v0.2.1

[caimichael]

New features

• A basic package structure to be able to utilize the sequence_jacobian as a python package. Top-level functions and modules located within sequence_jacobian/__init__.py.

• steady_state, which automatically solves for the steady state values of the parameters/variables for a given model specification, e.g. a set of blocks, fixed parameter/variable values, unknowns, and targets.

  Implemented a rudimentary new block type, HelperBlock, which allows users to include partial steady state solutions as a "block" in the blocks argument provided to steady_state to speed up the computation of the rest of the steady state values, e.g. in the Krusell Smith model, the firm's problem can be solved so steady_state effectively only needs to solve the household problem.

• Enhanced SimpleBlock with more flexible displacement handlers (Ignore, Displace, and the newly implemented IgnoreVector and AccumulatedDerivative [replacing Reporter and Perturb]) to allow for more general specifications of equations within a SimpleBlock.

  This includes the ability to call time displacements on variables newly defined within a SimpleBlock, to evaluate variables using arbitrary functions—single-argument functions are called by invoking the .apply(f) method of any variable within a SimpleBlock, and multi-argument functions are called by invoking apply_function(*args)—while maintaining displacement-relevant information, and to write equations that involve arbitrary compositions of newly-defined variables with input arguments into the SimpleBlock.

• A pytest testing suite located in sequence_jacobian/tests/base, which unit tests for the basic functionality of the Sequence Space Jacobian package. This includes testing most of the computations displaced in the Jupyter notebooks included with this package.

• hetoutput—functions of the individual-level policy variables from a given HetBlock—can now be included as outputs users get from steady state and transitional dynamic computations by using the add_hetoutput features in sequence_jacobian/blocks/het_block.py. Currently hetoutput are limited to being the same dimension as the cross-sectional distribution D and hence there is no capability for "custom" aggregation of hetoutputs. All hetoutputs are aggregated linearly across D, just as the standard individual-level policy variables are aggregated across D.

  An example of a hetoutput is the adjustment_costs of agents across the two-dimensional asset distribution in the two asset HANK model located in sequence_jacobian/models/two_asset.py. This hetoutput can be added to the household block using household.add_hetoutput(adjustment_costs).

Work-in-progress features

These features exist in the code currently but in an unfinished state.

• CombinedBlock in sequence_jacobian/blocks/combined_block.py currently works when included in the set of blocks called in steady_statei, i.e. .ss for a concrete instance of the block works; however, transitional dynamics (.td) and calculation of the Jacobian (.jac) have yet to be implemented.
• HelperBlock in sequence_jacobian/blocks/helper_block.py currently works when included in the set of blocks for steady state, transitional dynamics, and Jacobian calculations; however, most of its functionality is not contained within the HelperBlock class itself, but rather in other code such as block_sort, which checks if blocks are of HelperBlock when deciding how to conduct the sort. May revisit the design principle behind having a block that is virtually identical to a SimpleBlock albeit with more limited abilities as a stand-alone class simple as a way for methods to inspect differently.
• Block, and Model in sequence_jacobian/primitives.py are meant to be abstract base classes to provide more unifying concepts within the Sequence Space Jacobian package.
• Performance-based tests in sequence_jacobian/tests/performance and more granular unit tests in sequence_jacobian/tests/base and sequence_jacobian/tests/robustness.

Breaking changes

• Re-organized most of the file locations when implementing the basic package structure. Absolute/relative imports of local modules changed due to this.
  In particular, sequence_jacobian/utils.py has been broken up into separate modules contained in the sequence_jacobian/utilities directory to make it easier for users to locate utilities and define new ones.
• Moved SimpleSparse class to sequence_jacobian/jacobian.py
• Changed all keyword arguments noisy, set to True if the user wanted a particular function to provide print statements of its functionality, to verbose to be more descriptive.
• Changed attach_hetinput to add_hetinput and instead of returning a new instance of the HetBlock, it modifies the existing HetBlock in place. Note add_hetinput has an overwrite kwarg to allow users to switch out existing hetinputs for new ones.
• Changed the names of the various attributes of HetBlock to be slightly more descriptive. Some examples include: non_back_outputs -> non_back_iter_outputs, and inputs_p -> inputs_to_be_primed.

[caimichael]

Updated with fixes to notebook code and unit tests! Released as the patch v0.2.1

[caimichael]

I will introduce these changes onto a new branch in the repository for now, so as to not disrupt work currently happening on the `develop' branch