Configs for Ax API entry point class

Summary: The new Ax API will rely on configs to bundle together related information while the user sets up their optimization. These classes will have a stable and backwards compatible API (in contrast to their ax.core counterparts), better reflect how a user conceptualizes their optimization without leaking in "implementation details", and be a centralized place for validation.

Reviewed By: lena-kashtelyan

Differential Revision: D58022351

ax/preview/api/configs.py

@@ -0,0 +1,108 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+# This source code is licensed under the MIT license found in the
+# LICENSE file in the root directory of this source tree.
+
+from dataclasses import dataclass, field
+from enum import Enum
+from typing import Dict, List, Optional, Union
+
+from ax.core.types import TParamValue
+
+# Note: I'm not sold these should be dataclasses, just using this as a placeholder
+
+
+class DomainType(Enum):
+    """
+    The DomainType enum allows the ParameterConfig to know whether to expect inputs for
+    a RangeParameter or ChoiceParameter (or FixedParameter) during the parameter
+    instantiation and validation process.
+    """
+
+    RANGE = "range"
+    CHOICE = "choice"
+
+
+class ParameterType(Enum):
+    """
+    The ParameterType enum allows users to specify the type of a parameter. This can often
+    """
+
+    INT = "int"
+    FLOAT = "float"
+    BOOL = "bool"
+    STR = "str"
+
+
+class ParameterScaling(Enum):
+    """
+    The ParameterScaling enum allows users to specify which scaling to apply during
+    candidate generation. This is useful for parameters that should not be explored
+    on the same scale, such as learning rates and batch sizes.
+    """
+
+    LINEAR = "linear"
+    LOG = "log"
+
+
+@dataclass
+class ParameterConfig:
+    """
+    ParameterConfig allows users to specify the parameters of an experiment and will
+    internally validate the inputs to ensure they are valid for the given DomainType.
+    """
+
+    name: str
+    domain_type: DomainType
+    parameter_type: ParameterType | None = None
+
+    # Fields for RANGE
+    bounds: Optional[tuple[float, float]] = None
+    step_size: Optional[float] = None
+    scaling: Optional[ParameterScaling] = None
+
+    # Fields for CHOICE ("FIXED" is Choice with len(values) == 1)
+    values: Optional[Union[List[float], List[str], List[bool]]] = None
+    is_ordered: Optional[bool] = None
+    dependent_parameters: Optional[Dict[TParamValue, str]] = None
+
+
+@dataclass
+class ExperimentConfig:
+    """
+    ExperimentConfig allows users to specify the SearchSpace and OptimizationConfig of
+    an Experiment and validates their inputs jointly.
+
+    This will also be the construct that handles transforming string-based inputs (the
+    objective, parameter constraints, and output constraints) into their corresponding
+    Ax class using SymPy.
+    """
+
+    name: str
+    parameters: List[ParameterConfig]
+    # Parameter constraints will be parsed via SymPy
+    # Ex: "num_layers1 <= num_layers2", "compound_a + compound_b <= 1"
+    parameter_constraints: List[str] = field(default_factory=list)
+
+    description: str | None = None
+    owner: str | None = None
+
+
+@dataclass
+class GenerationStrategyConfig:
+    # This will hold the args to choose_generation_strategy
+    num_trials: Optional[int] = None
+    num_initialization_trials: Optional[int] = None
+    maximum_parallelism: Optional[int] = None
+
+
+@dataclass
+class OrchestrationConfig:
+    parallelism: int = 1
+    tolerated_trial_failure_rate: float = 0.5
+    seconds_between_polls: float = 1.0
+
+
+@dataclass
+class DatabaseConfig:
+    url: str