Adding binary (binary_big_M) decomposition type#50
Conversation
Updating error to note that absolute_value not valid for cvxpy, as neither is DCP-compliant Allowing binary_big_M for cvxpy with gurobi
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #50 +/- ##
==========================================
+ Coverage 94.38% 94.63% +0.24%
==========================================
Files 10 10
Lines 1603 1659 +56
==========================================
+ Hits 1513 1570 +57
+ Misses 90 89 -1 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
fletchapin
left a comment
fletchapin
left a comment
There was a problem hiding this comment.
Looks great to me! @arao53 might have comments since he's more familiar with MILP
| .. code-block:: python | ||
|
|
||
| from electric_emission_cost import costs | ||
| from eeco import costs |
| - "absolute_value": Uses max(x, 0) constraints. Creates nonlinear problem | ||
| for Pyomo due to abs() constraint. | ||
| - "binary_big_M": Uses binary indicator with Big-M constraints. | ||
| Creates a MILP (mixed-integer linear program) for Pyomo. |
There was a problem hiding this comment.
Should this be "for Pyomo or CVXPY" since binary_big_M is supported for both?
| - "absolute_value": Uses max(x, 0) constraints. Creates nonlinear problem | ||
| for Pyomo due to abs() constraint. | ||
| - "binary_big_M": Uses binary indicator with Big-M constraints. | ||
| Creates a MILP (mixed-integer linear program) for Pyomo. |
There was a problem hiding this comment.
Should this be "for Pyomo or CVXPY" since binary_big_M is supported for both?
|
(Still working on this PR, it can be merged AFTER #51 if that's approved) |
- check if charge_array is cp.Expression and access charge_array.size rather than len(charge_array) if so -Don't check for consumption_max when building in lines 641-644 if cp.Expression - Update CVXPY branch gate conditions lines 694-711. When demand is below the tier limit, consumption_data - limit < 0, so cp.max(demand_charged) < 0, and cp.maximum(... -prev_demand_cost, 0) = 0. - in calculate_cost lines 1291 - 1298, get n by which to divide demand consumption estimate based on .size if cp.Expression
… tuple from decompose_consumption. Previously was using it in the "model" entry for pyomo, but constraints weren't properly getting added. Now there will be a separate "_" in the return tuple from calculate_cost and calculate_itemized cost. Alternatively could try to group both items under "model_or_constraints" to not change compatibility with previous uses of eeco? Or add backwards compatibility?
| ) | ||
|
|
||
| return results_dict, model | ||
| return results_dict, model, cvxpy_constraints |
There was a problem hiding this comment.
@fletchapin when using the binary_big_M decomposition method for cvxpy, it has to create 3 new constraints in _decompose_binary_cvx that get passed back up to calculate_cost. Do you think it's better to:
- Return cvxpy_constraints as a separate item in this tuple (current implementation here)
- Return results_dict, model_objects where "model_objects" could either be a pyomo model or a cvxpy constraint list?
Option 2 would prevent this update breaking older code that calls calculate_cost, but will take a little more work to change some of the checks for model being "None"
There was a problem hiding this comment.
Updated to option 2 which should avoid any issues with existing calls to calculate_cost!
There was a problem hiding this comment.
@arao53 Interested in any thoughts you have on the MILP functions here _decompose_binary_cvx and _decompose_binary_pyo. No rush on this since I'll make any further updates when I get back mid-April
Adding codecov for new lines Allowing big_m as optional argument for decomposition_type
Removing outdated binary_variable references
…ict from prior data. Helper for use in other repositories that need to pass in a prev_demand_dict to calculate_cost
…cts instead of model. (model_objects can be Pyomo model m OR a list of cvxpy constraints) update calculate_itemized_cost to preserve the model_objects returned by calculate_cost
3 participants
Updating error to note that absolute_value not valid for cvxpy, as neither is DCP-compliant
Allowing binary_big_M for cvxpy with gurobi
Addresses #45 and #46
Also added a helper function get_prev_demand_dict that can be called by external codebases by passing in charge data and a charge dict
Pull request recommendations:
Thanks for contributing!