feat: infer input dimensions (#162)

Author: mtth

pyproject.toml

@@ -14,7 +14,7 @@ requires = ['poetry-core']
 build-backend = 'poetry.core.masonry.api'
 
 [tool.poetry]
-version = '0.23.5rc2'
+version = '0.24.0rc1'
 packages = [{include = 'opvious', from = 'src'}]
 
 [tool.poetry.dependencies]

src/opvious/data/solves.py

@@ -19,6 +19,7 @@
     UnboundedOutcome,
 )
 from .outlines import Label, ObjectiveSense, ProblemOutline, SourceBinding
+from .tensors import KeyItem
 
 
 @dataclasses.dataclass(frozen=True)
@@ -177,37 +178,58 @@ class SolveInputs:
     problem_outline: ProblemOutline
     """Target model metadata"""
 
-    raw_parameters: list[Json] = dataclasses.field(repr=False)
+    raw_parameters: Sequence[Json] = dataclasses.field(repr=False)
     """All parameters in raw format"""
 
-    raw_dimensions: list[Json] | None = dataclasses.field(repr=False)
+    raw_dimensions: Sequence[Json] | None = dataclasses.field(repr=False)
     """All dimensions in raw format"""
 
-    def parameter(self, label: Label, coerce: bool = True) -> pd.DataFrame:
-        """Returns the parameter for a given label as a pandas DataFrame
+    def parameter(
+        self,
+        label: Label,
+        coerce: bool = True,
+        index: pd.Index | Sequence[KeyItem] | None = None,
+    ) -> pd.DataFrame:
+        """Returns the parameter for a given label as a pandas dataframe
 
         The returned dataframe has a `value` column with the parameter's values
         (0 values may be omitted).
 
         Args:
             label: Parameter label to retrieve
             coerce: Round integral parameters
+            index: Returned dataframe index
         """
         for param in self.raw_parameters:
             if param["label"] == label:
                 outline = self.problem_outline.parameters[label]
-                return _entries_dataframe(
-                    param["entries"],
+                return _tensor_json_dataframe(
+                    param,
                     outline.bindings,
+                    index=index,
                     round_values=coerce and outline.is_integral,
                 )
         raise Exception(f"Unknown parameter: {label}")
 
     def dimension(self, label: Label) -> pd.Index:
         """Returns the dimension for a given label as a pandas Index"""
-        for dim in self.raw_dimensions or []:
-            if dim["label"] == label:
-                return pd.Index(dim["items"])
+        if self.raw_dimensions is not None:
+            for dim in self.raw_dimensions:
+                if dim["label"] == label:
+                    return pd.Index(dim["items"], name=label)
+        else:
+            items = set()
+            has_binding = False
+            for param in self.raw_parameters:
+                outline = self.problem_outline.parameters[param["label"]]
+                for i, binding in enumerate(outline.bindings):
+                    if binding.dimension_label != label:
+                        continue
+                    has_binding = True
+                    for entry in param["entries"]:
+                        items.add(entry["key"][i])
+            if has_binding:
+                return pd.Index(items, name=label).sort_values()
         raise Exception(f"Unknown dimension: {label}")
 
 
@@ -218,13 +240,18 @@ class SolveOutputs:
     problem_outline: ProblemOutline
     """Solved model metadata"""
 
-    raw_variables: list[Json] = dataclasses.field(repr=False)
+    raw_variables: Sequence[Json] = dataclasses.field(repr=False)
     """All variables in raw format"""
 
-    raw_constraints: list[Json] = dataclasses.field(repr=False)
+    raw_constraints: Sequence[Json] = dataclasses.field(repr=False)
     """All constraints in raw format"""
 
-    def variable(self, label: Label, coerce: bool = True) -> pd.DataFrame:
+    def variable(
+        self,
+        label: Label,
+        coerce: bool = True,
+        index: pd.Index | Sequence[KeyItem] | None = None,
+    ) -> pd.DataFrame:
         """Returns variable results for a given label
 
         The returned dataframe always has a `value` column with the variable's
@@ -234,14 +261,16 @@ def variable(self, label: Label, coerce: bool = True) -> pd.DataFrame:
         Args:
             label: Variable label to retrieve
             coerce: Round integral variables
+            index: Returned dataframe index
         """
         for res in self.raw_variables:
             if res["label"] == label:
                 outline = self.problem_outline.variables[label]
-                return _entries_dataframe(
-                    res["entries"],
+                return _tensor_json_dataframe(
+                    res,
                     outline.bindings,
                     dual_value_name="reduced_cost",
+                    index=index,
                     round_values=coerce and outline.is_integral,
                 )
         raise Exception(f"Unknown variable {label}")
@@ -255,29 +284,35 @@ def constraint(self, label: Label) -> pd.DataFrame:
         """
         for res in self.raw_constraints:
             if res["label"] == label:
-                return _entries_dataframe(
-                    res["entries"],
+                return _tensor_json_dataframe(
+                    res,
                     self.problem_outline.constraints[label].bindings,
                     value_name="slack",
                     dual_value_name="shadow_price",
                 )
         raise Exception(f"Unknown constraint {label}")
 
