def load_code_from_file(self,filename):
# Clear out the old references
self.canvas = self.controller = self.exec_model = None
self.exec_model = ExecutionModel.from_file(filename)
self.controller = BlockGraphController(execution_model = self.exec_model)
self.canvas = BlockCanvas(graph_controller=self.controller)
self.controller.canvas = self.canvas
def __init__(self, code=None, shared_context=None, *args, **kwargs):
super(Experiment, self).__init__(*args, **kwargs)
if code is None:
self.exec_model = ExecutionModel()
else:
self.exec_model = ExecutionModel.from_code(code)
self.controller = BlockGraphController(execution_model = self.exec_model)
self.canvas = BlockCanvas(graph_controller=self.controller)
self.controller.canvas = self.canvas
self._shared_context = shared_context
self._local_context = DataContext(name = self._LOCAL_NAME_TEMPLATE())
self._update_exec_context()
def load_from_config(self, config, dirname, project=None):
""" Loads the experiment. The existing state of the experiment is
completely modified.
Parameters
----------
config: a dict-like object
The keys should correspond to the configspec for experiments as
defined in project_config_spec.
dirname: a string
the absolute path to the subdirectory of the project that
holds the experiment's saved data
project: Project instance
If provided, the project is used to hook up references to
the shared context.
"""
join = os.path.join
self.name = config.get("name", "")
if "code_file" in config:
# Clear out the old references
self.canvas = self.controller = self.exec_model = None
self.exec_model = ExecutionModel.from_file(join(dirname, config["code_file"]))
self.controller = BlockGraphController(execution_model = self.exec_model)
self.canvas = BlockCanvas(graph_controller=self.controller)
self.controller.canvas = self.canvas
if "layout_file" in config:
self.canvas.load_layout(join(dirname, config["layout_file"]))
if "local_context" in config:
self._local_context = DataContext.load(join(dirname, config["local_context"]))
shared_context = None
if project is not None:
name = config.get("shared_context")
if name != "":
shared_context = project.find_context(name)
self._shared_context = shared_context
self._update_exec_context()
class Experiment(HasTraits):
"""
An Experiment represents a particular workflow and its associated data.
It is associated with a particular context, but any data it produces
is placed into a shadowing "local" context.
It encapsulates an execution model, its associated canvas, and the
local context.
"""
# The name of the experiment.
name = String()
# The Execution Model object
exec_model = Instance(ExecutionModel)
# The Block Canvas
canvas = Instance(BlockCanvas)
# The controller between the canvas and execution model
controller = Instance(BlockGraphController)
# The execution context
context = Instance(ExecutingContext)
# A reference to a context that might be shared with other experiments.
# This is usually a context from the project's list of contexts. If None,
# then all of the data in the experiment is kept "locally".
#
# Note: For most purposes, you should use the **context** attribute
# and not this one, unless you really know what you are doing.
shared_context = Property(Instance(IListenableContext))
#---------------------------------------------------------------------
# Private Traits
#---------------------------------------------------------------------
# A shadow trait for **shared_context** property
_shared_context = Instance(IListenableContext, adapt='yes',
rich_compare=False)
# The context in which all of our execution model's generated values are
# stored. This context is exposed as part of self.context. This remains
# constant even as the _shared_context gets changed.
_local_context = Instance(IListenableContext, rich_compare=False)
# Class-level generator for the name of the local context; takes
# **self** as an argument.
_LOCAL_NAME_TEMPLATE = lambda x: "%s_local_%d" % (x.name, id(x))
#---------------------------------------------------------------------
# Public methods
#---------------------------------------------------------------------
def __init__(self, code=None, shared_context=None, *args, **kwargs):
super(Experiment, self).__init__(*args, **kwargs)
if code is None:
self.exec_model = ExecutionModel()
else:
self.exec_model = ExecutionModel.from_code(code)
self.controller = BlockGraphController(execution_model = self.exec_model)
self.canvas = BlockCanvas(graph_controller=self.controller)
self.controller.canvas = self.canvas
self._shared_context = shared_context
self._local_context = DataContext(name = self._LOCAL_NAME_TEMPLATE())
self._update_exec_context()
#---------------------------------------------------------------------
# Persistence
#---------------------------------------------------------------------
def load_code_from_file(self,filename):
# Clear out the old references
self.canvas = self.controller = self.exec_model = None
self.exec_model = ExecutionModel.from_file(filename)
self.controller = BlockGraphController(execution_model = self.exec_model)
self.canvas = BlockCanvas(graph_controller=self.controller)
self.controller.canvas = self.canvas
def load_from_config(self, config, dirname, project=None):
""" Loads the experiment. The existing state of the experiment is
completely modified.
