Source code for schrodinger.stepper.sideinputs
"""
Utility steps for creating side inputs in stepper workflows.
For example, in a workflow with steps A, B, C, and D, a ForkStep and JoinStep
can be set up so that all outputs from A are passed along to D. This allows
outputs from A to get to D even if B or C would normally filter those inputs.
Example::
class MyWorkflow(stepper.Chain):
def buildChain(self):
a = A()
self.addStep(a)
fork = ForkStep(step=a)
self.addStep(fork)
self.addStep(B())
self.addStep(C())
self.addStep(JoinStep(fork=fork))
self.addStep(D())
"""
import os
import uuid
from schrodinger.models import parameters
from schrodinger import stepper
SCHRODINGER_RUN = os.path.join(os.environ['SCHRODINGER'], 'run')
[docs]class ForkStep(stepper.UnbatchedReduceStep):
"""
A step to save some inputs to be reprocessed again. See the module
docstring for more info and an example.
"""
[docs] def __init__(self, step):
self.Input = self.Output = step.Output
self.InputSerializer = self.OutputSerializer = step.OutputSerializer
self._pipe_fname = f'.{str(uuid.uuid4())}.forkfile'
super().__init__()
[docs] def reduceFunction(self, inps):
serializer = self.getOutputSerializer()
with open(self._pipe_fname, 'w') as outfile:
for inp in inps:
outfile.write(f"{serializer.toString(inp)}\n")
yield inp
[docs] def getPipeFilename(self):
return self._pipe_fname
[docs] def report(self, prefix=''):
stepper.logger.info(f'{prefix} - {self.getStepId()}')
class JoinStep(stepper.UnbatchedReduceStep):
"""
A step to read some inputs saved by a preceding ForkStep. See the
module docstring for more info and an example.
"""
def __init__(self, fork):
self._fork = fork
self.InputSerializer = self.OutputSerializer = fork.OutputSerializer
self._in_fname = fork.getPipeFilename()
super().__init__()
@property
def Input(self):
return self._fork.Input
@property
def Output(self):
return self._fork.Output
[docs] def reduceFunction(self, inps):
yield from inps
yield from self.getOutputSerializer().deserialize(self._in_fname)
[docs] def report(self, prefix=''):
stepper.logger.info(
f'{prefix} - {self.getStepId()} <- {self._fork.getStepId()}')
class JoinFromFileStep(stepper.UnbatchedReduceStep):
"""
A step for injecting inputs read from a file into a chain. To use, add into
your chain and set the step's `join_file` setting to the path of your
datafile.
"""
[docs] class Settings(parameters.CompoundParam):
join_file: stepper.StepperFile = None
def __init__(self, Input=None, InputSerializer=None, **kwargs):
if Input is None and InputSerializer is None:
raise TypeError("Must set either Input or InputSerializer at "
"step initialization time.")
elif Input is not None and InputSerializer is not None:
raise TypeError("Can't set both Input _and_ InputSerializer")
if InputSerializer:
Input = InputSerializer.DataType
self.InputSerializer = self.OutputSerializer = InputSerializer
self.Input = self.Output = Input
super().__init__(**kwargs)
[docs] def reduceFunction(self, inps):
serializer = self._getInputSerializer()
yield from serializer.deserialize(self.settings.join_file)
yield from inps
#==============================================================================
# PUBSUB FUNCTIONALITY
#
# The below code adds functionality to the Join* so extra inputs will just
# be added to the input topics. This optimizes the Join* steps so they don't
# unnecessarily read topics just to append a few extra inputs.
#
# NOTE: This section is designed to be 'transparent', meaning if a chain
# is not using pubsub, all JoinSteps will behave normally. Additionally,
# if this section of code is removed, the base functionality of the Join*
# will still work.
#==============================================================================
[docs]class JoinStep(stepper.PubsubEnabledStepMixin, JoinStep):
[docs] def outputs(self):
if self.usingPubsub():
inp_topic = self.getInputTopic()
self.setOutputTopic(inp_topic)
extra_outputs = self.getOutputSerializer().deserialize(
self._in_fname)
self._uploadToTopic(extra_outputs, self.getOutputSerializer(),
inp_topic)
return self._deserializeFromOutputTopic()
else:
return super().outputs()
[docs] def usingPubsub(self):
return bool(self.getInputTopic())
[docs]class JoinFromFileStep(stepper.PubsubEnabledStepMixin, JoinFromFileStep):
[docs] def outputs(self):
if self.usingPubsub():
inp_topic = self.getInputTopic()
self.setOutputTopic(inp_topic)
extra_outputs = self.getOutputSerializer().deserialize(
self.settings.join_file)
self._uploadToTopic(extra_outputs, self.getOutputSerializer(),
inp_topic)
return self._deserializeFromOutputTopic()
else:
return super().outputs()
[docs] def usingPubsub(self):
return bool(self.getInputTopic())