 
-def _entries_dataframe(
-    entries: Sequence[Json],
+def _tensor_json_dataframe(
+    tensor_json: Json,
     bindings: Sequence[SourceBinding],
     *,
     value_name: str = "value",
     dual_value_name: str | None = None,
+    index: pd.Index | Sequence[KeyItem] | None = None,
     round_values: bool = False,
 ) -> pd.DataFrame:
+    entries = tensor_json["entries"]
+    default_values = {
+        value_name: decode_extended_float(tensor_json.get("defaultValue", 0)),
+    }
     if dual_value_name:
         data = (
             (decode_extended_float(e["value"]), e.get("dualValue"))
             for e in entries
         )
         columns = [value_name, dual_value_name]
+        default_values[dual_value_name] = 0
     else:
         data = (decode_extended_float(e["value"]) for e in entries)
         columns = [value_name]
@@ -287,11 +322,11 @@ def _entries_dataframe(
         index=_entry_index(entries, bindings),
     )
     if dual_value_name and df[dual_value_name].isnull().all():
-        df.drop(dual_value_name, axis=1, inplace=True)
+        df = df.drop(dual_value_name, axis=1)
+    df = df.sort_index() if index is None else df.reindex(cast(Any, index))
+    df = df.fillna(default_values)
     if round_values:
         df[value_name] = df[value_name].round(0).astype(np.int64)
-    df.fillna(0, inplace=True)
-    df.sort_index(inplace=True)
     return df
 
 
@@ -463,7 +498,7 @@ class SolveStrategy:
     sense: ObjectiveSense | None = None
     """Optimization sense"""
 
-    epsilon_constraints: list[EpsilonConstraint] = dataclasses.field(
+    epsilon_constraints: Sequence[EpsilonConstraint] = dataclasses.field(
         default_factory=lambda: []
     )
     """All epsilon-constraints to apply"""

src/opvious/data/tensors.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from collections.abc import Iterable, Mapping
+from collections.abc import Iterable, Mapping, Sequence
 import dataclasses
 import math
 from typing import Any, Self
@@ -40,12 +40,12 @@ def is_value(arg: Any) -> bool:
 )
 
 
-@dataclasses.dataclass
+@dataclasses.dataclass(frozen=True)
 class Tensor:
     """An n-dimensional matrix"""
 
-    entries: list[Any]
-    """Raw list of matrix entries"""
+    entries: Sequence[Any]
+    """Raw matrix entries"""
 
     default_value: ExtendedFloat = 0
     """Value to use for missing key"""

src/opvious/modeling/definitions.py

@@ -214,6 +214,9 @@ def total_product_count(self):
     The number of arguments must match the tensor's quantification.
     """
 
+    # TODO: Add map method, which appends to _mappers array of transformations.
+    # Once implemented, remove the negate arguments to transformations.
+
     def __init__(
         self,
         image: Image,

src/opvious/modeling/fragments.py

@@ -625,6 +625,8 @@ def activated_variable(
     indicator_projection: Projection = -1,
     upper_bound: ExpressionLike | None = None,
     negate: bool = False,
+    force_activation: bool = True,
+    force_deactivation: bool = True,
     name: Name | None = None,
 ) -> Callable[[TensorLike], ActivatedVariable]:
     """Wraps a method into an :class:`ActivatedVariable` fragment
@@ -640,6 +642,8 @@ def wrapper(fn: TensorLike) -> ActivatedVariable:
             indicator_projection=indicator_projection,
             upper_bound=upper_bound,
             negate=negate,
+            force_activation=force_activation,
+            force_deactivation=force_deactivation,
             name=name,
         )
 

tests/client_test.py

@@ -86,6 +86,8 @@ async def test_queue_diet_solve(self):
             "salad": 9,
             "caviar": 23,
         }
+        nutrients = input_data.dimension("nutrients")
+        assert list(nutrients) == ["carbs", "fibers", "vitamins"]
 
         output_data = await client.fetch_solve_outputs(uuid)
         quantities = output_data.variable("quantityOfRecipe")