Skip to main content

Changelog

0.5.3​

  • cai_causal_graph.utils.get_variable_name_and_lag now allows new lines in the names of variables.
  • Passing input_list, output_list, and fully_connected as arguments to the constructor of a CausalGraph is now deprecated, and will be removed in future versions. To maintain this behavior, call add_fully_connected_nodes after construction.
  • Upgraded poetry version from 1.8.2 to 1.8.3 in the GitHub workflows.

0.5.2​

0.5.1​

0.5.0​

NOTE: Backwards compatibility warning! Global metadata to TimeSeriesCausalGraph and CausalGraph, which means their serialized state has changed.

NOTE: Backwards compatibility warning! The default value of the include_all_parents argument to extend_graph has been changed to True (from False).

0.4.10​

  • Improved documentation by cleaning up a few syntax issues to support the new docs building process.

0.4.9​

  • Fixed a bug where metadata on floating nodes would not be correctly carried over to the minimal graph when calling get_minimal_graph. This also fixes an issue where is_minimal_graph could return False on minimal graphs that contain floating nodes.

0.4.8​

  • Upgraded docs-builder dependency to "~0.2.1" in the Makefile and updated syntax to support newer poetry.

0.4.7​

  • Added the cai_causal_graph.causal_graph.identify_utils.identity_colliders utility function, which allows you to identify a list of colliders in a CausalGraph.
  • Improved efficiency of get_nodes_at_lag and get_nodes_for_variable_name by adding caching.
  • Updated networkx dependency from ">=3.0.0, <4.0.0" to ">=3.0.0, <3.3.0" and networkx have claimed to fix a bug with simple paths and other related functions, but this now introduces paths of length 1 which we do not want. Also updated pyproject.toml to allow different networkx versions for python 3.8 and 3.9 - 3.12 as networkx3.1 is the last one that supports 3.8.
  • Upgraded poetry version from 1.7.1 to 1.8.2 in the GitHub workflows.

0.4.6​

  • Added the Boolean keyword argument include_all_parents to extend_graph. When True extra nodes may be added. Nodes and edges will be added as far back that all nodes up to backward_steps in the past have all their parents and inbound edges. This means that the extended graph may now have nodes at lags further back than backward_steps. Default is False, meaning the default behavior of the method has not changed.
  • Added the Boolean keyword argument construct_minimal to cai_causal_graph.time_series_causal_graph.TimeSeriesCausalGraph.from_adjacency_matrices to allow the user to specify whether to construct a minimal graph from adjacency matrices. Default is True, making the change backwards compatible.
  • Fixed a bug in the cai_causal_graph.time_series_causal_graph.TimeSeriesCausalGraph.from_adjacency_matrices method of TimeSeriesCausalGraph that was introduced in the previous release. The method was not properly handling undirected edges in the adjacency matrices.
  • Fixed a bug in the to_numpy_by_lag method of TimeSeriesCausalGraph where it was not always returning the correct variable names.
  • Fixed a bug in the adjacency_matrices property of TimeSeriesCausalGraph where it was not looking at the variable names of the minimal graph, as expected, but the variables names of the full graph.
  • Removed the return_minimal argument from the adjacency_matrices property of TimeSeriesCausalGraph as this was never working.

0.4.5​

  • Fixed a bug in cai_causal_graph.time_series_causal_graph.TimeSeriesCausalGraph.from_adjacency_matrices for TimeSeriesCausalGraph where the method was not handling properly undirected edges in the adjacency matrices.
  • Added mypy-extensions dependency as "^1.0.0" and moved mypy back to dev dependency (setting it as "^1.8.0").

0.4.4​

  • Relaxed mypy dependency to "*".

