modelx v0.13.0 (23 Feb 2021)#

This release introduces the following enhancements.

Enhancements#

Introduction of the new_module method#

modelx has allowed to reference modules from within Models just by assigning modules to References. For example, to reference numpy from a Space, the user can just assign the module to a Reference:

>>> import numpy as np

>>> space.np = np

This works for modules in the Python standard library and in third-party packages, such as math and numpy. When Models referencing these modules are saved, those modules themselves are not saved within the Models but the names of the modules are. When the Models are read back, the Python interpreter is able to find those modules thanks to Python’s import system, and the modules are properly referenced again.

However, the user would also want to reference modules written by the user (user modules). User modules are usually not installed as packages in the Python’s import system. Rather, they are located in the user’s current directory. Referencing such modules in Models by the same assignment operation is problematic, because when such Models are saved and read back, the current directory may have changed or the referenced modules may have been moved or deleted by the user.

To make user modules portable, the Model.new_module and UserSpace.new_module methods are introduced. The methods allow the user to define a Reference, assign a user module to the Reference, and associate the user module with a source file of the module in the model directory, so that the module’s source code can be saved within the containing model.

Example

Suppose the following code is saved in “sample.py” in the current directory.

def triple(x)
    return 3 * x

The code below creates a Reference named “foo” in space:

>>> space.new_module("foo", "modules/sample.py", "sample.py")

The module becomes accessible as foo in space:

>>> space.foo
<module 'sample' from 'C:\\path\\to\\samplemodule.py'>

>>> @mx.defcells(space)
... def bar(y):
        return foo.triple(y)

>>> space.foo.bar(3)
9

Let model be the ultimate parent model of space. The next code creates a directory named “model” under the current directory, and within the “model” directory, the module is saved as “sample.py” in the “modules” sub-directory of the “model” dir, as specified by the path parameter to this method.

>>> model.write("model")

Direct assignment of DataFrame/Series by new_pandas#

UserSpace.new_pandas ,the previously introduced method, as well as Model.new_pandas now has a new default behaviour of assigning a pandas DataFrame/Series object passed as the data parameter to a Reference, instead of assigning the PandasData object associated with the pandas object. By passing False to the newly introduced expose_data parameter, the default behaviour can be altered to be consistent with the previous behavior, which is to assign the PandasData object to the Reference instead of the pandas object itself.

Backward Incompatible Changes#

  • Models saved by the previous modelx version (v0.12.1) works perfectly fine with this version. However, the default behaviour of UserSpace.new_pandas and Model.new_pandas has changed and they assign pandas objects(DataFrame or Series) to References directly instead of assigning PandasData objects. If False is given to the exposed_data parameter, the behaviour of the methods are consistent with the previous version, which is to assign PandasData objects rather than pandas objects(DataFrame or Series) themselves.