The IBM Decision Optimization CPLEX Modeling for Python
Welcome to the IBM® Decision Optimization CPLEX Modeling for Python. Licensed under the Apache License v2.0.
With this library, you can quickly and easily add the power of optimization to your application. You can model your problems by using the Python API and solve them on the cloud with the IBM® Decision Optimization on Cloud service or on your computer with IBM® ILOG CPLEX Optimization Studio.
This library is composed of 2 modules:
- IBM® Decision Optimization CPLEX Optimizer Modeling for Python - with namespace docplex.mp
- IBM® Decision Optimization CP Optimizer Modeling for Python - with namespace docplex.cp
Solving with CPLEX locally requires that IBM® ILOG CPLEX Optimization Studio V12.7.1 or V12.8 is installed on your machine.
Solving with the IBM® Decision Optimization on Cloud service requires that you register for an account and get the API key.
This library is numpy friendly.
- pip install docplex
Changed in 2.8.125:
- Solving with solver agent ‘docloud’ is deprecated. Models are now preferably solved with local solver or the python source can be submitted to DOcplexcloud solve service. See https://ibm.biz/BdYhhK
- In docplex.mp:
- solve_lexicographic is being deprecated. In a future version, a new api will be available to support multi-objectives.
- In docplex.cp:
- Fix problem with boolean indicators in no_overlap(), always_constant() and always_equal().
- Allow model solution to be used directly as a starting point (ignores what is not integer or interval var).
- Add new examples with KPIs.
- Add methods domain_min(), domain_max(), domain_iterator() and domain_contains() on both CpoIntVar and CpoIntVarSolution.
- Default solver agent is now ‘local’ instead of ‘docloud’. All examples modified consequently.
Changed in 2.7.113:
- In docplex.mp:
- Multiplying a constant expression by a quadratic expression raised an exception. Now returns the product of the quadratic expression and the constant value.
- Model.solve_lexicographic() on cloud now send the previous pass solution as a MIP start (for MIP problems)
- The slack of quadratic constraints always returned zero. Now returns the correct value.
- Accessing the dual (or slack) of a constraint that is not added to the model returned zero; now it raises an exception. A constraint must belong to a model to return a valid dual (or slack) value
- Range constraints with infeasible domain (i.e. lb > ub) did not fail to solve. Now they raise a modeling exception.
- Multiplying two absolute value expressions raised an exception. Now fixed.
- When using tuples in variable dictionaries, the default name generation used to generate non-LP-compliant names, because of ( and ). Now the name generator formats the tuples with a “_” separator without parentheses.
- In docplex.cp:
- Split fzn stuff in a separate package docplex.cp.fzn
- Optimize construction of arrays in FZN parser
- Enhance FZN parser and save 30% time
Changed in 2.6.94:
- In docplex.cp:
- Allow CpoModel.add() to accept list of constraints.
- Fix a bug in the conversion of an array of boolean constants into CPO expression.
- Extend CpoModel method set_parameters() to accept a dictionary and/or optional list of updates using named arguments.
- Method CpoModel.set_parameters() now clone the CpoParameters object given in arguments.
- Add a new method CpoModel.add_parameters() that updates parameters associated to the model.
- Fix wrong source location (not in real model source) when CpoModel.add() is called from another docplex.cp method.
- When constraint auto-naming is on (in particular for refine_conflict(), searchPhases are no more included in the process.
- Parameters mean_UB and mean_LB are now optional in standard_deviation()
- CpoModel.add() checks that the added expression is limited to constraint, boolean, objective or search phase.
- Add documented functions slope_piecewise_linear() and coordinate__piecewise_linear() in modeler.py.
- Remove default configuration settings for parameters TimeLimit and Workers.
Changed in 2.5.92:
- docplex.cli gains new features:
- option --details will display solve details as they are published on DOcplexcloud.
- options --url and -key allow specification of credentials without using a config file.
- In docplex.cp:
- Fix problem with min() and max() that did not support optional key.
- Add a Flatzinc parser capable of reading Minizinc Challenge problems.
- Move expression dependencies analysis from model to compiler side.
- No more constraint to have a unique name for model expressions. Compiler reallocate private names when needed.
- Multiple variables or expressions with the same public name is now allowed.
- Replace method CpoModel.get_expression() by CpoModel.get_named_expressions_dict().
- Make SolverProgressPanelListener work properly with Python 2
- Solve is automatically set to start/next loop when SolverProgressPanelListener is used.
