In this page, we are going to demonstrate how to use Python APIs.
Since OACIS is implemented in Ruby, we developed a library to call Ruby methods from Python. Using this library, almost all methods in Ruby is directly translated into Python.
In the following, we assume that a Simulator “my_simulator” is registered on OACIS, which has three parameters “p1”, “p2”, and “p3”. We assume that you know basics of Python programming language.
To use OACIS, Python3 is necessary. (As of OACIS v3.2.1, Python 3.7.0 or later is not supported since fibers library does not work with it.) It also requires the following libraries.
$ pip install mprpc fibers
New in v2.13.0“fibers” library is required since v2.13.0.
New in v3.6.2“mprpc” library is required since v3.6.2.
For OACIS v3.6.1 or prior, the following packages are needed.
$ pip install msgpack-python==0.4.8 msgpack-rpc-python==0.4.1 fibers
To use OACIS APIs, we can either write a Python script or execute in an interactive environment. We recommend the interactive environment for testing and debugging. Once you fix your procedure, we recommend to use a script.
Run bin/oacis_python
command in the repository of OACIS and you’ll find an interactive shell.
Import oacis
module to access OACIS APIs.
$ ~/oacis/bin/oacis_python
Python 3.5.2 |Continuum Analytics, Inc.| (default, Jul 2 2016, 17:52:12)
[GCC 4.2.1 Compatible Apple LLVM 4.2 (clang-425.0.28)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import oacis
or you can use ipython or Jupyter if you prefer. In that case, set PYTHONPATH
environment variable such that you can import oacis.py
.
$ export PYTHONPATH="/path/to/oacis:$PYTHONPATH"
$ ipython
You can run a script by giving the path to your script to oacis_python
command.
$ echo 'import oacis; print( oacis.Simulator.first().name() )' > test.py # preparing test.rb
$ ~/oacis/bin/oacis_python test.py
"my_simulator"
To operate OACIS, we are going to use the methods of the following classes. Major methods of these classes are shown in the following.
These APIs are available for OACIS 2.9.0 or later.
sim = oacis.Simulator.find("...ID...")
sim = oacis.Simulator.find_by_name("my_simulator")
# If you are using OACIS v2.9.0 or prior, use the following.
# sim = oacis.Simulator.where(name="my_simulator").first()
sim.name() #=> "my_simulator"
for d in sim.parameter_definitions():
print( d.inspect() ) # If you call `inspect` method against an instance of RubyObject, you'll have a more human-friendly output
#=>
[#<ParameterDefinition _id: 522d751f899e533149000003, key: "p1", type: "Integer", default: 1, description: "first parameter">,
#<ParameterDefinition _id: 522d751f899e533149000004, key: "p2", type: "Integer", default: 1, description: "second parameter">,
#<ParameterDefinition _id: 522d751f899e533149000005, key: "p3", type: "Float", default: 0.0, description: "third parameter">]
sim.default_parameters()
#=> {'p1': 1.0, 'p2': 2.0, 'p3': 3.0}
ps = oacis.ParameterSet.find("...ID...")
Simulator#find_parameter_set
returns None if a matching ParameterSet is not found.sim = oacis.Simulator.where(name: "my_simulator").first()
ps = sim.find_parameter_set( {"p1":1.0, "p2":2.0, "p3":3.0} )
#=> RubyObject( ParameterSet, 70231167310880 )
sim = Simulator.where(name: "my_simulator").first()
[ ps.id().to_s() for ps in sim.parameter_sets().where({"v.p1":1.0,"v.p2":2.0}) ]
#=> ['5805c089b93f969922b863a9']
The above code is searching for a ParameterSet under a given Simulator. The values of parameters are stored in the field “v”. Queries on the sub-elements of “v” is used for searching.
After matching ParameterSets are found, an iteration over them is conducted by for
syntax.
v()
returns the values of parameters as a Hash.dir()
returns the path to the directory where the data are stored.average_result
returns the average (and the number of runs) of the specified resultps.v() #=> {'p1': 1.0, 'p2': 2.0, 'p3': 3.0}
ps.dir().to_s() # => '/path/oacis/public/Result_development/5805c082b93f969922b863a1/5805c089b93f969922b863a9'
ps.average_result("result1") # => [0.25, 5] Average of "result1" is 0.25, which is averaged over 5 runs
sim.find_or_create_parameter_set( {"p1":1.0, "p2":2.0, "p3": 3.0} ) #=> RubyObject( ParameterSet, 70231135107120 )
If a ParameterSet of {"p1"=>1, "p2"=>2.0}
already exists, it return that ParameterSet.
