"""
Macro and Model classes are the main classes for loading and interacting
with a PySD model. Model class allows loading and running a PySD model.
Several methods and propierties are inherited from Macro class, which
allows integrating a model or a Macro expression (set of functions in
a separate file).
"""
import time
import warnings
import inspect
import pickle
from pathlib import Path
from typing import Union
import numpy as np
import xarray as xr
import pandas as pd
from pysd._version import __version__
from . import utils
from .statefuls import DynamicStateful, Stateful
from .external import External, Excels, ExtLookup, ExtData
from .cache import Cache, constant_cache
from .data import TabData
from .lookups import HardcodedLookups
from .components import Components, Time
from .output import ModelOutput
[docs]class Macro(DynamicStateful):
"""
The Macro class implements a stateful representation of the system,
and contains the majority of methods for accessing and modifying
components.
When the instance in question also serves as the root model object
(as opposed to a macro or submodel within another model) it will have
added methods to facilitate execution.
The Macro object will be created with components drawn from a
translated Python model file.
Parameters
----------
py_model_file: str or pathlib.Path
Filename of a model or macro which has already been converted
into a Python format.
params: dict or None (optional)
Dictionary of the macro parameters. Default is None.
return_func: str or None (optional)
The name of the function to return from the macro. Default is None.
time: components.Time or None (optional)
Time object for integration. If None a new time object will
be generated (for models), if passed the time object will be
used (for macros). Default is None.
time_initialization: callable or None
Time to set at the begginning of the Macro. Default is None.
data_files: dict or list or str or None
The dictionary with keys the name of file and variables to
load the data from. Or the list of names or name of the file
to search the data in. Only works for TabData type object
and it is neccessary to provide it. Default is None.
py_name: str or None
The name of the Macro object. Default is None.
"""
def __init__(self, py_model_file, params=None, return_func=None,
time=None, time_initialization=None, data_files=None,
py_name=None):
super().__init__()
self.time = time
self.time_initialization = time_initialization
# Initialize the cache object
self.cache = Cache()
# Python name of the object (for Macros)
self.py_name = py_name
# Booleans to avoid loading again external data or lookups
self.external_loaded = False
self.lookups_loaded = False
# Functions with constant cache
self._constant_funcs = set()
# Load model/macro from file and save in components
self.components = Components(str(py_model_file), self.set_components)
if __version__.split(".")[0]\
!= self.get_pysd_compiler_version().split(".")[0]:
raise ImportError(
"\n\nNot able to import the model. "
+ "The model was translated with a "
+ "not compatible version of PySD:"
+ "\n\tPySD " + self.get_pysd_compiler_version()
+ "\n\nThe current version of PySd is:"
+ "\n\tPySD " + __version__ + "\n\n"
+ "Please translate again the model with the function"
+ " read_vensim or read_xmile.")
# Assing some protected attributes for easier access
self._namespace = self.components._components.component.namespace
self._dependencies =\
self.components._components.component.dependencies.copy()
self._subscript_dict = getattr(
self.components._components, "_subscript_dict", {})
self._modules = getattr(
self.components._components, "_modules", {})
self._doc = self._build_doc()
if params is not None:
# add params to namespace
self._namespace.update(self.components._components._params)
# create new components with the params
self.set_components(params, new=True)
# update dependencies
for param in params:
self._dependencies[
self._namespace[param]] = {"time"}
# Get the collections of stateful elements and external elements
self._stateful_elements = {
name: getattr(self.components, name)
for name in dir(self.components)
if isinstance(getattr(self.components, name), Stateful)
}
self._dynamicstateful_elements = [
getattr(self.components, name) for name in dir(self.components)
if isinstance(getattr(self.components, name), DynamicStateful)
]
self._external_elements = [
getattr(self.components, name) for name in dir(self.components)
if isinstance(getattr(self.components, name), External)
]
self._macro_elements = [
getattr(self.components, name) for name in dir(self.components)
if isinstance(getattr(self.components, name), Macro)
]
self._data_elements = [
getattr(self.components, name) for name in dir(self.components)
if isinstance(getattr(self.components, name), TabData)
]
self._lookup_elements = [
getattr(self.components, name) for name in dir(self.components)
if isinstance(getattr(self.components, name), HardcodedLookups)
]
# Load data files
if data_files:
self._get_data(data_files)
# Assign the cache type to each variable
self._assign_cache_type()
# Get the initialization order of Stateful elements
self._get_initialize_order()
if return_func is not None:
# Assign the return value of Macros
self.return_func = getattr(self.components, return_func)
else:
self.return_func = lambda: 0
self.py_model_file = str(py_model_file)
def __call__(self):
return self.return_func()
@property
def doc(self) -> pd.DataFrame:
"""
The documentation of the model.
"""
return self._doc.copy()
@property
def namespace(self) -> dict:
"""
The namespace dictionary of the model.
"""
return self._namespace.copy()
@property
def dependencies(self) -> dict:
"""
The dependencies dictionary of the model.
"""
return self._dependencies.copy()
@property
def subscripts(self) -> dict:
"""
The subscripts dictionary of the model.
"""
return self._subscript_dict.copy()
@property
def modules(self) -> Union[dict, None]:
"""
The dictionary of modules of the model. If the model is not
split by modules it returns None.
"""
return self._modules.copy() or None
[docs] def clean_caches(self):
"""
Clean the cahce of the object and the macros objects that it
contains
"""
self.cache.clean()
# if nested macros
[macro.clean_caches() for macro in self._macro_elements]
def _get_data(self, data_files):
if isinstance(data_files, dict):
for data_file, vars in data_files.items():
for var in vars:
found = False
for element in self._data_elements:
if var in [element.py_name, element.real_name]:
element.load_data(data_file)
found = True
break
if not found:
raise ValueError(
f"'{var}' not found as model data variable")
else:
for element in self._data_elements:
element.load_data(data_files)
def _get_initialize_order(self):
"""
Get the initialization order of the stateful elements
and their the full dependencies.
