Run tensorflow for the first time

If all you are doing is training shit you are legit worst that a undergrad with 4 years of work experience. Yikes

because to engineer the hyperparameters (network structure etc) and to keep up with the current research you gotta have a minimum of math knowledge. but not necessarily a "really good grasp" as other user said, just a good grasp. this shit is mostly keeping up with the state of the art empirical research.
this. wtf do you mean "not accounted for"? do you think you need to manually tag those terms for the botnet to figure out it has to consider them together?
>Thanks for trying to demoralize me though.
heh, don't mention it

What the fuck no I don’t need anything that powerful for what I do.
It won’t pop because ML has practical applications many businesses haven’t considered. You ML dweebs are just mad no one cares about your ~science~. Newsflash. A bunch of stats mean jack shit to decision makers. Learn to communicate your findings better.
I wasn’t even thinking startup or diagnosis (since making a diagnosis is rather involved) but that’s for the idea.
If the model does work for me that I don’t have to do then idgaf

As in I already have 8 years of tagged shit I can use for a specific task and save ppl from another 8 years of monotonous suffering you absolute sperg.

>It won’t pop because ML has practical applications many businesses haven’t considered.
Literally everything that is possible has already been tried, thanks to the ML buzzword bubble that's been going on recently.
ML is nothing new, the math behind it existed for at least 40 years and all that improved is hardware, nothing new is being invented despite all that money flowing into industry.
It's a fucking meme of worst sort.

Look man I’m not trying to start some ground breaking business I’m literally just making prototypes to replace NPCs at my work out of spite. Calm down.

This. The worst part is going to be after it pops the other specialities are going yo be flooded with master students only capable of writing shit tier python.

What other languages are good for ML? Python has most of the libraries but is admittedly slow. Would you recommend something like C++ and if so, could you recommend reading material for C++ based ML?

>Learn to communicate your findings better.
You are literally retard tier. People working in medicine really have the lowest IQs and the most inflated egos.
You do realize the tech giants who are leading in machine learning have entire marketing departments filled with psychologists, marketers and business graduates dedicated to shilling their shit, right? Machine learning and it's various related statistics shit are used in industries and places you wouldn't even think about. Basically anywhere it's usable, it's already being used.
youtube.com/watch?v=GmxkzaAraGw

So your point is that nobody has ever tried ML on your dataset?

The problem isn't python is a bad language necessarily but because you don't need to know how to program to sling code and implement a library. You can use the same libraries with C++ just look at the library most have other language support like tensorflow supports c++ the logic is exactly the same its just different langyage but the same api calls.

It isn’t being used at my company. So I will exploit that. I’m not trying to make some new ground breaking thing. I’m just applying it in a company that still thinks SAS is the top tier of all time. Why is everyone so angry about ML? Christ.

The data set as it is used in my specific team in my specific company. Correct. If it was already used, then 3/4 of my team would not need to exist.

I see. I’m trying to learn more about the math behind logistic regression and individual neurons in tensorflow actually. I know a neuron sums values it receives, engages in matrix multiplication, then max(0, n)’s it. But I can’t find a good resource that explains it from a fundamental start. Any recommendations where I can really learn about the math behind neural networks (basic ones for now like convolutional) ?

The results with ML, unless you have google-level data, hardware and programmers are almost always so shit you need about as much menpower to verify and correct the results as to manually tag them in the first place. What are you trying to achieve anyways?

youtube.com/watch?v=aircAruvnKk&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi
this might be too basic though

the most basic one is a perceptron, actually
to learn the inner workings begin by implementing a solver for the NIST number challenge from scratch. then compare your results to the state of the art networks and get depressed

I want to replace NPCs (also known as paper pushers, administrative staff, data entry people, etc). From my experience in the medical research industry, these incompetent “people” are the main source of red tape preventing actual useful meds from reaching patients who need them.

It’s a start. Thank you user. Any recommendations to follow this vid?

it's a playlist

I know. I more meant things to literally read. Books, articles, etc.

They don't understand economics or reality. The idea that there is low-hanging fruit is completely outside of their worldview, which is why their biggest dream is to work for jews rather than start a business.

Idk I’m just tired of seeing the everyday be so boring. It’s the future. I want shit to be futuristic. So I’ll do what I can to make that a reality with a bunch of statistics formulas crammed into a class.

Except OP said he's an employee trying to make his employer happy...

You're mostly right, except neurons don't sum their inputs. Well, they do, but implicitly through the matrix multiplication. Also there's a bias (constant) added before the relu which also evolves due to backprop. This is important because allows a layer to create discontinuities anywhere.

You didn't mention backprop or optimization though which are the most important concepts. Make sure you understand SGD, how gradient calculations work vs a loss function (backprop/chain rule), and then try to understand something like Adam. You don't even need to mess with CNNs until you at least get this. Try just fitting a function like x^3 with a small network first and watch what it does.