If such ParameterSet does not exist, a new ParameterSet is created.
ps.discard()
Call discard
method to remove a ParameterSet. This method automatically removes its sub-elements (i.e., runs and analysis) as well even if remote jobs are running on a remote host.
run = oacis.Run.find("...ID...")
or
run = parameter_set.runs()[0] # getting the first Run of a PS
for run in ps.runs().where( {'status': 'finished'} ):
print( run.id() )
You can get information of a Run as the following.
run = ps.runs().where( {'status': 'finished'} )[0]
run.status() #=> either 'created', 'submitted', 'running', 'finished', or 'failed'
run.submitted_to() #=> The host to which a job is submitted to.
run.host_parameters()#=> {"ppn": "1", "walltime": "1:00:00"}
run.mpi_procs() #=> 1
run.omp_threads() #=> 1
run.priority() #=> 1
run.result() #=> {"result1": -0.016298, "result2": 0.0264882}
host = oacis.Host.where(name='localhost').first()
host_param = {'ppn':"4",'walltime':"1:00:00"}
# To get default host parameters, the following method is available.
# host.default_host_parameters
runs = ps.find_or_create_runs_upto( 10, submitted_to=host, host_param=host_param, mpi_procs=4, priority=0 )
# The keyword arguments "host_param", "mpi_procs", "omp_threads", "priority" are optional.
# For priority, you can choose from [0,1,2], where "0" has the highest priority.
ParameterSet#find_or_create_runs_upto
method create runs until the number of Runs becomes the specified number.
If there already exists enough number of Runs, runs are not newly created.
The returned value is an array of Runs.
You can also specify “HostGroup” as follows.
host_group = oacis.HostGroup.where(name="my_host_group").first
runs = ps.find_or_create_runs_upto( 10, host_group=host_group)
run.discard()
host = oacis.Host.find_by_name("localhost")
# If you are using OACIS v2.9.0 or prior, use the following.
# host = oacis.Host.where(name="localhost").first()
host.status() #=> either 'enabled' or 'disabled' is returned
host.user() #=> user name
host.port() #=> 22
host.ssh_key() #=> '~/.ssh/id_rsa'
host.host_parameter_definitions().inspect()
=> [#<HostParameterDefinition _id: 57babbb46b696d52bf240000, key: "ppn", default: "1", format: "^[1-9]\\d*$">,
#<HostParameterDefinition _id: 57babbb46b696d52bf250000, key: "walltime", default: "1:00:00", format: "^\\d+:\\d{2}:\\d{2}$">]
For other available fields, refer to app/models/host.rb.
azr = oacis.Analyzer.find("...ID...")
azr = sim.find_analyzer_by_name("my_analyzer")
# If you are using OACIS v2.9.0 or prior, use the following.
# azr = sim.analyzers().where(name:"my_analyzer").first()
azr.support_mpi() #=> true/false
azr.support_omp() #=> true/false
azr.command() #=> execution command
anl = oacis.Analysis.find("...ID...")
To find an Analysis on a ParameterSet, you search for it as parameter_set.analyses().where()
.
For an Analysis on a Run, use run.analyses().where()
idiom.
sim = oacis.Simulator.find("...ID...")
azr = sim.analyzers().where( {"name":"my_analyzer"} ).first
for anl in ps.analyses.where( {"analyzer": azr, "status": "finished"} ):
print( anl.id() )
Almost same set of APIs to those of Runs are available.
anl.status() #=> one of [:created,:submitted,:running,:failed,:finished] is returned.
anl.submitted_to() #=> #<Host _id: 53a3f583b93f964b7f0000fc, ...>
anl.host_parameters() #=> {"ppn"=>"1","walltime"=>"1:00:00"}
anl.result() #=> {"result1"=>-0.016298, "result2"=>0.0264882}
When creating an Analysis, we need to specify its Analyzer, host, and host_parameters.
host_param = {"ppn": "1", "walltime": "1:00:00"}
ps.analyses().create( analyzer=azr, submitted_to=host, host_parameters=host_param )
anl.discard()