"""
# get the full set of dependencies to initialize an stateful object
# includying all levels
self.stateful_initial_dependencies = {
ext: set()
for ext in self._dependencies
if (ext.startswith("_") and not ext.startswith("_active_initial_"))
}
for element in self.stateful_initial_dependencies:
self._get_full_dependencies(
element, self.stateful_initial_dependencies[element],
"initial")
# get the full dependencies of stateful objects taking into account
# only other objects
current_deps = {
element: [
dep for dep in deps
if dep in self.stateful_initial_dependencies
] for element, deps in self.stateful_initial_dependencies.items()
}
# get initialization order of the stateful elements
self.initialize_order = []
delete = True
while delete:
delete = []
for element in current_deps:
if not current_deps[element]:
# if stateful element has no deps on others
# add to the queue to initialize
self.initialize_order.append(element)
delete.append(element)
for element2 in current_deps:
# remove dependency on the initialized element
if element in current_deps[element2]:
current_deps[element2].remove(element)
# delete visited elements
for element in delete:
del current_deps[element]
if current_deps:
# if current_deps is not an empty set there is a circular
# reference between stateful objects
raise ValueError(
'Circular initialization...\n'
+ 'Not able to initialize the following objects:\n\t'
+ '\n\t'.join(current_deps))
def _get_full_dependencies(self, element, dep_set, stateful_deps):
"""
Get all dependencies of an element, i.e., also get the dependencies
of the dependencies. When finding an stateful element only dependencies
for initialization are considered.
Parameters
----------
element: str
Element to get the full dependencies.
dep_set: set
Set to include the dependencies of the element.
stateful_deps: "initial" or "step"
The type of dependencies to take in the case of stateful objects.
Returns
-------
None
"""
deps = self._dependencies[element]
if element.startswith("_"):
deps = deps[stateful_deps]
for dep in deps:
if dep not in dep_set and not dep.startswith("__")\
and dep != "time":
dep_set.add(dep)
self._get_full_dependencies(dep, dep_set, stateful_deps)
def _add_constant_cache(self):
for element, cache_type in self.cache_type.items():
if cache_type == "run":
self.components._set_component(
element,
constant_cache(getattr(self.components, element))
)
self._constant_funcs.add(element)
def _remove_constant_cache(self):
for element in self._constant_funcs:
self.components._set_component(
element,
getattr(self.components, element).function)
self._constant_funcs.clear()
def _assign_cache_type(self):
"""
Assigns the cache type to all the elements from the namespace.
"""
self.cache_type = {"time": None}
for element in self._namespace.values():
if element not in self.cache_type\
and element in self._dependencies:
self._assign_cache(element)
for element, cache_type in self.cache_type.items():
if cache_type is not None:
if element not in self.cache.cached_funcs\
and self._count_calls(element) > 1:
self.components._set_component(
element,
self.cache(getattr(self.components, element)))
self.cache.cached_funcs.add(element)
def _count_calls(self, element):
n_calls = 0
for subelement in self._dependencies:
if subelement.startswith("_") and\
element in self._dependencies[subelement]["step"]:
if element in\
self._dependencies[subelement]["initial"]:
n_calls +=\
2*self._dependencies[subelement]["step"][element]
else:
n_calls +=\
self._dependencies[subelement]["step"][element]
elif (not subelement.startswith("_") and
element in self._dependencies[subelement]):
n_calls +=\
self._dependencies[subelement][element]
return n_calls
def _assign_cache(self, element):
"""
Assigns the cache type to the given element and its dependencies if
needed.
Parameters
----------
element: str
Element name.
Returns
-------
None
"""
if not self._dependencies[element]:
self.cache_type[element] = "run"
elif "__lookup__" in self._dependencies[element]:
self.cache_type[element] = None
elif self._isdynamic(self._dependencies[element]):
self.cache_type[element] = "step"
else:
self.cache_type[element] = "run"
for subelement in self._dependencies[element]:
if subelement.startswith("_initial_")\
or subelement.startswith("__"):
continue
if subelement not in self.cache_type:
self._assign_cache(subelement)
if self.cache_type[subelement] == "step":
self.cache_type[element] = "step"
break
def _isdynamic(self, dependencies):
"""
Parameters
----------
dependencies: iterable
List of dependencies.
Returns
-------
isdynamic: bool
True if 'time' or a dynamic stateful objects is in dependencies.
"""
if "time" in dependencies:
return True
for dep in dependencies:
if dep.startswith("_") and not dep.startswith("_initial_")\
and not dep.startswith("__"):
return True
return False
[docs] def get_pysd_compiler_version(self):
"""
Returns the version of pysd complier that used for generating
this model
"""
return self.components.__pysd_version__
[docs] def initialize(self):
"""
This function initializes the external objects and stateful objects
in the given order.
"""
# Initialize time
if self.time is None:
self.time = self.time_initialization()
# Reset time to the initial one
self.time.reset()
self.cache.clean()
self.components._init_outer_references({
'scope': self,
'time': self.time
})
if not self.lookups_loaded:
# Initialize HardcodedLookups elements
for element in self._lookup_elements:
element.initialize()
self.lookups_loaded = True
if not self.external_loaded:
# Initialize external elements
self.initialize_external_data()
# Initialize stateful objects
for element_name in self.initialize_order:
self._stateful_elements[element_name].initialize()
def ddt(self):
return np.array([component.ddt() for component
in self._dynamicstateful_elements], dtype=object)
@property
def state(self):
return np.array([component.state for component
in self._dynamicstateful_elements], dtype=object)
@state.setter
def state(self, new_value):
[component.update(val) for component, val
in zip(self._dynamicstateful_elements, new_value)]
[docs] def export(self, file_name):
"""
Export stateful values to pickle file.
Parameters
----------
file_name: str or pathlib.Path
Name of the file to export the values.
"""
warnings.warn(
"\nCompatibility of exported states could be broken between"
" different versions of PySD or xarray, current versions:\n"
f"\tPySD {__version__}\n\txarray {xr.__version__}\n"
)
stateful_elements = {
name: element.export()
for name, element in self._stateful_elements.items()
}
with open(file_name, 'wb') as file:
pickle.dump(
(self.time(),
stateful_elements,
{'pysd': __version__, 'xarray': xr.__version__}
), file)
[docs] def import_pickle(self, file_name):
"""
Import stateful values from pickle file.
Parameters
----------
file_name: str or pathlib.Path
Name of the file to import the values from.
"""
with open(file_name, 'rb') as file:
time, stateful_dict, metadata = pickle.load(file)
if __version__ != metadata['pysd']\
or xr.__version__ != metadata['xarray']: # pragma: no cover
warnings.warn(
"\nCompatibility of exported states could be broken between"
" different versions of PySD or xarray. Current versions:\n"
f"\tPySD {__version__}\n\txarray {xr.__version__}\n"
"Loaded versions:\n"
f"\tPySD {metadata['pysd']}\n\txarray {metadata['xarray']}\n"
)
self.set_stateful(stateful_dict)
self.time.set_control_vars(initial_time=time)
[docs] def initialize_external_data(self, externals=None):
"""
Initializes external data.