0.4.3​

  • Added the replace_edge method to the CausalGraph class, which allows you to replace an existing edge with a new one.
  • Added the get_nodes_for_variable_name method to the TimeSeriesCausalGraph class.
  • Added the is_source_node and is_sink_node methods to the Node, indicating whether the node is a source node or sink node respectively.
  • Changed the following methods in CausalGraph from static to class methods:
    • cai_causal_graph.causal_graph.CausalGraph.from_skeleton
    • cai_causal_graph.causal_graph.CausalGraph.from_networkx
    • cai_causal_graph.causal_graph.CausalGraph.from_gml_string
    • Note that cai_causal_graph.causal_graph.CausalGraph.from_dict and cai_causal_graph.causal_graph.CausalGraph.from_adjacency_matrix were already class methods. Now all the from_ methods are consistent. This change is transparent to the user.
    • This allowed us to remove them from TimeSeriesCausalGraph. This change is transparent to the user.
  • Updated Skeleton such that its nodes match the class type of its graph.
    • Added ability to pass graph_class to the following methods in Skeleton such that the node classes will match accordingly when the new Skeleton is constructed:
      • cai_causal_graph.causal_graph.Skeleton.from_dict
      • cai_causal_graph.causal_graph.Skeleton.from_adjacency_matrix
      • cai_causal_graph.causal_graph.Skeleton.from_networkx
      • cai_causal_graph.causal_graph.Skeleton.from_gml_string
      • Note that cai_causal_graph.causal_graph.Skeleton.from_dict was already a class method but the other three were also changed to class methods for consistency. This change is transparent to the user.
  • Fixed a bug where the NodeVariableType was not synced between the CausalGraph and Skeleton.
  • Changed cai_causal_graph.time_series_causal_graph.TimeSeriesCausalGraph.from_adjacency_matrices from a static method to a class method to align with other from_ methods.
  • Updated numpy dependency from "^1.18.0" to "^1.20.0" and allow different numpy versions in the poetry.lock for python 3.8 and 3.9 - 3.12. This is to allow specific versions for 3.8 and the others as there is no version of numpy that supports them all.
  • Upgraded poetry version from 1.4.2 to 1.7.1 in the GitHub workflows.
  • Moved mypy from a dev dependency to a dependency as it is now used in source code.
  • Updated actions in GitHub workflows to transition from Node 16 to Node 20.

0.4.2​

  • Added the validate flag to add_edge, add_edges_from, add_edges_from_paths, add_edge_by_pair, cai_causal_graph.causal_graph.CausalGraph.from_dict, cai_causal_graph.time_series_causal_graph.TimeSeriesCausalGraph.add_edge, and add_time_edge which, if set to False, will disable validation checks. Currently, this disables cyclicity checks when adding edges. Default is True, making the change backwards compatible. There is no guarantees about the behavior of the resulting graph if this is disabled specifically to introduce cycles. This should only be used to speed up this method in situations where it is known the resulting graph is still valid, for example when copying a graph.

0.4.0​

0.3.14​

  • Fixed a bug in cai_causal_graph.identify_utils.identify_confounders where an empty confounding set would be returned in the edge case where all causal paths from the true confounders to node_1 were blocked by ancestors of node_2, or vice versa. This comes at a slight performance cost.

0.3.13​

  • Improved the get_topological_order method in TimeSeriesCausalGraph to improve performance. Added a new keyword argument respect_time_ordering to allow the user to specify whether the topological order must respect the time ordering of the nodes. If respect_time_ordering=True, the topological order will respect the time ordering, otherwise it may not. For example, if the graph is 'Y lag(n=1)' -> 'Y' <- 'X', then ['X', 'Y lag(n=1)', 'Y'] and ['Y lag(n=1)', 'X', 'Y'] are both valid topological orders. However, only the second one would respect time ordering. If both return_all and respect_time_ordering are True, then only all topological orders that respect time are returned, not all valid topological orders. The default is respect_time_ordering=True, matching previous behavior.

0.3.12​

  • Improved efficiency of cai_causal_graph.identify_utils.identify_confounders by performing all operations directly using networkx, removing the need to copy graphs and improving recursive logic, such that only the minimal confounders of the specified nodes are calculated (rather than recursively calculating minimal confounders of each parent). This results in significant speedups (in the order of hundreds of times in some cases).

