BANpipe is a logic-based scripting language designed to model complex compositions of time consuming analyses. The language (which itself is written in Logtalk) supports complex pipelines of Prolog programs, PRISM models and other types of programs through rules which specify dependencies between computations.
Pipelines for machine learning, natural language processing and biological sequence processing are often complex compositions of time consuming analyses, including calls to external resources and databases. The expected output, intermediate results and original input data are often huge files. BANpipe is designed to facilitate such computations.
BANpipe is a general pipeline programming language, but it is designed to support a special kind of annotation pipelines which we call Bayesian Annotation Networks (BANs). A Bayesian Network is a directed acyclic graph where the nodes are conditional probability distributions and the edges represent conditional dependencies. A BAN is a Bayesian Network where nodes are instead (probabilistic) annotation programs and edges are input/output dependencies between programs. Inference in BANs is performed iteratively by evaluating each program at a time and using its output annotation as input for dependent programs. This is not only similar to the way forward analysis takes place in Bayesian networks, but also a nice fit to the pipeline paradigm.
BANpipe is implemented in Logtalk and is portable across multiple Prolog systems.
It has been tested on OSX using B-Prolog and SWI-Prolog and the latest stable Logtalk. Windows support seems to be flaky still.
To run BANpipe, you need Logtalk plus at least one Prolog system installed.
BANpipe has been developed within the LoSt research project. BANpipe is based on what was once called "The LoSt framework", which is available on github. In addition to being the predecessor of BANpipe, the Lost framework contains a lot of libraries and modules specific to biological sequences and is developed exclusively for PRISM as host language.
If you have questions, please don't hesitate to contact me - Christian Theil Have
Copyright (C) 2012 Christian Theil Have
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
To install BANpipe you first need to download Logtalk, available from logtalk.org.
Next, you need to download the source code for BANpipe. You can download the latest version as a tgz or a zip file here.
Alternative you can clone the git repository, i.e.,
git clone https://github.com/cth/banpipe.git
Place the source code in any directory that you see fit, e.g., /home/myaccount/banpipe.
After downloading banpipe, you will need to setup BANpipe as a Logtalk library:
You can edit the file settings.lgt in the BANpipe directory to reflect your installation path
and then move the settings.lgt to $LOGTALKUSER.
The environment variable BANPIPE_PATH must be set to the banpipe directory:
export BANPIPE_PATH=/home/myacccount/banpipe
You also need to add any directory which contains BANpipe modules to the
$BANPIPE_MODULE_PATH variable. For instance, to add the example modules,
export BANPIPE_MODULE_PATH=/home/myacccount/banpipe/examples
The $BANPIPE_MODULE_PATH environment variable works like the normal PATH variable and you can add multiple directories separated by ':'
Now, you should be able to start Logtalk (e.g., using swilgt) and call,
{banpipe(loader)}.
which will load the BANpipe library.
You can now load BANpipe scripts, e.g. the helloworld.pl examples, by calling the goal,
banpipe::load('/home/myaccount/examples/helloworld.pl').
and subsequently run the script using
banpipe::run(helloworld(english),Outfile).
Outfile will be unified to name of the output file.
Once installation of BANpipe is completed, you can load BANpipe using the command
{banpipe(loader)}.
A BANpipe script is loaded using the command
banpipe::load(Filename).
where Filename is a relative or absolute file name in the filesystem.
You need to supply the full filename including possible extensions, e.g. .pl
To run a goal in a BANpipe script the command,
banpipe::run(Goal).
is used. This will run the scripts sequentially in a bottom-up, left-to-right fashion, i.e., dependencies are computed first left-most dependencies in a task are computed before right-most dependencies.
To get the filename corresponding to the goal, call instead,
banpipe::run(Goal,File).
This will unify File to the generated file corresponding to Goal.
When banpipe::run(Goal) is called, scripts are run in a sequential fashion. It is possible also to run scripts with automatic parallelization, so that all independent tasks will be run simultaneously. This is achieved by calling,
banpipe::prun(Goal).