If a path to a netCDF file containing serialized values of some or
all of the model external data is passed in the external argument,
those will be loaded from the file.
To get the full performance gain of loading the externals from a netCDF
file, the model should be loaded with initialize=False first.
Parameters
----------
externals: str or pathlib.Path (optional)
Path to the netCDF file that contains the model external objects.
Returns
-------
None
"""
if not externals:
for ext in self._external_elements:
ext.initialize()
# Remove Excel data from memory
Excels.clean()
self.external_loaded = True
return
externals = Path(externals)
if not externals.is_file():
raise FileNotFoundError(f"Invalid file path ({str(externals)})")
ds = xr.open_dataset(externals)
for ext in self._external_elements:
if ext.py_name in ds.data_vars.keys():
# Initialize external from nc file
da = ds.data_vars[ext.py_name]
if isinstance(ext, ExtData):
# Rename again time dimension
time_dim = [dim for dim in da.dims if dim.startswith(
"time_#")][0]
da = da.rename({time_dim: "time"})
elif isinstance(ext, ExtLookup):
# Rename again lookup_dim dimension
lookup_dim = [dim for dim in da.dims if dim.startswith(
"lookup_dim_#")][0]
da = da.rename({lookup_dim: "lookup_dim"})
# Assign the value
if da.dims:
ext.data = da
else:
ext.data = float(da.data)
else:
# Initialize external from original file
ext.initialize()
Excels.clean()
self.external_loaded = True
[docs] def serialize_externals(self, export_path="externals.nc",
include_externals="all", exclude_externals=None):
"""
Stores a netCDF file with the data and metadata for all model external
objects.
This method is useful for models with lots of external inputs, which
are slow to load into memory. Once exported, the resulting netCDF file
can be passed as argument to the initialize_external_data method.
Names of variables should be those in the model (python safe,
without the _ext_type_ string in front).
Parameters
----------
export_path: str or pathlib.Path (optional)
Path of the resulting *.nc* file.
include_externals: list or str (optional)
External objects to export to netCDF.
If 'all', then all externals are exported to the *.nc* file.
The argument also accepts a list containing spreadsheet file names,
external variable names or a combination of both. If a spreadsheet
file path is passed, all external objects defined in it will be
included in the *.nc* file. Spreadsheet tab names are not currently
supported, because the same may be used in different files.
Better customisation can be achieved by combining the
include_externals and exclude_externals (see description below)
arguments.
exclude_externals: list or None (optional)
Exclude external objects from being included in the exported nc
file. It accepts either variable names, spreadsheet files or a
combination of both.
Returns
-------
None
"""
data = {}
metadata = {}
lookup_dims = utils.UniqueDims("lookup_dim")
data_dims = utils.UniqueDims("time")
if isinstance(export_path, str):
export_path = Path(export_path)
if not include_externals:
raise ValueError("include_externals argument must not be None.")
# Generate a Dataframe to make simpler the search of external
# objects to include
# TODO include also checking the original name
py_names = []
externals_dict = {"py_var_name": [], "file": [], "ext": []}
for ext in self._external_elements:
py_names.append(ext.py_name)
externals_dict["py_var_name"].append(
self.__get_varname_from_ext_name(ext.py_name))
externals_dict["file"].append(set(ext.files))
externals_dict["ext"].append(ext)
exts_df = pd.DataFrame(index=py_names, data=externals_dict)
if include_externals != "all":
if not isinstance(include_externals, (list, set)):
raise TypeError(
"include_externals must be 'all', or a list, or a set.")
# subset only to the externals to include
exts_df = exts_df[[
name in include_externals
or var_name in include_externals
or bool(file.intersection(include_externals))
for name, (var_name, file)
in exts_df[["py_var_name", "file"]].iterrows()
]]
if exclude_externals:
if not isinstance(exclude_externals, (list, set)):
raise TypeError("exclude_externals must be a list or a set.")
# subset only to the externals to include
exts_df = exts_df[[
name not in exclude_externals
and var_name not in exclude_externals
and not bool(file.intersection(exclude_externals))
for name, (var_name, file)
in exts_df[["py_var_name", "file"]].iterrows()
]]
for _, (ext, var_name) in exts_df[["ext", "py_var_name"]].iterrows():
self.__include_for_serialization(
ext, var_name, data, metadata, lookup_dims, data_dims
)
# create description to be used as global attribute of the dataset
description = {
"description": f"External objects for {self.py_model_file} "
f"exported on {time.ctime(time.time())} "
f"using PySD version {__version__}"
}
# create Dataset
ds = xr.Dataset(data_vars=data, attrs=description)
# add data_vars attributes
for key, values in metadata.items():
ds[key].attrs = values
ds.to_netcdf(export_path)
def __include_for_serialization(self, ext, py_name_clean, data, metadata,
lookup_dims, data_dims):
"""
Initialize the external object and get the data and metadata for
inclusion in the netCDF.
This function updates the metadata dict with the metadata corresponding
to each external object, which is collected from model.doc. It also
updates the data dict, with the data from the external object (ext).
It renames the "time" dimension of ExtData types by appending _#
followed by a unique number at the end of it. For large models, this
prevents having an unnecessary large time dimension, which then causes
all ExtData objects to have many nans when stored in a xarray Dataset.
It does the same for the "lookup_dim" of all ExtLookup objects.
NOTE: Though subscripts can be read from Excel, they are hardcoded
during the model building process. Therefore they will not be
serialized.
Parameters
----------
ext: pysd.py_backend.externals.External
External object. It can be any of the External subclasses
(ExtConstant, ExtData, ExtLookup)
py_name_clean: str
Name of the variable without _ext_[constant|data|lookup] prefix.
data: dict
Collects all the data for each external, which is later used to
build the xarray Dataset.
metadata: dict
Collects the metadata for each external, which is later included
as data_vars attributes in the xarray Dataset.
lookup_dims: utils.UniqueDims
UniqueDims object for "lookup_dim" dimension.
data_dims: utils.UniqueDims
UniqueDims object for "time" dimension.