0.3.11​

0.3.10​

0.3.9​

0.3.8​

0.3.7​

  • Fixed a bug in cai_causal_graph.time_series_causal_graph.TimeSeriesCausalGraph.from_adjacency_matrices for TimeSeriesCausalGraph where the method was not adding floating nodes to the graph. Now any floating nodes at time lag 0 will be added.

0.3.6​

0.3.5​

  • Added the cai_causal_graph.identify_utils.identify_markov_boundary utility function, which allows you to identify the Markov boundary of a node in a CausalGraph or in a Skeleton.
  • Added get_neighbor_nodes and get_neighbors methods to CausalGraph and Skeleton. get_neighbor_nodes returns the nodes neighboring the specified node while get_neighbors returns the identifiers of the neighboring nodes. Note: For a CausalGraph, it does not matter what the edge type is, as long as there is an edge between the specified node and another node, that other node is considered its neighbor.

0.3.4​

  • Extended documentation to provide further information regarding the types of mixed graphs that can be defined in a CausalGraph.

0.3.3​

  • Fixed a bug in cai_causal_graph.identify_utils.identify_instruments and cai_causal_graph.identify_utils.identify_mediators, where an unclear error was raised if the source node was a descendant of the destination node. Instead, these methods now return an empty list in that case.
  • Extended the quickstart documentation to describe how to set the variable_type when adding a Node / TimeSeriesNode to a CausalGraph / TimeSeriesCausalGraph, respectively.

0.3.2​

  • Improved documentation.

0.3.1​

0.3.0​

NOTE: Backwards compatibility warning! The definition of the EdgeConstraint enumeration has changed.

  • Updated the EdgeConstraint enumeration to simplify the enumeration members and exposed it at the root level, so it can be imported as from cai_causal_graph import EdgeConstraint. EdgeConstraint is not used by the cai-causal-graph package but any packages that rely on its definition, must be updated to reflect the new members.

0.2.17​

  • Fixed the docstrings of cai_causal_graph.time_series_causal_graph.TimeSeriesCausalGraph.from_adjacency_matrices, such that the examples render properly.

0.2.16​

  • Fixed a bug where cai_causal_graph.utils.get_variable_name_and_lag would not match variable names with non-alphanumeric characters, and would not match variable names with the string lag or future in them.

0.2.15​

  • Improved performance of checking for cycles when adding edges by avoiding repeated checks.

0.2.14​

  • Modified cai_causal_graph.time_series_causal_graph.TimeSeriesCausalGraph.add_edge in TimeSeriesCausalGraph to always order source and destination nodes by time_lag even when the edge is not directed. For example, add_edge('c', 'a (lag=1)', edge_type=EdgeType.UNDIRECTED_EDGE) will add an undirected edge a (lag=1) -- c instead of c -- a (lag=1). This is done just for convenience and to avoid confusion.

0.2.13​

  • Improved documentation.

0.2.12​

0.2.11​

0.2.10​

0.2.9​

  • Added support for python version 3.12.

0.2.8​

  • Changed the order of the documentation in the sidebar to ensure Quickstart is at the top.

0.2.7​

  • Improved internal logic for how an Edge is instantiated from a dictionary.

0.2.6​

  • Improved the copy method in CausalGraph such that it is more general and preserves the subclass type. As such, the .copy method was removed from the TimeSeriesCausalGraph class.
  • Extended equality methods for the Skeleton, CausalGraph, and TimeSeriesCausalGraph classes. A new keyword parameter deep has been added. If deep=True, deep equality checks are also done on all nodes and edges in the graphs. To call you must do graph_1.__eq__(graph_2, deep=True) as graph_1 == graph_2 still matches previous behavior.