Runnings scripts with parallization is only available when running BANpipe with Logtalk on a Prolog system which has support for threads, e.g., SWI-Prolog.
Scripts are not type checked when calling banpipe::runor banpipe::prun. Is perfectly possible to
run a script that is not well-typed. However, type checking a script in advance may help to avoid
errors. To typecheck a script use the goal,
banpipe::typecheck(Goal).
or similarly
banpipe::typecheck(Goal,Type).
to get the Type corresponding to Goal. banpipe::typecheck/{1,2} fails with a report of the type error,
if the script is not well-typed.
A call graph shows the interdependencies between tasks and may be used as a way to debug scripts. Even though it is a graph, it is displayed as a tree where nodes which are shared in the graph are shown as multiple branches of the tree. A call graph is obtained by calling,
banpipe::callgraph(Goal).
For instance, for the hello world example, the following call graph is produced:
?- banpipe::callgraph(helloworld(english)).
+1 ready helloworld::helloworld([hello(english),world(english)],[])
|--+3 ready helloworld::world([],[language(english)])
|--+2 ready helloworld::hello([],[language(english)])
Scripts can be run through an interactive trace. This is enabled by typing banpipe::trace.
To disable tracing, type banpipe::notrace.
There is really nothing special about change propagation. Applying change propagation is
achieved by re-running goals. However, changes to modules are not detected automatically.
Instead, each task declaration in a module may include a version option.
When the the version option is changed, the task gets a new signature and it
will be recomputed (as will depending tasks implied by goal) the next time a goal involving the task is called.
BANpipe modules are not to be confused with Prolog modules. A BANpipe module is simply a collection of source files residing in a particular directory. The name of the module is the same as the directory, e.g.,
if a module is placed in /home/user/modules/foo then the name of the module is foo.
In the module directory, BANpipe expects to find a file called interface.pl. This is a Prolog file provided by the the module creator which contains declarations and predicates for invoking the module.
The interface file contains declarations of task predicates provided by the module. An task declaration has the form,
:- task(<taskname>(<input-file-types>,<options>,<output-file-types>)).
The declaration specifies that the module provides a task predicate with the name <taskname>.
Such predicates always takes three arguments, a list of input files, a list of options and a list of output files. <input-file-types> is a Prolog list of terms which specifies the number and types of input files that the task predicate expects. The types are used defined and any Prolog term is a valid type. Similarly, <output-file-types> is a list of terms. The terms may contain variables, which may be shared between input types and output types. In this way, an output file type, may depend on an input file type. <options> is a list of terms of arity one. The functor serves as key and the argument serves as value. Options may also contain variables.
Suppose that the interface file /home/user/modules/foo/interface.pl, i.e. of the module foo includes a task declaration:
:- task(bar([_],[],[_])
The name of task is bar and it takes one input file (of unspecified type _), it takes no options and it writes one output file (of unspecified type _). Furthermore, the interface file contains a task predicate,
bar([InFile],_,[OutFile]) :- ...
This predicate will be called (see below) with ground filenames. BANpipe guarantees that InFile is available when the predicate is called and expects are OutFile is available (has been written) when the predicate terminates.
The task predicate gets called when a goal of a dependency rule in the BANpipe script refers to the task predicate in its body, e.g.,
foobar <- foo::bar(['file:://tmp/infile']).
When the goal banpipe::run(foobar,File) is called then a new Prolog process will be started and the /home/user/modules/foo/interface.pl will be consulted (from the /home/user/modules/foo/ directory). This prolog process will then call the bar task predicate with Infile='/tmp/infile' and OutFile being a unique filename generated by the system. When banpipe::run(foobar,File) finishes, File is unified to this unique file.
Interface files may also contain an optional invoke_with/1 declaration, i.e., the declaration
:- invoke_with(swipl)
specifies that the interface.pl should be loaded with SWI-prolog. The current list of valid arguments for invoke_with/1 are
- prism (PRISM - this is the default)
- bp (B-Prolog)
- swipl (SWI-Prolog)
- gprolog (GNU-Prolog)
Documented examples are available in the examples/ folder of BANpipe (on github).