Returns
-------
None
"""
ext.initialize()
# collecting variable metadata from model._doc
var_meta = {
col:
self._doc.loc[self._doc["Py Name"] == py_name_clean, col].values[0]
or "Missing"
for col in self._doc.columns
}
var_meta["files"] = ";".join(ext.files)
var_meta["sheets"] = ";".join(ext.sheets)
var_meta["cells"] = ";".join(ext.cells)
# TODO: add also time_row_or_cols
da = ext.data
# Renaming shared dims by all ExtData ("time") and ExtLookup
# ("lookup_dim") external objects
if isinstance(ext, ExtData):
new_name = data_dims.name_new_dim("time", da.coords["time"].values)
da = da.rename({"time": new_name})
if isinstance(ext, ExtLookup):
new_name = lookup_dims.name_new_dim("lookup_dim",
da.coords["lookup_dim"].values)
da = da.rename({"lookup_dim": new_name})
metadata.update({ext.py_name: var_meta})
data.update({ext.py_name: da})
# TODO use a logger
print(f"Finished processing variable {py_name_clean}.")
def __get_varname_from_ext_name(self, varname):
"""
Returns the name of the variable that depends on the external object
named varname. If that is not possible (see warning in the code to
understand when that may happen), it gets the name by removing the
_ext_[constant|data|lookup] prefix from the varname.
Parameters
----------
varname: str
Variable name to which the External object is assigned.
Returns
-------
var: str
Name of the variable that calls the variable with name varname.
"""
for var, deps in self._dependencies.items():
for _, ext_name in deps.items():
if varname == ext_name:
return var
warnings.warn(
f"No variable depends upon '{varname}'. This is likely due "
f"to the fact that '{varname}' is defined using a mix of "
"DATA and CONSTANT. Though Vensim allows it, it is "
"not recommended."
)
return "_".join(varname.split("_")[3:])
[docs] def get_args(self, param):
"""
Returns the arguments of a model element.
Parameters
----------
param: str or func
The model element name or function.
Returns
-------
args: list
List of arguments of the function.
Examples
--------
>>> model.get_args('birth_rate')
>>> model.get_args('Birth Rate')
"""
if isinstance(param, str):
func_name = utils.get_key_and_value_by_insensitive_key_or_value(
param,
self._namespace)[1] or param
func = getattr(self.components, func_name)
else:
func = param
if hasattr(func, 'args'):
# cached functions
return func.args
else:
# regular functions
args = inspect.getfullargspec(func)[0]
if 'self' in args:
args.remove('self')
return args
[docs] def get_coords(self, param):
"""
Returns the coordinates and dims of a model element.
Parameters
----------
param: str or func
The model element name or function.
Returns
-------
(coords, dims) or None: (dict, list) or None
The coords and the dimensions of the element if it has.
Otherwise, returns None.
Examples
--------
>>> model.get_coords('birth_rate')
>>> model.get_coords('Birth Rate')
"""
if isinstance(param, str):
func_name = utils.get_key_and_value_by_insensitive_key_or_value(
param,
self._namespace)[1] or param
func = getattr(self.components, func_name)
else:
func = param
if hasattr(func, "subscripts"):
dims = func.subscripts
if not dims:
return None
coords = {dim: self.components._subscript_dict[dim]
for dim in dims}
return coords, dims
elif hasattr(func, "state") and isinstance(func.state, xr.DataArray):
value = func()
else:
return None
dims = list(value.dims)
coords = {coord: list(value.coords[coord].values)
for coord in value.coords}
return coords, dims
def __getitem__(self, param):
"""
Returns the current value of a model component.
Parameters
----------
param: str or func
The model element name.
Returns
-------
value: float or xarray.DataArray
The value of the model component.
Examples
--------
>>> model['birth_rate']
>>> model['Birth Rate']
Note
----
It will crash if the model component takes arguments.
"""
func_name = utils.get_key_and_value_by_insensitive_key_or_value(
param,
self._namespace)[1] or param
if self.get_args(getattr(self.components, func_name)):
raise ValueError(
"Trying to get the current value of a lookup "
"to get all the values with the series data use "
"model.get_series_data(param)\n\n")
return getattr(self.components, func_name)()
[docs] def get_series_data(self, param):
"""
Returns the original values of a model lookup/data component.
Parameters
----------
param: str
The model lookup/data element name.
Returns
-------
value: xarray.DataArray
Array with the value of the interpolating series
in the first dimension.
Examples
--------
>>> model['room_temperature']
>>> model['Room temperature']
"""
func_name = utils.get_key_and_value_by_insensitive_key_or_value(
param,
self._namespace)[1] or param
if func_name.startswith("_ext_"):
return getattr(self.components, func_name).data
elif "__data__" in self._dependencies[func_name]:
return getattr(
self.components,
self._dependencies[func_name]["__data__"]
).data
elif "__lookup__" in self._dependencies[func_name]:
return getattr(
self.components,
self._dependencies[func_name]["__lookup__"]
).data
else:
raise ValueError(
"Trying to get the values of a constant variable. "
"'model.get_series_data' only works lookups/data objects.\n\n")
[docs] def set_components(self, params, new=False):
""" Set the value of exogenous model elements.
Element values can be passed as keyword=value pairs in the
function call. Values can be numeric type or pandas Series.
Series will be interpolated by integrator.
Examples
--------
>>> model.set_components({'birth_rate': 10})
>>> model.set_components({'Birth Rate': 10})
>>> br = pandas.Series(index=range(30), values=np.sin(range(30))
>>> model.set_components({'birth_rate': br})
"""
# TODO: allow the params argument to take a pandas dataframe, where
# column names are variable names. However some variables may be
# constant or have no values for some index. This should be processed.
# TODO: make this compatible with loading outputs from other files
for key, value in params.items():
func_name = utils.get_key_and_value_by_insensitive_key_or_value(
key,
self._namespace)[1]
if isinstance(value, np.ndarray) or isinstance(value, list):
raise TypeError(
'When setting ' + key + '\n'
'Setting subscripted must be done using a xarray.DataArray'
' with the correct dimensions or a constant value '
'(https://pysd.readthedocs.io/en/master/'
'getting_started.html)')
if func_name is None:
raise NameError(
"\n'%s' is not recognized as a model component."
