metexplore / miom Goto Github PK

.get_solver_status() in BaseModel should return information about the total num of seconds for the search, the result (infeasible, optimal, etc) and the objective score

Add a method in _Variables to obtain both the raw indicator variables (2 per each positive weighted and reversible reaction, 1 for each negative weighted reaction or positive non-reversible reaction), and the reaction activity (1 if the reaction is active in any direction, 0 otherwise). Return these values with the method get_values() instead of the raw indicators, as they are hard to interpret for users.

As @pablormier suggested, we should think about using a common status for both PICOS and Python-MIP.

Possible solver status for PICOS: https://picos-api.gitlab.io/picos/api/picos.modeling.solution.html

Possible solver status for Python-Mip: https://python-mip.readthedocs.io/en/latest/classes.html (OptimizationStatus class)

Time returned with get_solver_status() does not match the values shown by Python-MIP (tested with CBC)

Add an integer cut to the current problem, taking by default the current assignations for the integer vars (if the problem was solved) or taking an optional vector with the values to cut from the problem.

after copy the method subset_selection has no effect

Allow warm start for both PICOS and Python-MIP:

• PICOS: https://picos-api.gitlab.io/picos/api/picos.modeling.options.html (hotstart option)
• Python-MIP: https://docs.python-mip.com/en/latest/custom.html#providing-initial-feasible-solutions

Implement a strategy to auto-select the right backend and solver. PICOS automatically decides the right solver based on the problem formulation. Switch to PICOS+GUROBI/CPLEX if they are available

model.copy() should return a new instance with a copy of the constraints and variables. This will allow to reuse partial instantiated models (e.g. not re-adding the steady state constraints)

This method would make it even easier to select the active subnetwork after a FBA/subset selection problem. Use flux thresholds if no indicator variables are present.

Adding independent constraints has a huge memory impact. Change to vectorized format to see if it improves the problem building step

Issues with cffi lib with Google Colab, make Mip lib optional and PICOS the default lib for modeling

Create indicators for all reactions even when the weights are 0. This will allow to add constraints on every reaction (e.g for maximization/minimization of differences).