The IBM Decision Optimization CPLEX Modeling for Python

## Project description

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 need IBM ILOG CPLEX Optimization Studio to solve the models.

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 V12.8 or later is installed on your machine.

This library is numpy friendly.

## Installation

`pip install docplex`

## Changelog

### Changed in 2.18.200:

- Latest supported CPLEX Optimization Studio is now 20.1

### Changed in 2.17.196 (2020.11.#2):

- In
`docplex.mp`: - Fixed a bug with pickling: edition of a constraint in a pickled model raised an error
- Fixed a bug with pickling: models with piecewise-linear constraints could not be pickled

- In
- In
`docplex.cp`: - Add environment variable DOCPLEX_CP_CONTEXT to modify configuration
- Add new module check_list that print a report on execution environment
- Remove DOcloud from documentation (including code)
- Rework customization of configuration and better support of default directory
- Add new configuration parameter model.sort_names to drive sort of variable declarations in CPO file format.
- Fix a problem that may crash Python in case of abort_search with local solve.

- In

### Changed in 2.16.196 (2020.11):

- In
`docplex.mp`: - add Model.add_quadratic_constraints() to add a batch of quadratic constraints
- add Model.populate_solution_pool() for a native support of solurtion pools
- support of CPLEX 20.1
- compatible with Python 3.8 (only with CPLEX 20.1)
- enable changing absolute and relative tolerances for multi-objectives
- Optimization of Model.if_then: when condition is of the form b==1 (or 0), no additional boolean variable is generated
- For solving, docplex.mp now uses the cplex module if it has been installed. If not, docplex.mp checks for the latest installed version of CPLEX Optimization Studio (COS) (using the CPLEX_STUDIO_DIRXXX environment variables) and use the cplex if a COS is found, unless the configuration of the engine states otherwise.

- In
- In
`docplex.cp`: - Add pngfile= parameter to visu.show() method to store in a PNG file instead of display on screen.
- Make parameters and solver infos also available in refine_conflict solution.
- Add a IntegerDomain class used to represent domain of integer variables, with a specific __str__ method
- Add new functions ceil(), floor(), trunc(), round() and sgn()
- Remove all warnings generated by Python 38
- Optimize creation of add expressions with CumulExpr and zero
- Implement configurable factorization of common model expressions when generating CPO format
- Add checking of scal_prod() array sizes at modeling time
- Add strict_lexicographic() and checking of strict_lexicographic() and lexicographic() array sizes at modeling time.
- Add failure explanation as new method explain_failure() allowing to log failure tags or get details on one or several failures.
- Enhance management of local solve sub-process timeout with detailed error and configurable timeout delay
- For solving, docplex.cp now uses the cpoptimizer executable if it has been installed. If not, docplex.cp checks for the latest installed version of CPLEX Optimization Studio (COS) (using the CPLEX_STUDIO_DIRXXX environment variables) and use the cpoptimizer if a COS is found, unless the configuration of the engine states otherwise.
- Support last optimal solution in search/next sequence
- Support of solver parameters in all next() solutions
- Add solver version in process info attached to a run result

- In

### Changed in 2.15.194 (2020.07):

- In
`docplex.mp`: - add Model.quadratic_dual_slacks()
- Fixed a bug in multi-objectives: objectives were incorrectly rounded
- Fixed a bug in Model.report(): multiple objective values were not displayed()

- In
- In
`docplex.cp`: - Add conflict in CPO format in refine conflict result
- Fix problem when parsing KPIs section of a CPO model
- Add method add_constraint() to model for compatibility with docplex.mp
- Comment method get_fail_status() of SolveResult as deprecated.
- Fix problem of wrong import of deque in collections.abc

- In

### Changed in 2.14.186 (2020.05):

- Updated tracking events in Watson studio notebooks.
- In
`docplex.mp`: - Model.solve() will not use solve on cloud unless agent is specifically set to ‘docloud`.

- In

### Changed in 2.13.184 (2020.03):

- Removed dependency to the docloud package. Now you need to explicitely install the package using pip install docloud to use DOcplexcloud.
- In
`docplex.mp`: - added Model.export_as_mps_string(), Model.export_as_sav_string()
- fixed a bug with dettime_limit: solving with a deterministic time limit
- was mis-interpreted as a solve failure, returning None.

- fixed bug on cplexcloud solve: number of nodes processed was always zero.
- repeated solves incorrectly restarted from start of search, now start from where the last solve stopped.
- added keyword argument ‘time_limit’ to Model.solve() to set a temporary time limit.
- added new method SolveSolution.is_valid_solution()
- fixed a bug in ModelReader: ranged constraints bounds were inverted when reading from SAV or MPS.
- fixed a bug in Model.set_lex_multiobj(): arguments abstols, reltols were ignored.
- added proper type-checking for Model.add_indicator_constraints()
- added docplex.mp.check_list/py to check local installation.