% key)
if new:
func = None
dims = None
else:
func = getattr(self.components, func_name)
_, dims = self.get_coords(func) or (None, None)
# if the variable is a lookup or a data we perform the change in
# the object they call
func_type = getattr(func, "type", None)
if func_type in ["Lookup", "Data"]:
# getting the object from original dependencies
obj = self._dependencies[func_name][f"__{func_type.lower()}__"]
getattr(
self.components,
obj
).set_values(value)
# Update dependencies
if func_type == "Data":
if isinstance(value, pd.Series):
self._dependencies[func_name] = {
"time": 1, "__data__": obj
}
else:
self._dependencies[func_name] = {"__data__": obj}
continue
if isinstance(value, pd.Series):
new_function, deps = self._timeseries_component(
value, dims)
self._dependencies[func_name] = deps
elif callable(value):
new_function = value
# Using step cache adding time as dependency
# TODO it would be better if we can parse the content
# of the function to get all the dependencies
self._dependencies[func_name] = {"time": 1}
else:
new_function = self._constant_component(value, dims)
self._dependencies[func_name] = {}
# this won't handle other statefuls...
if '_integ_' + func_name in dir(self.components):
warnings.warn("Replacing the equation of stock "
"'{}' with params...".format(key),
stacklevel=2)
new_function.__name__ = func_name
if dims:
new_function.dims = dims
self.components._set_component(func_name, new_function)
if func_name in self.cache.cached_funcs:
self.cache.cached_funcs.remove(func_name)
def _timeseries_component(self, series, dims):
""" Internal function for creating a timeseries model element """
# this is only called if the set_component function recognizes a
# pandas series
# TODO: raise a warning if extrapolating from the end of the series.
# TODO: data type variables should be creted using a Data object
# lookup type variables should be created using a Lookup object
if isinstance(series.values[0], xr.DataArray):
# the interpolation will be time dependent
return lambda: utils.rearrange(xr.concat(
series.values,
series.index).interp(concat_dim=self.time()).reset_coords(
'concat_dim', drop=True),
dims, self._subscript_dict), {'time': 1}
elif dims:
# the interpolation will be time dependent
return lambda: utils.rearrange(
np.interp(self.time(), series.index, series.values),
dims, self._subscript_dict), {'time': 1}
else:
# the interpolation will be time dependent
return lambda:\
np.interp(self.time(), series.index, series.values),\
{'time': 1}
def _constant_component(self, value, dims):
""" Internal function for creating a constant model element """
if dims:
return lambda: utils.rearrange(
value, dims, self._subscript_dict)
else:
return lambda: value
[docs] def set_initial_value(self, t, initial_value):
""" Set the system initial value.
Parameters
----------
t : numeric
The system time
initial_value : dict
A (possibly partial) dictionary of the system initial values.
The keys to this dictionary may be either pysafe names or
original model file names
"""
self.time.set_control_vars(initial_time=t)
stateful_name = "_NONE"
modified_statefuls = set()
for key, value in initial_value.items():
component_name =\
utils.get_key_and_value_by_insensitive_key_or_value(
key, self._namespace)[1]
if component_name is not None:
if self._dependencies[component_name]:
deps = list(self._dependencies[component_name])
if len(deps) == 1 and deps[0] in self.initialize_order:
stateful_name = deps[0]
else:
component_name = key
stateful_name = key
try:
_, dims = self.get_coords(component_name)
except TypeError:
dims = None
if isinstance(value, xr.DataArray)\
and not set(value.dims).issubset(set(dims)):
raise ValueError(
f"\nInvalid dimensions for {component_name}."
f"It should be a subset of {dims}, "
f"but passed value has {list(value.dims)}")
if isinstance(value, np.ndarray) or isinstance(value, list):
raise TypeError(
'When setting ' + key + '\n'
'Setting subscripted must be done using a xarray.DataArray'
' with the correct dimensions or a constant value '
'(https://pysd.readthedocs.io/en/master/'
'getting_started.html)')
# Try to update stateful component
try:
element = getattr(self.components, stateful_name)
if dims:
value = utils.rearrange(
value, dims,
self._subscript_dict)
element.initialize(value)
modified_statefuls.add(stateful_name)
except NameError:
# Try to override component
raise ValueError(
f"\nUnrecognized stateful '{component_name}'. If you want"
" to set a value of a regular component. Use params={"
f"'{component_name}': {value}" + "} instead.")
self.clean_caches()
# get the elements to initialize
elements_to_initialize =\
self._get_elements_to_initialize(modified_statefuls)
# Initialize remaining stateful objects
for element_name in self.initialize_order:
if element_name in elements_to_initialize:
self._stateful_elements[element_name].initialize()
def _get_elements_to_initialize(self, modified_statefuls):
elements_to_initialize = set()
for stateful, deps in self.stateful_initial_dependencies.items():
if stateful in modified_statefuls:
# if elements initial conditions have been modified
# we should not modify it
continue
for modified_sateteful in modified_statefuls:
if modified_sateteful in deps:
# if element has dependencies on a modified element
# we should re-initialize it
elements_to_initialize.add(stateful)
continue
return elements_to_initialize
[docs] def set_stateful(self, stateful_dict):
"""
Set stateful values.
Parameters
----------
stateful_dict: dict
Dictionary of the stateful elements and the attributes to change.
"""
for element, attrs in stateful_dict.items():
for attr, value in attrs.items():
setattr(getattr(self.components, element), attr, value)
def _build_doc(self):
"""
Formats a table of documentation strings to help users remember
variable names, and understand how they are translated into
Python safe names.
Returns
-------
docs_df: pandas dataframe
Dataframe with columns for the model components:
- Real names
- Python safe identifiers (as used in model.components)
- Units string
- Documentation strings from the original model file
"""
collector = []
for name, pyname in self._namespace.items():
element = getattr(self.components, pyname)
collector.append({
'Real Name': name,
'Py Name': pyname,
'Subscripts': element.subscripts,
'Units': element.units,
'Limits': element.limits,
'Type': element.type,
'Subtype': element.subtype,
'Comment': element.__doc__.strip().strip("\n").strip()
if element.__doc__ else None
})
return pd.DataFrame(
collector
).sort_values(by="Real Name").reset_index(drop=True)
def __str__(self):
""" Return model source files """
# JT: Might be helpful to return not only the source file, but
# also how the instance differs from that source file. This
# would give a more accurate view of the current model.
string = 'Translated Model File: ' + self.py_model_file
if hasattr(self, 'mdl_file'):
string += '\n Original Model File: ' + self.mdl_file
return string
[docs]class Model(Macro):
"""
The Model class implements a stateful representation of the system.
