Alright /gee what deep learning and big data projects y'all been working on.
Next thread I make, will add resources for some of you who might be interested in getting started.
>this gets 0 replies
>nvidia gaming post #124 is probably in the making as I'm writing this
/g/ - Technology everyone
Daily reminder that "data science" is just a marketing term for statistics.
>my ultra deep neural learning machine vision network will change the way we work with data!
>is literally just linear regression in Python
Can you post anything that's not Python shit?
>>54428688
Have some patience, dumb shit. Your thread wasn't even 5 minutes old when you started complaining about people not responding.
>>54428621
>>54428688
Nigga you made a post about a niche as fuck subject. If you had put the resource links in this time instead of being lazy and saying you'll do it next time, you'd probably have gotten more replies.
Tl;Dr post resources, homo.
>>54428751
Sorry to disappoint m80 I was on my phone or else I would have posted links if I had my laptop near me. Forgive me /g I shall not disappoint you again.
>>54428734
That's not OP, I am OP here. I didn't self bump lmao.
deep learning and big data are a meme
next thing i'm working on is a bot to play a certain japanese rhythm game, might use some conv-nets to learn the visual cues or i may just select the features myself since that will be 10x easier and work 10x faster
>PHD in statistics
>90k$ starting
>any R job I want
>>54428621
AI is getting better as we speak
>>54429370
>deep learning and big data are a meme
>next thing i'm working on is a bot to play a certain japanese rhythm game
of course they're "a meme" to you, you're just some bedroom programmer loser
>>54429817
Sauce
>>54429989
ask the op
What kind of rigs do you guys have? Finna drop a fat wad on a desktop. Should I go for a mobo that can support 128 GB of RAM?
I use R and Python.
>>54428704
Statistics without knowing statistics. Just dump data into a black box.
>>54428621
I don't know much about deep learning outside of some of the results it can provide.
Would it be possible to apply deep learning in order to single out features from a georeferenced orthomosaic?
Specifically, it would be useful to have a tool that detects and draws georeferenced shapefiles (polygons) over unwanted vegetation like weeds in any kind of crop, it is both useful in the process of its removal and assessment of its coverage area.
The orthomosaic attached was captured and stitched using a NIR-GB sensor aboard an UAV, I've attempted to perform semi automated, supervised, and unsupervised classifications on this orthomosaic with bad results, the software always ends up interpreting some weeds as corn, and some corn as weeds.
>>54428621
This image, kek. Sorry to disappoint you OP but in current business world bigdata is simply feeding mapreduce machine learning with gigabytes of data about users.
I actually worked as a fullstack in bigdata integration. Through 11 months time on website of 13 kk RU on various sample groups we hade about 9% ctr raise and pretty much no other benefits. Depending on the chosen algorithm we even had downs on, for example, dwell time.
>>54428621
I've been working on what I guess you could call an AI for a few months.
I've been feeding a literary analytical algorithm I wrote eBooks to analyze writing styles of the great and not-so great authors with the end goal of the project is for this thing to use what it learns to write first short stories and then books.
its been chewing on classic fiction for about a month and its still spitting out incoherent but syntactically correct sentences. Its getting better though kind of creepy
Couchbase is my ODS running on a 4 NUC cluster and I have another cluster of 3 NUCs running my C and erlang server program which do the analytics and I/O to and from Couchbase.
>>54430509
>its still spitting out incoherent but syntactically correct sentences. Its getting better though kind of creepy
Post some pastebins, anon
>>54430417
I am pretty sure this is possible. What you would need though is a large number of training images. An interesting problem will be that vegetation and landscape vary widely from region to region, so any solution you would come up with would probably have to be tailored to a specific region or even farm. You could use something like a CNN with MANY images of large patches ( from the perspective of your UAV), then train using MANY test images where said patches of unwanted vegetation exist (i.e. would need to label them yourself). If you want an idea of how this works I would check out the two CNN
TensorBoard tutorials:
https://www.tensorflow.org/versions/r0.8/tutorials/index.html
>>54430685
Thanks anon, I'll give it a go, my programming and maths background is moderate, but I think I can invest enough time to make up for it.
>>54430355
I don't think you need 128GB of RAM.
>>54431330
64 okay? I'm being cautious since RAM is cheap enough.
>>54430355
What sort of inefficient implementations could you be using where you would require anything more than 8 GB of RAM? You should be focusing on getting a good CUDA card that run tons of comps/sec
>>54430961
>>54431467
When the meme so supreme you let out a scream
>>54429272
why even make the thread to begin with if that's the case, sigh
>>54428704
except its not retard
source: degrees in stats and cs
>>54430417
>I've attempted to perform semi automated, supervised, and unsupervised classifications on this orthomosaic with bad results, the software always ends up interpreting some weeds as corn, and some corn as weeds
This is very likely due to bad feature selection. Don't know what you've used but I've seen really simple algorithms (even ID3) blow away NN's in certain cases with some clever feature selection/extraction tricks.
You could use e.g. the fact that your image is a skewed version due to the perspective to come up with some fancy filter.
>>54428621
Learning about it right now for my course project.
Does anybody have a good resource to learn about Boltzmann machines? I'm kinda interested in seeing if they can help me in my image recognition task.
I used this book to learn everything I know at this point: http://neuralnetworksanddeeplearning.com/index.html
I tried reading through this guy's other book, but I can't be bothered to reread the first two sections, and the third one just defeats me with excessive math, and I'm not sure whether this means I need to learn more math or if he just went crazy on it because this one actually will get published. Is there a resource that explains it in simpler and more comprehensive terms, or should I just suck it up and wade through the math?
>>54430355
Just get 128, get that ram disk on niqqa
