blei-lab / hdp Goto Github PK

Please give a proper training set and a good readme. Thanks in advance

Where can I find the python code for the classical SVI paper? The following link is not valid any more.
"http://www.cs.princeton.edu/~blei/topicmodeling.html"
"Hoffman, M., Blei, D. M., Wang, C., & Paisley, J. (2012). Stochastic Variational Inference. 14, 1303–1347. Retrieved from http://arxiv.org/abs/1206.7051"

I am trying to install hdp, and I get this error:

In file included from state.h:4:0,
from state.cpp:1:
corpus.h: In constructor ‘document::document()’:
corpus.h:18:17: error: ‘NULL’ was not declared in this scope
words = NULL;
^
corpus.h: In destructor ‘document::~document()’:
corpus.h:34:22: error: ‘NULL’ was not declared in this scope
if (words != NULL)
^
In file included from state.cpp:1:0:
state.h: At global scope:
state.h:146:63: error: ‘NULL’ was not declared in this scope
hdp_state * target_state=NULL);
^
In file included from state.h:4:0,
from hdp.h:4,
from hdp.cpp:1:
corpus.h: In constructor ‘document::document()’:
corpus.h:18:17: error: ‘NULL’ was not declared in this scope
words = NULL;
^
corpus.h: In destructor ‘document::~document()’:
corpus.h:34:22: error: ‘NULL’ was not declared in this scope
if (words != NULL)
^
In file included from hdp.h:4:0,
from hdp.cpp:1:
state.h: At global scope:
state.h:146:63: error: ‘NULL’ was not declared in this scope
hdp_state * target_state=NULL);
^
hdp.cpp: In member function ‘void hdp::run(const char*)’:
hdp.cpp:147:97: error: call to ‘double hdp_state::split_sampling(int, int, int, int, int, hdp_state*)’ uses the default argument for parameter 6, which is not yet defined
double prob_split = proposed_state->split_sampling(num_scans, d0, d1, t0, t1);
^
Makefile:13: recipe for target 'hdp' failed
make[1]: *** [hdp] Error 1
make[1]: Leaving directory '/srv/scratch/shared/surya/imk1/TFBindingPredictionProject/src/hdp/hdp'
Makefile:4: recipe for target 'make_all' failed
make: *** [make_all] Error 2

I installed the latest version of gsl, and I think that I updated the Makefiles in hdp and hdp-faster correctly. Do you know what might be causing this problem?

Thanks so much!

In the code of state.cpp, the way to calculate the likelihood in table assignment is different from that in the word assignment, and I didn't find the details on the paper. confused by this for days, and still find little information, really appreciate it if someone could advice some details or related materials. Thank you!

gsl: ../gsl/gsl_rng.h:200: ERROR: invalid n, either 0 or exceeds maximum value of generator
Default GSL error handler invoked.
how can I fix it
thanks!

Hello, I have been testing the function of hdp recently, but in the process of testing, I will encounter the problem that the process is killed due to the excessive occupation of environmental resources. Is there something wrong with my usage or is the test failure caused by my own environmental restrictions?

ls data.txt

3 2 6
0:1 1:2 2:3
0:2 1:1 2:3
0:3 1:1 2:2
0:1 1:3 2:2
0:2 1:2 2:2
0:1 1:2 2:3

hdp --algorithm train --data data.txt  --directory train_dir

Program starts with following parameters:
algorithm:          = train
data_path:          = data.txt
directory:          = train_dir
max_iter            = 1000
save_lag            = 100
init_topics         = 0
random_seed         = 1716260426
gamma_a             = 1.00
gamma_b             = 1.00
alpha_a             = 1.00
alpha_b             = 1.00
eta                 = 0.50
#restricted_scans   = 5
split-merge         = no
sampling hyperparam = no

reading data from data.txt
Killed

cat /etc/os-release
NAME="CentOS Linux"
VERSION="8 (Core)"
ID="centos"
ID_LIKE="rhel fedora"
VERSION_ID="8"
PLATFORM_ID="platform:el8"
PRETTY_NAME="CentOS Linux 8 (Core)"
ANSI_COLOR="0;31"
CPE_NAME="cpe:/o:centos:centos:8"
HOME_URL="https://www.centos.org/"
BUG_REPORT_URL="https://bugs.centos.org/"

CENTOS_MANTISBT_PROJECT="CentOS-8"
CENTOS_MANTISBT_PROJECT_VERSION="8"
REDHAT_SUPPORT_PRODUCT="centos"
REDHAT_SUPPORT_PRODUCT_VERSION="8"

uname -a
Linux simon28li 4.18.0-383.el8.aarch64 #1 SMP Wed Apr 20 15:39:57 UTC 2022 aarch64 aarch64 aarch64 GNU/Linux

I've made attempts to run hdp and hdp-faster on Ubuntu 16.04 and Windows 10. On Windows, hdp raises a segmentation fault and hdp-faster triggers std::bad_alloc. On Ubuntu, training sessions are killed after a few minutes.

I encountered that with some datasets, the final results of the training phase aren't stored under --directory if I use a random_seed of 13712 while hyper-sampling the concentration parameters as well. Only the file state.log would be produced, but not any other output files.

To reproduce the problem:

1. Download this training corpus from PAN @ CLEF 2017 competition
2. Run the regular hdp (not the fast variant) on the LDA-C corpus of the fifth training problem set, like:

hdp.exe --data ..\pan17_train\problem005\ldac_corpus.dat --algorithm train --directory ..\output --sample_hyper yes --save_lag -1 --random_seed 13712

(I used gensim to generate the LDA-C corpora)

The program will run smoothly and no error would be raised. However, the output directory would contain only the state.log file and the interim outputs, where we expect also mode.bin, mode-topics.dat and mode-word-assignments.dat. As far as I can tell, the combination of --sample_hyper yes and --random_seed 13712 is causing this fault to occur on selected datasets.