It inherits methods from the Macro class to integrate the model and
access and modify model components. It also contains the main
methods for running the model.
The Model object will be created with components drawn from a
translated Python model file.
Parameters
----------
py_model_file: str or pathlib.Path
Filename of a model which has already been converted into a
Python format.
data_files: dict or list or str or None
The dictionary with keys the name of file and variables to
load the data from there. Or the list of names or name of the
file to search the data in. Only works for TabData type object
and it is neccessary to provide it. Default is None.
initialize: bool
If False, the model will not be initialize when it is loaded.
Default is True.
missing_values : str ("warning", "error", "ignore", "keep") (optional)
What to do with missing values. If "warning" (default)
shows a warning message and interpolates the values.
If "raise" raises an error. If "ignore" interpolates
the values without showing anything. If "keep" it will keep
the missing values, this option may cause the integration to
fail, but it may be used to check the quality of the data.
"""
def __init__(self, py_model_file, data_files, initialize, missing_values):
""" Sets up the Python objects """
super().__init__(py_model_file, None, None, Time(),
data_files=data_files)
self.time.stage = 'Load'
self.time.set_control_vars(**self.components._control_vars)
self.data_files = data_files
self.missing_values = missing_values
self.progress = None
if initialize:
self.initialize()
[docs] def initialize(self):
""" Initializes the simulation model """
self.time.stage = 'Initialization'
External.missing = self.missing_values
super().initialize()
[docs] def run(self, params=None, return_columns=None, return_timestamps=None,
initial_condition='original', final_time=None, time_step=None,
saveper=None, reload=False, progress=False, flatten_output=True,
cache_output=True, output_file=None):
"""
Simulate the model's behavior over time.
Return a pandas dataframe with timestamps as rows,
model elements as columns.
Parameters
----------
params: dict (optional)
Keys are strings of model component names.
Values are numeric or pandas Series.
Numeric values represent constants over the model integration.
Timeseries will be interpolated to give time-varying input.
return_timestamps: list, numeric, ndarray (1D) (optional)
Timestamps in model execution at which to return state information.
Defaults to model-file specified timesteps.
return_columns: list, 'step' or None (optional)
List of string model component names, returned dataframe
will have corresponding columns. If 'step' only variables with
cache step will be returned. If None, variables with cache step
and run will be returned. Default is None.
initial_condition: str or (float, dict) (optional)
The starting time, and the state of the system (the values of
all the stocks) at that starting time. 'original' or 'o'uses
model-file specified initial condition. 'current' or 'c' uses
the state of the model after the previous execution. Other str
objects, loads initial conditions from the pickle file with the
given name.(float, dict) tuple lets the user specify a starting
time (float) and (possibly partial) dictionary of initial values
for stock (stateful) objects. Default is 'original'.
final_time: float or None
Final time of the simulation. If float, the given value will be
used to compute the return_timestamps (if not given) and as a
final time. If None the last value of return_timestamps will be
used as a final time. Default is None.
time_step: float or None
Time step of the simulation. If float, the given value will be
used to compute the return_timestamps (if not given) and
euler time series. If None the default value from components
will be used. Default is None.
saveper: float or None
Saving step of the simulation. If float, the given value will be
used to compute the return_timestamps (if not given). If None
the default value from components will be used. Default is None.
reload : bool (optional)
If True, reloads the model from the translated model file
before making changes. Default is False.
progress : bool (optional)
If True, a progressbar will be shown during integration.
Default is False.
flatten_output: bool (optional)
If True, once the output dataframe has been formatted will
split the xarrays in new columns following Vensim's naming
to make a totally flat output. Default is True.
This argument will be ignored when passing a netCDF4 file
path in the output_file argument.
cache_output: bool (optional)
If True, the number of calls of outputs variables will be increased
in 1. This helps caching output variables if they are called only
once. For performance reasons, if time step = saveper it is
recommended to activate this feature, if time step << saveper
it is recommended to deactivate it. Default is True.
output_file: str, pathlib.Path or None (optional)
Path of the file in which to save simulation results.
Currently, csv, tab and nc (netCDF4) files are supported.
Examples
--------
>>> model.run(params={'exogenous_constant': 42})
>>> model.run(params={'exogenous_variable': timeseries_input})
>>> model.run(return_timestamps=[1, 2, 3, 4, 10])
>>> model.run(return_timestamps=10)
>>> model.run(return_timestamps=np.linspace(1, 10, 20))
>>> model.run(output_file="results.nc")
See Also
--------
pysd.set_components : handles setting model parameters
pysd.set_initial_condition : handles setting initial conditions
"""
if reload:
self.reload()
self.time.add_return_timestamps(return_timestamps)
if self.time.return_timestamps is not None and not final_time:
# if not final time given the model will end in the list
# return timestamp (the list is reversed for popping)
if self.time.return_timestamps:
final_time = self.time.return_timestamps[0]
else:
final_time = self.time.next_return
self.time.set_control_vars(
final_time=final_time, time_step=time_step, saveper=saveper)
if params:
self.set_components(params)
# update cache types after setting params
self._assign_cache_type()
self.set_initial_condition(initial_condition)
# set progressbar
if progress and (self.cache_type["final_time"] == "step" or
self.cache_type["time_step"] == "step"):
warnings.warn(
"The progressbar is not compatible with dynamic "
"final time or time step. Both variables must be "
"constants to prompt progress."
)
progress = False
self.progress = progress
if return_columns is None or isinstance(return_columns, str):
return_columns = self._default_return_columns(return_columns)
capture_elements, return_addresses = utils.get_return_elements(
return_columns, self._namespace)
# create a dictionary splitting run cached and others
capture_elements = self._split_capture_elements(capture_elements)
# include outputs in cache if needed
self._dependencies["OUTPUTS"] = {
element: 1 for element in capture_elements["step"]
}
if cache_output:
# udate the cache type taking into account the outputs
self._assign_cache_type()
# check validitty of output_file. This could be done inside the
# ModelOutput class, but it feels too late
if output_file:
if not isinstance(output_file, (str, Path)):
raise TypeError(
"Paths must be strings or pathlib Path objects.")
if isinstance(output_file, str):
output_file = Path(output_file)
file_extension = output_file.suffix
if file_extension not in ModelOutput.valid_output_files:
raise ValueError(
f"Unsupported output file format {file_extension}")
# add constant cache to thosa variable that are constants
self._add_constant_cache()
# Run the model
self.time.stage = 'Run'
# need to clean cache to remove the values from active_initial
self.clean_caches()
# instantiating output object
output = ModelOutput(self, capture_elements['step'], output_file)
self._integrate(output)
del self._dependencies["OUTPUTS"]
output.add_run_elements(self, capture_elements['run'])
self._remove_constant_cache()
return output.postprocess(
return_addresses=return_addresses, flatten=flatten_output)
[docs] def select_submodel(self, vars=[], modules=[], exogenous_components={}):
"""
Select a submodel from the original model. After selecting a submodel
only the necessary stateful objects for integrating this submodel will
be computed.
