Skip to main content

Tools for building streamcorpus objects, such as those used in TREC.

Project description


kba.pipeline is a document processing pipeline that assembles
streamcorpus objects from raw data sets for us in TREC KBA.

1006305073 all-stream-ids.suc.txt
884018982 all-stream-ids.doc_ids.suc.txt


The kba.pipeline python module contains tools for processing
streamcorpus.StreamItems stored in Chunks. It includes transform
functions for getting clean_html, clean_visible text, creating labels
from hyperlinks to particular sites (e.g. Wikipedia), and taggers like
LingPipe and Stanford CoreNLP, that make Tokens and Sentences.

To create a python2.7 virtualenv, do this:

tar xzf Python-2.7.3.tgz
cd Python-2.7
./configure --prefix /data/trec-kba/installs/py27
make install
cd ..
tar xzf virtualenv-1.8.4.tar.gz
cd virtualenv-1.8.4/
/data/trec-kba/installs/py27/bin/python install
cd ..
/data/trec-kba/installs/py27/bin/virtualenv --distribute -p /data/trec-kba/installs/py27/bin/python py27env


Easiest to put this entire repo at a path like


which is hardcoded into these three files:


Then, you need these two other directories:

/data/trec-kba/keys ---- from the trec-kba-secret-keys.tar.gz that is in the Dropbox
/data/trec-kba/third/lingpipe-4.1.0 --- also in the dropbox

As a test run this:

## first go inside the virtualenv
source /data/trec-kba/installs/py27env/bin/activate

## install all the python libraries
make install

## run a simple test
make john-smith-simple

and if that works, then try

make john-smith

To try doing the real pull/push from AWS, you can put the input paths here:
zcat spinn3r-transform-input-paths.txt.gz | split -l 150 -a 4
b=0; for a in `ls ?????`; do mv $a input.$b; let b=$b+1; done;

and then locally as a test:

cat /data/trec-kba/installs/trec-kba-data/spinn3r-transform/input.0 | python -m configs/spinn3r-transform.yaml

and then, after seeing that work edit the submit script to have as
many jobs as their are input files:

condor_submit scripts/spinn3r-transform.submit

There is one key problem with this, which we discussed on the phone:
when the job dies, it starts over on the input list. Let's discuss
using the zookeeper "task_queue" stage.

running on task_queue: zookeeper

To use the zookeeper task queue, you must install zookeeper on a
computer that your cluster can access. Here is an example zookeeper

# The number of milliseconds of each tick
# The number of ticks that the initial
# synchronization phase can take
# The number of ticks that can pass between
# sending a request and getting an acknowledgement
# the directory where the snapshot is stored.
# the port at which the clients will connect

Note the large maxClientCnxns for running with many nodes in condor,
and also not the 10sec tickTime, which is needed to avoid frequent
session timeouts from condor slots that are working hard.

To make a job run off the zookeeper task queue, make these changes:


- task_queue: stdin
+ #task_queue: stdin
+ task_queue: zookeeper
+ zookeeper:
+ namespace: spinn3r-transform
+ zookeeper_address:


-Input = /data/trec-kba/spinn3r-transform/input.$(PROCESS)
+## disable stdin because we are using task_queue: zookeeper
+#Input = /data/trec-kba/spinn3r-transform/input.$(PROCESS)

Also update the number of jobs at the end of the .submit file.

and then do these steps on the command line:

## see the help text
python -m kba.pipeline.load configs/spinn3r-transform.yaml -h

## load the data
python -m kba.pipeline.load configs/spinn3r-transform.yaml --load spinn3r-transform-input-paths.txt

## check the counts -- might take a bit to run, so background and come back to it
python -m kba.pipeline.load configs/spinn3r-transform.yaml --counts >& counts &

## launch the jobs
condor_submit scripts/spinn3r-transform.submit

## watch the logs for the jobs
tail -f ../spinn3r-transform/{err,out}*

Periodically check the --counts on the queue and see how fast it is
going. Do we need to turn off the lingpipe stage?

Project details

Release history Release notifications

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Filename, size & hash SHA256 hash help File type Python version Upload date
streamcorpus_pipeline-0.4.2.dev1-py2.7.egg (777.7 kB) Copy SHA256 hash SHA256 Egg 2.7
streamcorpus_pipeline-0.4.2.dev1.tar.gz (445.9 kB) Copy SHA256 hash SHA256 Source None

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page