bgamari / bayes-stack Goto Github PK
View Code? Open in Web Editor NEWFramework for Gibbs sampling of probabilistic models
License: BSD 3-Clause "New" or "Revised" License
Framework for Gibbs sampling of probabilistic models
License: BSD 3-Clause "New" or "Revised" License
If the last iteration is 4000
the last output is
Sweep 3980: -40.879208771211445
You should also print the last iteration.
$ bayes-stack-st --help
run-ci - learn citation influence model
not ci.
Usage: bayes-stack-st ...
Learn citation influence model
No, shared taste model
Available options:
-e,--edges FILE File containing arcs
are you sure you want to call it edges (as opposed to arcs)?
provide an option to set all Dirichlet Hyperparameters (e.g. alpha_lambda, alpha_gamma_own)
They are either intial values to hyperparam estimation, or -- if hyper is off -- used throughout the inference loop.
It is unclear where diagnostic information is written to.
Which file is the per iteration model likelihood written?
Which file is the hyperparameter value as well before/after likelihood written to?
Maybe these files should be configurable - although I am also happy if they end up in the sweeps directory.
Usage: bayes-stack-dump-lda [-n|--n-elems ARG](-d|--dist ARG) (-s|--sweeps ARG) [-n|--number ARG]
-n,--n-elems ARG Number of elements to output from each distribution
n-elems is weird, please rename to --top
Probability
should probably just be a newtype
wrapper around a numeric type to indicate that it is a normalized quantity.
$ bayes-stack-st
Read 0 stopwords
bayes-stack-st: arcs: openFile: does not exist (No such file or directory)
the call should print usage information.
(works fine for lda and ci)
Just to confirm: if parameter --n-elems is missing, the whole range should be printed
if I have a topic model with 10 topics, and I call
bayes-stack-dump-lda --dist=theta
it should have the same effect as printing
bayes-stack-dump-lda --dist=theta --number 10
params="--burnin=100 --lag=20 --iterations=200 --update-block=1 --hyper-burnin=20 --hyper-lag=20"
LDA and ST
Sweep 3980: -29.191862221605028
I only asked for 200 iterations
I can't test CI as just brings up the usage screen but refuses to do anything useful :(
Usage: bayes-stack-dump-lda [-n|--n-elems ARG](-d|--dist ARG) (-s|--sweeps ARG) [-n|--number ARG]
Dump distributions from an LDA sweep
Available options:
-n,--n-elems ARG Number of elements to output from each distribution
-n,--number ARG The sweep number to dump
which is -n ?
-u,--update-block N Number of update diffs to batch before updating global state
rename to --diff-batch
In the documentation I write:
One global diff-worker will apply the diffs in batches of XXX
While the diff queue and work queue shouldn't be terribly hot, it wouldn't hurt to use a lockfree deque.
One shouldn't need to resort to RTS options to change this.
[aristoteles ~/bayesstack/exp] $ bayes-stack-dump-st --dist=gammas -n20 $lastsweep
dump-lda - Dump distributions from an shared taste model sweep
Usage: bayes-stack-dump-st [-n|--n-elems ARG](-d|--dist ARG) FILE
Dump distributions from an shared taste model sweep
Available options:
-h,--help Show this help text
-n,--n-elems ARG Number of elements to output from each distribution
-d,--dist ARG Which distribution to output (psis, phis, lambdas, or omegas)
Currently import code expects nodes to be given by numeric node IDs. This is unlike items, where the UniqueKey
infrastructure is used to map string names to numeric keys. This should be fixed.
[aristoteles ~/bayesstack/exp] $ echo $lastsweep
stsweeps/00140
[aristoteles ~/bayesstack/exp] $ bayes-stack-dump-st --dist=phis -n20 $lastsweep
bayes-stack-dump-st: sweeps/item-map: openBinaryFile: does not exist (No such file or directory)
it is using sweeps instead of stsweeps
(same for lda and ci)
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.