Parameters
----------
vars: set or list of strings (optional)
Variables to include in the new submodel.
It can be an empty list if the submodel is only selected by
module names. Default is an empty list.
modules: set or list of strings (optional)
Modules to include in the new submodel.
It can be an empty list if the submodel is only selected by
variable names. Default is an empty list. Can select a full
module or a submodule by passing the path without the .py, e.g.:
"view_1/submodule1".
exogenous_components: dictionary of parameters (optional)
Exogenous value to fix to the model variables that are needed
to run the selected submodel. The exogenous_components should
be passed as a dictionary in the same way it is done for
set_components method. By default it is an empty dict and
the needed exogenous components will be set to a numpy.nan value.
Returns
-------
None
Notes
-----
modules can be only passed when the model has been split in
different files during translation.
Examples
--------
>>> model.select_submodel(
... vars=["Room Temperature", "Teacup temperature"])
UserWarning: Selecting submodel, to run the full model again use model.reload()
>>> model.select_submodel(
... modules=["view_1", "view_2/subview_1"])
UserWarning: Selecting submodel, to run the full model again use model.reload()
UserWarning: Exogenous components for the following variables are necessary but not given:
initial_value_stock1, stock3
>>> model.select_submodel(
... vars=["stock3"],
... modules=["view_1", "view_2/subview_1"])
UserWarning: Selecting submodel, to run the full model again use model.reload()
UserWarning: Exogenous components for the following variables are necessary but not given:
initial_value_stock1, initial_value_stock3
Please, set them before running the model using set_components method...
>>> model.select_submodel(
... vars=["stock3"],
... modules=["view_1", "view_2/subview_1"],
... exogenous_components={
... "initial_value_stock1": 3,
... "initial_value_stock3": 5})
UserWarning: Selecting submodel, to run the full model again use model.reload()
"""
deps = self.get_dependencies(vars, modules)
warnings.warn(
"Selecting submodel, "
"to run the full model again use model.reload()")
# get set of all dependencies and all variables to select
all_deps = deps.d_deps["initial"].copy()
all_deps.update(deps.d_deps["step"])
all_deps.update(deps.d_deps["lookup"])
all_vars = all_deps.copy()
all_vars.update(deps.c_vars)
# clean dependendies and namespace dictionaries, and remove
# the rows from the documentation
for real_name, py_name in self._namespace.copy().items():
if py_name not in all_vars:
del self._namespace[real_name]
del self._dependencies[py_name]
self._doc.drop(
self._doc.index[self._doc["Real Name"] == real_name],
inplace=True
)
for py_name in self._dependencies.copy().keys():
if py_name.startswith("_") and py_name not in deps.s_deps:
del self._dependencies[py_name]
# remove active initial from s_deps as they are "fake" objects
# in dependencies
deps.s_deps = {
dep for dep in deps.s_deps if not dep.startswith("_active_initial")
}
# reassing the dictionary and lists of needed stateful objects
self._stateful_elements = {
name: getattr(self.components, name)
for name in deps.s_deps
if isinstance(getattr(self.components, name), Stateful)
}
self._dynamicstateful_elements = [
getattr(self.components, name) for name in deps.s_deps
if isinstance(getattr(self.components, name), DynamicStateful)
]
self._macro_elements = [
getattr(self.components, name) for name in deps.s_deps
if isinstance(getattr(self.components, name), Macro)
]
# keeping only needed external objects
ext_deps = set()
for values in self._dependencies.values():
if "__external__" in values:
ext_deps.add(values["__external__"])
self._external_elements = [
getattr(self.components, name) for name in ext_deps
if isinstance(getattr(self.components, name), External)
]
# set all exogenous values to np.nan by default
new_components = {element: np.nan for element in all_deps}
# update exogenous values with the user input
[new_components.update(
{
utils.get_key_and_value_by_insensitive_key_or_value(
key,
self._namespace)[1]: value
}) for key, value in exogenous_components.items()]
self.set_components(new_components)
# show a warning message if exogenous values are needed for a
# dependency
new_components = [
key for key, value in new_components.items() if value is np.nan]
if new_components:
warnings.warn(
"Exogenous components for the following variables are "
f"necessary but not given:\n\t{', '.join(new_components)}"
"\n\n Please, set them before running the model using "
"set_components method...")
# re-assign the cache_type and initialization order
self._assign_cache_type()
self._get_initialize_order()
[docs] def get_dependencies(self, vars=[], modules=[]):
"""
Get the dependencies of a set of variables or modules.
Parameters
----------
vars: set or list of strings (optional)
Variables to get the dependencies from.
It can be an empty list if the dependencies are computed only
using modules. Default is an empty list.
modules: set or list of strings (optional)
Modules to get the dependencies from.
It can be an empty list if the dependencies are computed only
using variables. Default is an empty list. Can select a full
module or a submodule by passing the path without the .py, e.g.:
"view_1/submodule1".
Returns
-------
dependencies: pysd.py_backend.utils.Dependencies
Dependencies data object.
Notes
-----
modules can be only passed when the model has been split in
different files during translation.