0.2.5​

  • Added the cai_causal_graph.identify_utils.identify_confounders utility function, which allows you to identify a list of confounders between two nodes in a CausalGraph.
  • Added the cai_causal_graph.identify_utils.identify_instruments utility function, which allows you to identify a list of instrumental variables between two nodes in a CausalGraph.
  • Added the cai_causal_graph.identify_utils.identify_mediators utility function, which allows you to identify a list of mediators between two nodes in a CausalGraph.

0.2.4​

  • Fixed formatting in the documentation.

0.2.3​

  • Added the deserialization method from_dict to the following classes: Node, TimeSeriesNode, and Edge.
  • Added the serialization method to_dict to TimeSeriesNode. Node and Edge already had it.
  • Changed behavior of add_node method in TimeSeriesCausalGraph such that when both identifier and (time_lag, variable_name) are provided. Now, if all are provided, the method will raise an error only if identifier is not equal to get_name_with_lag(time_lag, variable_name), that is, the correct name.
  • Extended equality methods for the Node, Edge and TimeSeriesNode classes. A new keyword parameter deep has been added. If deep=True, additional class attributes are also checked; see the docstrings for additional information. To call you must do node_1.__eq__(node_2, deep=True) as node_1 == node_2 still matches previous behavior.
  • Added edge_type property to the Edge class.

0.2.2​

  • Fixed typo in the quickstart documentation.

0.2.1​

0.2.0​

  • Added the TimeSeriesCausalGraph class to represent a time series causal graph. TimeSeriesCausalGraph is aware of the time relationships between the nodes in the graph while CausalGraph is not. Moreover, the TimeSeriesCausalGraph class has three new representations:
    • The minimal graph, which can be obtained via the get_minimal_graph method, defines the graph with the minimal number of nodes and edges that is required to capture all the information encoded in the original graph. This is because a time series causal graph may contain a lot of repetitive information. For example, if the original graph is x(t-2) -> x(t-1) -> x(t), then the minimal graph would be x(t-1) -> x(t). In other words, it is a graph that has no edges whose destination is not time 0.
    • The summary graph, which can be obtained via the get_summary_graph method, defines the graph collapsed in time so there is a single node per variable. For example, if the original graph is z(t-2) -> x(t) <- y(t) <- y(t-1) then the summary group would be z -> x <- y. Note, it is possible to have cycles in the summary graph. For example, a graph with edges y(t-1) -> x(t) and x(t-1) -> y(t) would have a summary graph of x <-> y.
    • The extended graph, which can be obtained via the extend_graph method, defines the graph obtained by extending backward and forward in time via the arguments backward_steps and forward_steps, respectively. This graph may contain lots of redundant information. For example, if the original graph is x(t-1) -> x(t) and backward_steps=2 and forward_steps=1, then the extended graph would be x(t-2) -> x(t-1) -> x(t) -> x(t+1).
  • The cai_causal_graph.time_series_causal_graph.TimeSeriesCausalGraph.from_adjacency_matrices method was added to instantiate an instance of TimeSeriesCausalGraph from a dictionary of adjacency matrices where the keys are the time lags.
  • Added the TimeSeriesNode class to extend the Node class to represent time information on the node. The following properties were added:
    • variable_name: The variable name of the time series node. For example, if the identifier of the time series node is 'X1 lag(n=1)', i.e., it is a lagged version of the variable 'X1', then variable_name would be 'X1'.
    • time_lag: The time lag of the time series node. For example, if the identifier of the time series node is 'X1 lag(n=1)', i.e., it is a lagged version of the variable 'X1', then time_lag would be -1.
  • Renamed EdgeTypeEnum to EdgeType.

0.1.3​

  • Fixed a syntax error in the docstring for the get_bidirected_edges method that was preventing the reference docs from being built.

0.1.2​

  • Improved README links so images appear on PyPI.
  • Upgraded poetry version from 1.2.2 to 1.4.2 in the GitHub workflows.

0.1.1​

  • Added security linting checks of source code using bandit.
  • Improved documentation packaging and publishing.

0.1.0​