- In CpoModel, add a method that allows to substitute a function by another in the whole model.
- Overwrite method __bool__ to avoid accidental use of CPO expressions as Python booleans.
- Add special cases to search for the local CP Optimizer Interactive executable.
- Allow methods min(), max(), min_of() and max_of() to support variable number of arguments.
- Allow method all_diff() to support variable number of arguments.
- Context parameter ‘length_for_rename’ is deprecated. Only length_for_alias is used.
- Add a method add_var() in CpoModelSolution as a shortcut to add_integer_var_solution() and add_interval_var_solution()
- Overwrite method __contains__() in CpoModelSolution to easily verify that a solution to a given variable is in the solution.
- When called on a model, export_model() and get_cpo_string() disable all model optimization options.
Changed in 2.4.61:
- Both docplex.mp & docplex.cp:
- Support for CPLEX engines 12.8. Some features of docplex2.4 are available only with engines >= 12.8.
- Adding new ports (AIX, plinux).
- Examples are now available as Zeppelin notebooks.
- In docplex.mp:
- Express a linear problem as a scikit-learn transformer by providing a numpy, a pandas or scipy matrix.
- Logical constraints: constraint equivalence, if-then & rshift operator.
- Meta-constraints: allow the use of discrete linear constraints in expressions, using their truth value.
- Solve hook to add a method to be called at each intermediate solution.
- KPIS automatically published at each intermediate solution if running on docplexcloud python worker.
- Support for scipy coo & csr matrixes.
- Fixed a bug in Model.add_constraints() when passing a string instead of a list of strings.
- In docplex.cp:
- add new method run_seeds() to execute a model multiple times, available with local solver 12.8.
- add support of new solver infos ‘SearchStatus’ and ‘SearchStopCause’.
- In method docplex.cp.model.CpoModel.propagate(), add possibility to add an optional constraint to the model.
- add domain iterator in integer variables and integer variables solutions, allowing to get domain as a list of individual integers.
- add possibility to identify some model variables as KPIs of the model.
- add abort_search() method on solver (not supported everywhere)
- Rework code generation to enhance performances and remove unused variables that was pointed by removed expressions.
- add possibility to add one or more CpoSolverListener to put some callback functions when solve is started, ended, or when a solution is found. Implementation is provided in new python module docplex.cp.solver.solver_listener that also contains sample listeners SolverProgressPanelListener and AutoStopListener.
- Using parameter context.solver.solve_with_start_next, enable solve() method to execute a start/next loop instead of standard solve. This enables, for optimization problems, usage of SolveListeners with a greater progress accuracy.
- Completely remove deprecated ‘angel’ to identify local solver.
- Deprecate usage of methods minimize() and maximize() on docplex.cp.CpoModel.
- Add methods get_objective_bounds() and get_objective_gaps() in solution objects.
Changed in 2.3.44 (2017.09):
- Module docplex.cp.model.solver_angel.py has been renamed solver_local.py. A shadow copy with previous name still exist to preserve ascending compatibility. Module docplex.cp.model.config.py is modified to refer this new module.
- Class docplex.cp.model.solver_local.SolverAngel has been renamed SolverLocal. A shadow copy with previous name still exist to preserve ascending compatibility.
- Class docplex.cp.model.solver_local.AngelException has been renamed LocalSolverException. A shadow copy with previous name still exist to preserve ascending compatibility.
- Functions logical_and() and logical_or() are able to accept a list of model boolean expressions.
- Fix defect on allowed_assignments() and forbiden_assignments() that was wrongly converting list of tupes into tuple_set.
- Update all examples to add comments and split them in sections data / prepare / model / solve
- Add new sched_RCPSPMM_json.py example that reads data from JSON file instead of raw data file.
- Rename all visu examples with more explicit names.
- Remove the object class CpoTupleSet. Tuple sets can be constructed only by calling tuple_set() method, or more simply by passing directly a Python iterable of iterables when a tupleset is required (in expressions allowed_assignments() and forbidden_assignments)
- Allow logical_and() and logical_or() to accept a list of boolean expressions.
- Add overloading of builtin functions all() and any() as other form of logical_and() and logical_or().
- In no_overlap() and state_function(), transition matrix can be passed directly as a Python iterable of iterables of integers,
- Editable transition matrix, created with a size only, is deprecated. However it is still available for ascending compatibility.