Examples
--------
>>> print(model.get_dependencies(
... vars=["Room Temperature", "Teacup temperature"]))
Selected variables (total 1):
room_temperature, teacup_temperature
Stateful objects integrated with the selected variables (total 1):
_integ_teacup_temperature
>>> print(model.get_dependencies(
... modules=["view_1", "view_2/subview_1"]))
Selected variables (total 4):
var1, var2, stock1, delay1
Dependencies for initialization only (total 1):
initial_value_stock1
Dependencies that may change over time (total 2):
stock3
Stateful objects integrated with the selected variables (total 1):
_integ_stock1, _delay_fixed_delay1
>>> print(model.get_dependencies(
... vars=["stock3"],
... modules=["view_1", "view_2/subview_1"]))
Selected variables (total 4):
var1, var2, stock1, stock3, delay1
Dependencies for initialization only (total 1):
initial_value_stock1, initial_value_stock3
Stateful objects integrated with the selected variables (total 1):
_integ_stock1, _integ_stock3, _delay_fixed_delay1
"""
def check_dep(deps_obj, deps, initial=False):
for dep in deps:
if dep in deps_obj.c_vars or dep.startswith("__"):
pass
elif dep.startswith("_"):
deps_obj.s_deps.add(dep)
dep = self._dependencies[dep]
check_dep(deps_obj, dep["initial"], True)
check_dep(deps_obj, dep["step"])
else:
if initial and dep not in deps_obj.d_deps["step"]\
and dep not in deps_obj.d_deps["lookup"]:
deps_obj.d_deps["initial"].add(dep)
else:
if dep in deps_obj.d_deps["initial"]:
deps_obj.d_deps["initial"].remove(dep)
if self.get_args(dep):
deps_obj.d_deps["lookup"].add(dep)
else:
deps_obj.d_deps["step"].add(dep)
dependencies = utils.Dependencies(
{"time", "time_step", "initial_time", "final_time", "saveper"},
{"initial": set(), "step": set(), "lookup": set()},
set()
)
for var in vars:
py_name = utils.get_key_and_value_by_insensitive_key_or_value(
var,
self._namespace)[1]
dependencies.c_vars.add(py_name)
for module in modules:
dependencies.c_vars.update(self.get_vars_in_module(module))
for var in dependencies.c_vars:
if var == "time":
continue
check_dep(dependencies, self._dependencies[var])
return dependencies
[docs] def get_vars_in_module(self, module):
"""
Return the name of Python vars in a module.
Parameters
----------
module: str
Name of the module to search in.
Returns
-------
vars: set
Set of varible names in the given module.
"""
if self._modules:
module_content = self._modules.copy()
else:
raise ValueError(
"Trying to get a module from a non-modularized model")
try:
# get the module or the submodule content
for submodule in module.split("/"):
module_content = module_content[submodule]
module_content = [module_content]
except KeyError:
raise NameError(
f"Module or submodule '{submodule}' not found...\n")
vars, new_content = set(), []
while module_content:
# find the vars in the module or the submodule
for content in module_content:
if isinstance(content, list):
vars.update(content)
else:
[new_content.append(value) for value in content.values()]
module_content, new_content = new_content, []
return vars
[docs] def reload(self):
"""
Reloads the model from the translated model file, so that all the
parameters are back to their original value.
"""
self.__init__(self.py_model_file, data_files=self.data_files,
initialize=True,
missing_values=self.missing_values)
def _default_return_columns(self, which):
"""
Return a list of the model elements tha change on time that
does not include lookup other functions that take parameters
or run-cached functions.
Parameters
----------
which: str or None
If it is 'step' only cache step elements will be returned.
Else cache 'step' and 'run' elements will be returned.
Default is None.
Returns
-------
return_columns: list
List of columns to return
"""
if which == 'step':
types = ['step']
else:
types = ['step', 'run']
return_columns = []
for key, pykey in self._namespace.items():
if pykey in self.cache_type and self.cache_type[pykey] in types\
and not self.get_args(pykey):
return_columns.append(key)
return return_columns
def _split_capture_elements(self, capture_elements):
"""
Splits the capture elements list between those with run cache
and others.
Parameters
----------
capture_elements: list
Captured elements list
Returns
-------
capture_dict: dict
Dictionary of list with keywords step and run.
"""
capture_dict = {'step': [], 'run': [], None: []}
[capture_dict[self.cache_type[element]].append(element)
for element in capture_elements]
return capture_dict
[docs] def set_initial_condition(self, initial_condition):
""" Set the initial conditions of the integration.
Parameters
----------
initial_condition : str or (float, dict) or pathlib.Path
The starting time, and the state of the system (the values of
all the stocks) at that starting time. 'original' or 'o'uses
model-file specified initial condition. 'current' or 'c' uses
the state of the model after the previous execution. Other str
objects, loads initial conditions from the pickle file with the
given name.(float, dict) tuple lets the user specify a starting
time (float) and (possibly partial) dictionary of initial values
for stock (stateful) objects.
Examples
--------
>>> model.set_initial_condition('original')
>>> model.set_initial_condition('current')
>>> model.set_initial_condition('exported_pickle.pic')
>>> model.set_initial_condition((10, {'teacup_temperature': 50}))
See Also
--------
model.set_initial_value()
"""
if isinstance(initial_condition, str)\
and initial_condition.lower() not in ["original", "o",
"current", "c"]:
initial_condition = Path(initial_condition)
if isinstance(initial_condition, tuple):
self.initialize()
self.set_initial_value(*initial_condition)
elif isinstance(initial_condition, Path):
self.import_pickle(initial_condition)
elif isinstance(initial_condition, str):
if initial_condition.lower() in ["original", "o"]:
self.time.set_control_vars(
initial_time=self.components._control_vars["initial_time"])
self.initialize()
else:
raise TypeError(
"Invalid initial conditions. "
+ "Check documentation for valid entries or use "
+ "'help(model.set_initial_condition)'.")
def _euler_step(self, dt):
"""
Performs a single step in the euler integration,
updating stateful components
Parameters
----------
dt : float
This is the amount to increase time by this step
"""
self.state = self.state + self.ddt() * dt
def _integrate(self, out_obj):
"""
Performs euler integration and writes results to the out_obj.
Parameters
----------
out_obj: pysd.ModelOutput
Returns
-------
None
"""
if self.progress:
# initialize progress bar
progressbar = utils.ProgressBar(
int((self.time.final_time()-self.time())/self.time.time_step())
)
else:
# when None is used the update will do nothing
progressbar = utils.ProgressBar(None)
# performs the time stepping
while self.time.in_bounds():
if self.time.in_return():
out_obj.update(self)
self._euler_step(self.time.time_step())
self.time.update(self.time()+self.time.time_step())
self.clean_caches()
progressbar.update()
# need to add one more time step, because we run only the state
# updates in the previous loop and thus may be one short.
if self.time.in_return():
out_obj.update(self)
progressbar.finish()