Petridish

Evaluater

archai.supergraph.algos.petridish.evaluater_petridish.filepath_ext(filepath: str) → str

Returns ‘.f’ for ‘/a/b/c/d.e.f’

archai.supergraph.algos.petridish.evaluater_petridish.filepath_name_only(filepath: str) → str

Returns ‘d.e’ for ‘/a/b/c/d.e.f’

archai.supergraph.algos.petridish.evaluater_petridish.append_to_filename(filepath: str, name_suffix: str, new_ext: str | None = None) → str

Returns ‘/a/b/c/h.f’ for filepath=’/a/b/c/d.e.f’, new_name=’h’

class archai.supergraph.algos.petridish.evaluater_petridish.EvaluaterPetridish

evaluate(conf_eval: Config, model_desc_builder: ModelDescBuilder) → EvalResult

Takes a folder of model descriptions output by search process and trains them in a distributed manner using ray with 1 gpu

Experiment Runner

archai.supergraph.algos.petridish.petridish_exp_runner.copy_dir(src_dir: str, dest_dir: str, use_shutil: bool = True) → None

class archai.supergraph.algos.petridish.petridish_exp_runner.PetridishExperimentRunner(config_filename: str, base_name: str, clean_expdir=False)

model_desc_builder() → PetridishModelBuilder

trainer_class() → Type[ArchTrainer] | None

searcher() → SearcherPetridish

evaluater() → EvaluaterPetridish

copy_search_to_eval() → None

Model Description Builder

class archai.supergraph.algos.petridish.petridish_model_desc_builder.PetridishModelBuilder

pre_build(conf_model_desc: Config) → None

hook for accomplishing any setup before build starts

Operators

class archai.supergraph.algos.petridish.petridish_op.StopForward

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class archai.supergraph.algos.petridish.petridish_op.StopGradient

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class archai.supergraph.algos.petridish.petridish_op.StopForwardReductionOp(op_desc: OpDesc, affine: bool)

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class archai.supergraph.algos.petridish.petridish_op.StopGradientReduction(op_desc: OpDesc, affine: bool)

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class archai.supergraph.algos.petridish.petridish_op.TempIdentityOp(op_desc)

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool

class archai.supergraph.algos.petridish.petridish_op.PetridishOp(op_desc: OpDesc, arch_params: ArchParams | None, reduction: bool, affine: bool)

PRIMITIVES = ['max_pool_3x3', 'avg_pool_3x3', 'skip_connect', 'sep_conv_3x3', 'sep_conv_5x5', 'dil_conv_3x3', 'dil_conv_5x5', 'none']

forward(x: List[Tensor])

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

finalize() → Tuple[OpDesc, float | None]

for trainable op, return final op and its rank

ops() → Iterator[Tuple[Op, float]]

Return contituent ops, if this op is primitive just return self

training: bool

Utilities

class archai.supergraph.algos.petridish.petridish_utils.JobStage(value)

An enumeration.

SEED = 1

SEED_TRAINED = 2

SEARCH = 3

SEARCH_TRAINED = 4

EVAL = 5

EVAL_TRAINED = 6

class archai.supergraph.algos.petridish.petridish_utils.ExperimentStage(value)

An enumeration.

SEARCH = 1

EVAL = 2

class archai.supergraph.algos.petridish.petridish_utils.ConvexHullPoint(job_stage: JobStage, parent_id: int, sampling_count: int, model_desc: ModelDesc, cells_reductions_nodes: Tuple[int, int, int], metrics: Metrics | None = None, model_stats: ModelStats | None = None)

is_trained_stage() → bool

next_stage() → JobStage

archai.supergraph.algos.petridish.petridish_utils.model_descs_on_front(hull_points: List[ConvexHullPoint], convex_hull_eps: float, stage: ExperimentStage, lower_hull: bool = True) → Tuple[List[ConvexHullPoint], List[ConvexHullPoint], List[float], List[float]]

archai.supergraph.algos.petridish.petridish_utils.hull_points2tsv(points: List[ConvexHullPoint]) → str

archai.supergraph.algos.petridish.petridish_utils.sample_from_hull(hull_points: List[ConvexHullPoint], convex_hull_eps: float, stage: ExperimentStage = ExperimentStage.SEARCH) → ConvexHullPoint

archai.supergraph.algos.petridish.petridish_utils.get_top1_for_stage(hull_points: List[ConvexHullPoint], stage: ExperimentStage) → List[float]

Return top1 accuracy according to the experiment stage (SEARCH or EVAL)

archai.supergraph.algos.petridish.petridish_utils.plot_frontier(hull_points: List[ConvexHullPoint], convex_hull_eps: float, expdir: str, stage: ExperimentStage = ExperimentStage.SEARCH) → None

archai.supergraph.algos.petridish.petridish_utils.plot_pool(hull_points: List[ConvexHullPoint], expdir: str, stage: ExperimentStage = ExperimentStage.SEARCH) → None

archai.supergraph.algos.petridish.petridish_utils.plot_seed_model_stats(seed_model_stats: List[ModelStats], expdir: str) → None

archai.supergraph.algos.petridish.petridish_utils.save_hull_frontier(hull_points: List[ConvexHullPoint], convex_hull_eps: float, final_desc_foldername: str, expdir: str, stage: ExperimentStage = ExperimentStage.SEARCH) → ConvexHullPoint

archai.supergraph.algos.petridish.petridish_utils.save_hull(hull_points: List[ConvexHullPoint], expdir: str) → None

Searcher

class archai.supergraph.algos.petridish.searcher_petridish.SearcherPetridish

search(conf_search: Config, model_desc_builder: ModelDescBuilder, trainer_class: Type[ArchTrainer] | None, finalizers: Finalizers) → SearchResult

search_model_desc_dist = <ray.remote_function.RemoteFunction object>

train_model_desc_dist = <ray.remote_function.RemoteFunction object>

build_model_desc(model_desc_builder: ModelDescBuilder, conf_model_desc: Config, reductions: int, cells: int, nodes: int) → ModelDesc