I actually was copy and pasting and there would be errors at times which I really had to sit down and ponder on what to do to make it work
- The debugging was stressful but once I got it I would be so happy! On top of the moon kind of happy! Although not sure if being on top of the moon would make me happy..
- for day 1 I was only meant to just organise my data set and then set up my Google Colab and load the data into it but I ended up getting to the training of the model stage , getting me ahead with my work - i really keep surprising myself! Can’t wait for tomorrow.
DAY 2[12th November 2024]
DAY 3 [13th November 2024]
- had a graph designed based on initial results
- based on the graph the training of the AI was good however when it came to it getting validated(so shown to unfamiliar images) the scores were very low it had an overall validation score of 0.24 which is quite low
- so went to find out how to increase its accuracy and added more layers with an extra dense layer added in as well
- despite increasing the layers the validation accuracy was still low and did not increase by a bit so now I will try to tweak the hyperparameter
- so I tried hyperparameter tuning and well damn…my RAM crashed (LOL..sorry can’t take this seriously)
- I will have to find ways to minimize power used and got lots of suggestions from perplexity AI