Parameters
----------
config: a dict-like object
The keys should correspond to the configspec for experiments as
defined in project_config_spec.
dirname: a string
the absolute path to the subdirectory of the project that
holds the experiment's saved data
project: Project instance
If provided, the project is used to hook up references to
the shared context.
"""
join = os.path.join
self.name = config.get("name", "")
if "code_file" in config:
# Clear out the old references
self.canvas = self.controller = self.exec_model = None
self.exec_model = ExecutionModel.from_file(join(dirname, config["code_file"]))
self.controller = BlockGraphController(execution_model = self.exec_model)
self.canvas = BlockCanvas(graph_controller=self.controller)
self.controller.canvas = self.canvas
if "layout_file" in config:
self.canvas.load_layout(join(dirname, config["layout_file"]))
if "local_context" in config:
self._local_context = DataContext.load(join(dirname, config["local_context"]))
shared_context = None
if project is not None:
name = config.get("shared_context")
if name != "":
shared_context = project.find_context(name)
self._shared_context = shared_context
self._update_exec_context()
def save(self, basename, dirname):
""" Saves the experiment into a directory.
Parameters
----------
basename: string
the name of the project directory
dirname: string
the name of the experiment's subdirectory
Returns
-------
A dict representing the saved state for this object. All path
references in the dict will be relative to the given root directory.
"""
# Make the directory using the absolute path.
fullpath = join(basename, dirname)
if not os.path.isdir(fullpath):
os.mkdir(fullpath)
config = {}
config["name"] = self.name
config["save_dir"] = dirname
# Save out the canvas
config["layout_file"] = "layout.dat"
self.canvas.save_layout(join(fullpath, "layout.dat"))
# Save the execution model's code
config["code_file"] = "code.txt"
f = file(join(fullpath, "code.txt"), "w")
try:
f.write(self.exec_model.code)
finally:
f.close()
# Clean context - too problematic
self.exec_model._clean_old_results_from_context(self.context.subcontext)
# Save the local context
config["local_context"] = "local_context.pickle"
self._local_context.save(join(fullpath, "local_context.pickle"))
# Store the name of the shared context
if self._shared_context is not None:
config["shared_context"] = self._shared_context.name
return config
def save_script(self,filename):
""" Save the execution model code as script. """
f = file(filename, "w")
try:
f.write(self.exec_model.code)
finally:
f.close()
def export_as_function(self,filename,func_name, mode="w"):
""" Export this experiment as a function for batch application.
It puts together the model (i.e. code) and the parameters
(i.e. context) to create a self-contained function to be used
in batch application.
"""
imports_and_locals = self.exec_model.imports_and_locals
sorted_statements = self.exec_model.sorted_statements
context = self.context
# Clean context
self.exec_model._clean_old_results_from_context(context.subcontext)
export_as_function(filename, func_name, \
imports_and_locals, sorted_statements, context, \
reserved_inputs=[], \
reserved_outputs=[], \
mode=mode)
def export_as_script(self,filename, mode="w"):
""" Export this experiment as a script for batch application.
It puts together the model (i.e. code) and the parameters
(i.e. context) to create a self-contained script to be used
in batch application.
"""
imports_and_locals = self.exec_model.imports_and_locals
sorted_statements = self.exec_model.sorted_statements
context = self.context
# Clean context
self.exec_model._clean_old_results_from_context(context.subcontext)
export_as_script(filename, \
imports_and_locals, sorted_statements, context, \
reserved_inputs=[], \
reserved_outputs=[], \
mode=mode)
#---------------------------------------------------------------------
# Trait Event Handlers
#---------------------------------------------------------------------
@on_trait_change('exec_model')
def _exec_model_changed(self, name, old, new):
""" Propagate the change to the objects under this one.
"""
if self.controller is not None:
self.controller.execution_model = new
if self.context is not None:
self.context.executable = new
#---------------------------------------------------------------------
# Property getters/setters
#---------------------------------------------------------------------
def _get_shared_context(self):
return self._shared_context
def _set_shared_context(self, newcontext):
subcontexts = self.context.subcontext.subcontexts
if self._shared_context is not None:
assert(self._shared_context in subcontexts)
if newcontext is None:
subcontexts.remove(self._shared_context)
self._shared_context = None
else:
# Swap out the new context for the old, in-place
ndx = subcontexts.index(self._shared_context)
subcontexts[ndx] = newcontext
self._shared_context = newcontext
elif newcontext is not None:
self._shared_context = newcontext
subcontexts.append(newcontext)
return
def _update_exec_context(self):
mc = MultiContext(
# Put the function filter in front so we don't dirty up the data
# context with function objects.
FunctionFilterContext(name='functions'),
self._local_context,
name='multi',
)
if self._shared_context is not None:
mc.subcontexts.append(self._shared_context)
self.context = ExecutingContext(
executable=self.exec_model,
subcontext=mc,
name='exec',
)
