All this money and attention will surely pump up the R stock/s like crazy. Not sure if these investors are also receiving any preferntial liquidation preference too. Is it not likely that they might dump stock to book huge profit and the retail Indian investors will be left holding the bag ?
No, that is the training agent implemented as an instance of quicksort. I don't know why the OP reports that it's "a few percent faster than quicksort" because I can't find where that is claimed in the paper.
In fact, as far as I understand it the paper claims that the learned model (i.e. the student agent) has learned to reproduce the behaviour of this teacher agent after training for a smaller number of steps on average than what the teacher needs to sort a list. That is what the entire claim about superior efficiency rests on (there is no example of a learned model and no asymptotic analysis of the program such a model represents).
As I understand it they designed quicksort like that to be able to train with it. It is quite clear from the video where it is called "quick sort agent" compared to the model one and function1 and function2 is in the stack trace.
"We found adding the previously executed action to the input state st is sufficient to handle dis-ambiguation for this quick sort implementation. Alg. 8 shows the converted quick sort scripted agent"
Right! I was hoping to see the body of one of their system induced algorithms. They don't seem to have included any - or maybe I am mistaken and their system is opaque and does not allow a generated algo. to be read out...
Good starter project to flex your JS muscle. But other than that - Zoom's built-in whiteboard is pretty feature-rich and meets my use-cases perfectly. My use-cases -: brainstorming with colleagues, interviews.
Will things like WP (not in particular) die out eventually because of poor developer experience ?
Corollary: frameworks liked by developers will eventually stick even if currently they have a modest userbase.
Stanford universiry manages a tool called Protege (cf protege.stanford.edu). It is now fully RDF/OWL, I presume. But in the last decade, there was a dedicated version for FRAMES. (I Think this link is still valid: https://protege.stanford.edu/download/protege/3.4/installany...)
I liked the boldness of this idea.
But 'something' needs to select the sklearn model, tune its hyper-params - how long can you keep it all hidden away from the user?
The training phase can be considerably long. Have you thought of some kind of an async wrapper that Smart Fruit might provide or will the user be expected to code it up?
This is more of a user experience comment - when the interface is designed to feel as if one is interacting with a DB / ORM the user may come to assume that the outcomes will be deterministic... While the returned results will remain deterministic given the training data, model and hyper-parameters remain the same - it won't feel as deterministic when either of these is updated... I am not sure if I communicated my concern clearly. I am trying to understand who the intended end-user is, of this package...
I would propose a potential user as someone interested in some of the meta considerations and patterns of statistical reasoning, aka machine learning. There are is a vast amount of particulars the second hand on my watch operates (e.g. vibrating quartz, digital), but I can use that mostly reliable device to investigate higher level phenomenom, like calculating distance of planets by timing their movement. This library opens a direct line to these algorithims such that one might intuit, and apply, their high level behavior; as I could not time planets if consumed with the fidelitity and reliability of resonating quartz, it would slow my ability to explore this kind of reasoning if concerned with the minutiae.
That said, all points taken. If this sparks interest in someone, as is stands, it would be on them to dig in to all the considerations you've outline.
A logistic regression model is typically used for a classification task. 'Fitting a logistic model' entails finding optimal coefficients / 'weights' of input features such that classification error is minimised.
For a binary classification task, one could simply calculate mean squared error between predicted values and actual labels (as in linear regression) and then proceed to find the optimal weights iteratively using gradient descent. But the sigmoid shape of the logistic function makes gradient descent a poor choice of an optimization technique (w.r.t. lack of guarantee of finding a global optimum).
A surer way to find globally optimal weights is using the Newton's method of calculating weight updates. This is a numerical optimization technique that requires one to calculate the 1st and 2nd order derivatives of the error function. The matrix that 'calculates' the 1st order derivative is called a Jacobian and the one that calculates the 2nd order derivative is called a Hessian...
