freenode/#lisp - IRC Chatlog
Search
2:09:09
PuercoPope
Xach: Don't know if you are aware but https://www.xach.com/clhs has been returning 502 for the last few days. Its my goto search method when outside of Emacs.
4:00:13
fiddlerwoaroof
whartung: before I learned CL, I implemented restarts myself in Python to facilitate a major data migration from a schemaless database to one with a strictly enforced schema
4:00:48
fiddlerwoaroof
It saved me numerous cases of "wait half an hour for a crash to happen, change the code and wait half an hour to see if the fix worked"
4:01:56
fiddlerwoaroof
Although this was a semi-interactive situation: when there was an exception, I'd inspect the data, add a rule to handle it to a collection of rules and then restart the migration process which could now handle the new category of malformed data automatically.
11:08:30
shka__
when two initargs are provided for the slot, which one takes the precedence if both are present in the make-instance call?
11:15:18
specbot
Rules for Initialization Arguments: http://www.lispworks.com/reference/HyperSpec/Body/07_ad.htm
11:40:35
Xach
PuercoPope: sorry about that. i have moved servers and there have been some regressions. I will try to fix asap.
12:18:53
katco
hey all! i'm learning about machine learning, and i was wondering if i'd be able to use CL as i go for prototyping and such. what's the state of ML in the CL space these days? cliki doesn't list many libs.
12:22:33
beach
katco: Common Lisp is a general-purpose language so you can certainly use it where other languages can be used as well.
12:22:51
beach
katco: Common Lisp has many advantages compared to other languages, independently of the domain.
12:23:26
beach
katco: Not many people use Common Lisp these days, including for machine learning, so it might be tricky to find libraries for that particular domain.
12:25:43
katco
i am aware of the unfortunate fact that CL has fell out of favor. and that's why i'm reaching out to the community to see if maybe there's some well-supported cffi bindings to one of the popular ML frameworks, or if anyone has had some success with something like burgled-batteries, or is even growing a well-supported CL framework
12:28:47
heisig
katco: https://github.com/bendudson/py4cl could also be useful for using Python ML libraries from CL.
12:30:48
katco
heisig: i don't use py bindings that much. is this the community-preferred way to consume python?
12:33:32
katco
i am brand new to the ML space, but it looks like tensorflow is working hard to expose all of their bindings via C, so i wonder if cffi might be the preferred way to consume that
12:37:59
katco
yeah, i came across mgl, and it looked interesting. unfortunately i'll be sharing my work with non lispers, so i think i need to be leveraging something they can use/understand in `(not :lisp)`
12:39:23
p_l
which means the pains of Python every time I hit something mentioning ML or one of the many names for applied statistics
12:40:34
heisig
katco: Maybe cl4py (https://github.com/marcoheisig/cl4py) can help you present your Lisp libraries to non-Lispers.
12:55:54
p_l
katco: CFFI to TensorFlow could be done pretty fast, the bigger issue would be that TF C API is, from my understanding, pretty low-level
12:59:38
katco
p_l: that surprises me a little. i assumed since they were putting the c api forth as the way for languages to interop with tf, that it would have the same level of semantics as the py bindings
13:01:11
katco
p_l: ah, i see. well, maybe that's the correct approach? every language has its idioms
13:01:52
p_l
to quote the docs, C API is made for regularity and consistence rather than ease of use
13:07:40
katco
i'm also trying to balance how productive i'd be in CL vs. yak-shaving just so i can use it haha
13:12:11
katco
my understanding is that pytorch is "better" for research, and is preferred by the academic crowd
13:19:53
katco
shka__: yeah? would you mind expounding on that? i'm very open to being influenced atm :)
13:24:08
katco
i was kind of leaning towards tf because it looked like the ops side of things was more production-ready, and the ecosystem was broader
13:34:19
katco
ha! i was also looking at keras because it was suggested that it was good for learning ML