opvious 0.12.10rc2

Opvious Python SDK

Opvious Python SDK

An optimization SDK for solving linear, mixed-integer, and quadratic models

Highlights

Declarative modeling API

• Extensive static validations
• Exportable to LaTeX
• Extensible support for high-level patterns (activation variables, masks, ...)

import opvious.modeling as om

class BinPacking(om.Model):
  items = om.Dimension()  # All items to bin
  weight = om.Parameter.non_negative(items)  # Weight per item
  bins = om.interval(1, om.size(items), name="B")  # Available bins
  max_weight = om.Parameter.non_negative()  # Maximum weight for each bin

  assigned = om.Variable.indicator(bins, items)  # Bin to item assignment
  used = om.fragments.ActivationIndicator(assigned, projection=1)  # 1 if a bin is used

  @om.constraint
  def each_item_is_assigned_once(self):
    for i in self.items:
      yield om.total(self.assigned(b, i) for b in self.bins) == 1

  @om.constraint
  def bin_weights_are_below_max(self):
    for b in self.bins:
      bin_weight = om.total(self.weight(i) * self.assigned(b, i) for i in self.items)
      yield bin_weight <= self.max_weight()

  @om.objective
  def minimize_bins_used(self):
    return om.total(self.used(b) for b in self.bins)

Auto-generated specification:

Transparent remote solves

• No local solver installation required
• Real-time progress notifications
• Seamless data import/export via native support for pandas
• Flexible multi-objective support: weighted sums, epsilon constraints, ...
• Built-in debugging capabilities: relaxations, fully annotated LP formatting, ...

import opvious

client = opvious.Client.from_environment()

response = await client.run_solve(
  specification=BinPacking().specification(),
  parameters={
    "weight": {"a": 10.5, "b": 22, "c": 48},
    "binMaxWeight": 50,
  },
)
solution = response.outputs.variable("assigned")  # Optimal assignment dataframe

Take a look at https://opvious.readthedocs.io for the full documentation or these notebooks to see the SDK in action.