- Add conditional() modeling function
- Parameter ‘AutomaticReplay’ is deprecated.
- Add get_search_status() and get_stop_cause() on object CpoSolveResult, available for solver COS12.8
- Improved performance of Var.reduced_cost() in docplex.mp.
Changed in 2.2.34 (2017.07):
- Methods docplex.cp.model.export_model() and docplex.cp.model.import_model() have been added to respectively generate or parse a model in CPO format.
- Methods docplex.cp.model.minimize() and docplex.cp.model.maximize() have been added to directly indicate an objective at model level.
- Notebook example scheduling_tuto.ipynb contains an extensive tutorial to solve scheduling problems with CP.
- Modeling method sum() now supports sum of cumul expressions.
- Methods docplex.cp.model.start_search() allows to start a new search sequence directly from the model object.
- When setting context.solver.auto_publish is set, and using the CPLEX engine, KPIs and current objective are automatically published when the script is run on DOcplexcloud Python worker.
- When setting context.solver.auto_publish is set, and using the CP engine, current objective is automatically published when the script is run on DOcplexcloud Python worker.
- docplex.util.environment.Environment.set_stop_callback and docplex.util.environment.Environment.get_stop_callback are added so that you can add a callback when the DOcplexcloud job is aborted.
Changed in 2.1.28:
- New methods Model.logical_or() and Model.logical_and() handle logical operations on binary variables.
- DOcplex now supports CPLEX 12.7.1 and Benders decomposition. Set annotations on constraints and variables using the benders_annotation property and use the proper CPLEX parameters governing Benders decomposition.
- CPLEX tutorials: in the documentation and as notebooks in the examples.
- Fixed a bug in docplex.mp.solution.SolveSolution.display() and in docplex.mp.solution.Model.report_kpi() when using unicode variable names.
- There’s now a simple command line interface for DOcplexcloud. It can be run in a terminal. python -m docplex.cli help for more info. That command line reads your DOcplexcloud credentials in your cplex_config.py file. It allows you to submit, list, delete jobs on DOcplexcloud. The cli is available in notebooks too, using the %docplex_cli magics. %docplex_cli help for some help. In a notebook, credentials can be passed using %docplex_url and %docplex_key magics.
- Removing constraints in 1 call
- Bug fixes when editing an existing model.
- Bug fix in the relaxation mechanism when using docplexcloud.
Changed in 2.0.15:
- Piecewise linear (PWL) functions are now supported. An API is now available on docplex.mp.model to create PWL functions and to create constraints using these PWL functions. PWL functions may be defined with breakpoints (default API) or by using slopes. Some simple arithmetic is also available to build new PWL functions by adding, subtracting, or scaling existing PWL functions.
- DOcplex has undergone a significant overhaul effort that has resulted in an average of 30-50% improvement of modeling run-time performance. All parts of the API benefit from the performance improvements: creation of variables and constraints, removal of constraints, computation of sums of variables, and so on.
- Constraints are now fully editable: the expressions of a constraint can be modified. Similarly, the objective expression can also be modified. This allows for complex workflows in which the model is modified after a solve and then solved again.
- docplex is now available on Anaconda cloud and can be installed via the conda installation packager. See the IBM Anaconda home CPLEX Community Edition for Python is also provided on Anaconda Cloud to get free local solving capabilities with limitations.
- Support of ~/.docplexrc configuration files for docplex.mp.context.Context is now dropped. This feature has been deprecated since 1.0.0.
- Known incompatibility: class docplex.mp.model.AbstractModel moved to docplex.mp.absmodel.AbstractModel. Samples using this class have been updated.
Changed in 1.0.630:
- Added support for CPLEX 12.7 and Python 3.5.
- Upgraded the DOcplexcloud client to version 1.0.202.
- Module docplex.mp.advmodel is now officially supported. This module provides support for efficient, specialized aggregator methods for large models.
- When solving on DOcplexcloud, proxies can now be specified with the context.solver.docloud.proxies property.
- When two constraints are defined with the same name, issue a warning instead of a fatal exception. The last constraint defined will take over the first one in the name directory.
- Fix ValueError when passing a pandas DataFrame as variable keys (using DataFrame indexes).
- Solution.get_values() returns a collection of variable values in one call.
- docplex.mp.model no longer imports docloud.status. Any status previously initialized as JobSolveStatus.UNKNOWN is now initialized as None.
- Minor improvements to notebooks and examples.