- In
- In
`docplex.cp`: - Enable reading of #line directives when parsing a CPO file
- Remove parameter LogSearchTags from public parameters
- Fix a minor problem concerning compilation of KPI expressions in CPO format

- In

### Changed in 2.12.182 (2019.12):

- In
`docplex.mp`: - Added a LinearRelaxer class to make a linearized copy of a MIP model (if possible). see class docplex.mp.relax_linear.LinearRelaxer
- Conflict refiner default behavior is now identical to CPLEX interactive (the new behavior is much faster).
- Bug fixed: expressions of the form k*x did not notify constraints when modified.
- Fixed: message “ignored keyword argument” was incorrectly printed when setting cts_by_name=True in model constructor.

- In

### Changed in 2.11.176 (2019.11):

- Added support for CPLEX 12.10
- In
`docplex.mp`: - Logical expressions, binary variables, and constraints can now be freely nested with logical operators.
- Fixed a print of ‘CPLEX Error 1217’ in log for multi-objective problems.
- Fixed a bug when setting log_output to a file name: file was created, but empty.

- In

### Changed in 2.10.155 (2019.08):

- Fixed bug in logical_and() when result var is set to 0.

### Changed in 2.10.154 (2019.07):

- Fixed TypeError occuring in python 3.7 in progressData initialization.

### Changed in 2.10.151 (2019.07):

- In
`docplex.mp`: - fixed a bug in ModelReader when reading SAV files with no names
- fixed a bug in mip starts, which prevented mip starts with piecewise functions to work properly.
- fixed bug on Model.add_indicators() using comprehensions (len() was called).
- Added support for the ‘!=’ (not equals) operator in expressions.
- Clarified four types of checker: on, off, numeric and full. Pass checker=<name> at model creation to specify which checker is used.
- fixed a bug in solution JSON encoder for nonconvex QP problems.
- Add direct support for lazy constraints, see Model.add_lazy_constraints()
- Add direct support for user cuts, see Model.add_user_cut_constraints()
- Get basis status of variables in LP problems, see Var.basis_status
- Read MIP start files (MST format)
- Allow to set the effort level for a MIP solution.
- Read basis status files (in BAS format)
- Read variable priority orders (in ORD format)
- fixed bug in functional KPIs, solution argument was not passed on.
- Enable constraint name dictionary at Model creation time: Model(cts_by_name=True)
- Multi-objective is now pickled correctly
- Multi-objective is now copied in Model.copy()
- Wrote full documentation on progress listeners
- Added Model.set_lp_start_basis() to provide an initial basis for LP problems.

- In
- In
`docplex.cp`: - When trying to access a solution member that does not exists, an exception is thrown instead of returning None.
- Add a new map_solution function that replace in a Python object all model expressions by their value in a solve result.
- In CPO parser, fix a problem reading #line statements in startingPoint section.
- In CPO parser, skip experimental section ‘expressions’ in ‘startingPoint’ section.
- Simplify writing of interval variable domains reduced to a single value.
- Adding a second objective function now raises an exception
- Add new experimental local solve with a shared library.
- Enable iterators to specify the domain of an integer variable
- Add global methods get_version_info() and get_solver_verion() in docplex.cp.solver.
- By default, generate CPO model without explicit format version.
- Add a method reset() on CpoParameters object.
- Modeling method allowed_assignments() and forbidden_assignments() can now accept an empty list of tuples.
- On CpoModelSolution object, add a function map_solution() thar replace variables by their value in a python object.
- Add parser for LP models
- Add possibility to import CPO, MZN and LP models in gzip and zip format.
- Enhance management of unexpected errors thrown by cpoptimizer.exe

- In

### Changed in 2.9.141 (2019.03):

- In
`docplex.mp`: - Removed links to rawgit.com as this service is going end of life.
- Model.solve_lexicographic() is deprecated. This method should be used to perform lexicographic solve with COS 12.8, but with COS 12.9, Model.set_multi_objective() should be used for solving problems with multiple objectives.

- In
- In
`docplex.cp`: - Add KPIs supported by CPO Solver 12.9
- Update CPO parser to read KPIs section for format 12.9
- Add new examples with KPIs.

- In

### Changed in 2.8.125 (2018.10):

- 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
- 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 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.

- In

### Changed in 2.7.113 (2018.07):

- 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
- 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

- In

### Changed in 2.6.94 (2018.04):

- 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.

- In

### Changed in 2.5.92 (2018.03):

`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.

- option

- 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.

- In

### Changed in 2.4.61 (2017.11):

- 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.

- Both
- 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
- 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.

- In

### 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.

## Project details

## Release history Release notifications | RSS feed

## Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Filename, size | File type | Python version | Upload date | Hashes |
---|---|---|---|---|

Filename, size docplex-2.18.200.tar.gz (603.9 kB) | File type Source | Python version None | Upload date | Hashes View |