dataHaskell/Lobby

Knowledge base: http://datahaskell.org/docs

Alejandro Alcalde

@elbaulp

@ocramz the section “Data Structures” in here for example. http://www.datahaskell.org/docs/community/current-environment.html#data-structures

Marco Z

@ocramz

@elbaulp the current-environment page is a directory of libraries generally built by others

Alejandro Alcalde

@elbaulp

@ocramz I see... I will focus by now in learning Haskell and later on will try to see if I can contribute.

Ada922

@Ada922

Tomorrow Algorand is hosting a live webinar on implementing the world’s first CBDC. I have submitted a few questions and will be attending. This is a great way to spend your quarantine :)

Karl-Oskar Rikås

@funrep

anyone familiar with random-fu who knows why gamma does not define a pdf?

3 replies

Alexey Kuleshevich

@lehins

@funrep Not sure, but I know a person who might know the answer to that question. @idontgetoutmuch any ideas? ^

Karl-Oskar Rikås

@funrep

not a big issue because it's defined in the statistics package, just wondered in case there is some mathematical reason im unaware of :)

anyone know how I can make a set a tensor.require_grad = True in hasktorch?

Karl-Oskar Rikås

@funrep

https://gist.github.com/funrep/ced4c79fd97cbc945e8a5a83c4b30dc8 not really sure I understand how to use the grad API, i want to perform AD using hasktorch to compute the gradient in stochastic gradient descent

would be cool to use the optimizer api directly, but starting out with just the grad as I didn't find any bindings to zero_grad

Austin Huang

@austinvhuang

Screenshot 2020-05-31 at 12.28.18 PM.png

@funrep Just saw this. I'm in the middle of writing up a step-by-step tutorial in fact... here's a rough draft of the ad intro

Screenshot 2020-05-31 at 12.28.36 PM.png

super early version... this is for untyped tensors, working on sections for typed tensors. will have more on optimization but hopefully this at least gets the basic ideas on AD and how requires_grad looks like from the Haskell side of things

Karl-Oskar Rikås

@funrep

@austinvhuang hey thanks a lot, where will the tutorial be published?

Marco Z

@ocramz

@funrep you could use the 'gamma' package and implement the Gamma pdf from the Gamma function

Austin Huang

@austinvhuang

@funrep will eventually get the writeup on arxiv, but I think I'll also make a markdown version to include in the repo. there's a section that's basically a port of pytorch's 60 min blitz.

masterdezign

@masterdezign

Hi @austinvhuang, just wonder if the tutorial is already available

Adam Conner-Sax

@adamConnerSax

Hi! I put a new version of “knit-haskell” (https://hackage.haskell.org/package/knit-haskell-0.8.0.0) on hackage today. Adds a few features. The most useful one, for my workflow, is a caching system. Allows caching of anything serializable to an in-memory map and also disk. Has some basic but useful support for tracking when cached items need to be rebuilt. See the readme and examples for more info! All feedback/ideas most welcome.

NotBru

@NotBru

Hi there! I'm a physics graduate trying to think of a way to delve into AI and data science, but I'm quite lost as to where to start

Tony Day

@tonyday567

@NotBru hi!

Tony Day

@tonyday567

I'm working on a really interesting, haskelly side-project called ephemeral.
https://github.com/tonyday567/ephemeral/blob/master/src/Ephemeral/Learning.hs#L15
I'm connecting up a GA library, moo, with an auto-differentiation library, backprop. Trying to nail down what it means to Learn.

That quote is like page 1 of the cousera course. And I think it's a great start. I'm trying to think in types.

NotBru

@NotBru

@tonyday567 thanks a lot for the answer! will check it out. I'm still just delving into Haskell, too, so it's gonna take me some time

Tony Day

@tonyday567

We have time!

masterdezign

@masterdezign

@NotBru Hi! I have a small blog on deep learning in Haskell http://penkovsky.com/neural-networks/

Austin Huang

@austinvhuang

@masterdezign - it's not online yet but I can share a wip draft with you.

Tony Day

@tonyday567

I'm finally getting a productive covid workflow and have a few announcements on library releases.

Tony Day

@tonyday567

https://hackage.haskell.org/package/numhask is an alternative prelude, that plugs in an industrial strength numeric class tower, if needed. It wraps protolude from the wonderful @sdiehl, is inclined towards Text as the string primitive, and has less partial functions than is good for you.

Everyone should use an alternative prelude, and this is mine.

Tony Day

@tonyday567

Upstream, I'm pleased to announce that https://hackage.haskell.org/package/chart-svg has become quite stable, and better documented. A chart server is included that is quite addictive to use in prettying up a chart. SVG provides an exact medium in which to render, and is a versatile but stable API. Give it a go!

Alexey Kuleshevich

@lehins

@tonyday567 chart-svg looks amazing!!!!!

NotBru

@NotBru

@NotBru Hi! I have a small blog on deep learning in Haskell http://penkovsky.com/neural-networks/

Thanks!!

Tony Day

@tonyday567

Thanks @lehins

Tony Day

@tonyday567

@lehins hook it up to massive and you could render a movie I reckon.

Alexey Kuleshevich

@lehins

:smile:

Alexey Kuleshevich

@lehins

Later on I will give chart-svg a shot for plotting image histograms in HIP. I don't want to use Chart library for this because it depends on lens, which brings the whole kmettoverse with it. So I am really stoked to get a lightweight solution for plots in pure Haskell.

Tony Day

@tonyday567

lens is a dependency, but I use it heavily in connection with OverloadedLabels, so there is zero lens usage requirement. If I ever had the inclination, I could store a Chart it in a dhall file.

In chart-svg, histograms are equally-spaced rectangles (RectA's). There's support in numhask-space for a histogram primitive, but Chart.Bar probably has everything you need for a nice-looking, lightweight chart.

Alexey Kuleshevich

@lehins

lens is a dependency

Dammit, didn't look at that. In that case I'll have to weigh if it is worth to switch from Charts. I prefer to keep my packages very light on dependencies and lens with all its baggage is not acceptable for me. I might release a separate package that just does the image histogram plotting (that my original plan for Charts at least).

Tony Day

@tonyday567

Yeah, it's easy. Just shapes!

Marco Z

@ocramz

what's up gang

Stefan Dresselhaus

@Drezil

waiting for haskell.love ;)

Marco Z

@ocramz

is it today?

Alexey Kuleshevich

@lehins

nah, over the next weekend

Austin Huang

@austinvhuang

@NickSeagull Do you see tintin being developed further in the future?

Nick Tchayka

@NickSeagull

@austinvhuang We're not actively developing this project. We're using it in a few other projects, like the AWS lambda runtime, so we'll be occasionally updating it if we find a non met requirement there, but we'd be glad to review and merge community contributions.

Randall Britten

@vertexcite

Bumped into this at ICFP today: https://dl.acm.org/doi/10.1145/3406088.3409019
See also https://github.com/bolt12/laop (I think).

Callan McGill

@Boarders

Does anyone familiar with massiv know how parallelization works with a function like this: https://hackage.haskell.org/package/massiv-0.5.4.0/docs/Data-Massiv-Array.html#v:mapIO_

Alexey Kuleshevich

@lehins

@Boarders it will work in exactly same way as when the delayed array is being loaded: split the array in as many chunks as there are cores and map an IO action over each of them in a row major fashion in parallel. I talk a bit about in here: https://youtu.be/AAx2a0bUsxA?t=2215

Where communities thrive

People

Repo info

Activity