dhg.experiments
Base Class
- class dhg.experiments.BaseTask(work_root, data, model_builder, train_builder, evaluator, device, structure_builder=None, study_name=None, overwrite=True)[source]
Bases:
object
The base class of Auto-experiment in DHG.
- Parameters
work_root (
Optional[Union[str, Path]]
) – User’s work root to store all studies.data (
dict
) – The dictionary to store input data that used in the experiment.model_builder (
Callable
) – The function to build a model with a fixed parametertrial
.train_builder (
Callable
) – The function to build a training configuration with two fixed parameterstrial
andmodel
.evaluator (
dhg.metrics.BaseEvaluator
) – The DHG evaluator object to evaluate performance of the model in the experiment.device (
torch.device
) – The target device to run the experiment.structure_builder (
Optional[Callable]
) – The function to build a structure with a fixed parametertrial
. The structure can bedhg.Graph
,dhg.DiGraph
,dhg.BiGraph
, anddhg.Hypergraph
.study_name (
Optional[str]
) – The name of this study. If set toNone
, the study name will be generated automatically according to current time. Defaults toNone
.overwrite (
bool
) – The flag that whether to overwrite the existing study. Different studies are identified by thestudy_name
. Defaults toTrue
.
- experiment(trial)[source]
Run the experiment for a given trial.
- Parameters
trial (
optuna.Trial
) – Theoptuna.Trial
object.
- run(max_epoch, num_trials=1, direction='maximize')[source]
Run experiments with automatically hyper-parameter tuning.
- Parameters
max_epoch (
int
) – The maximum number of epochs to train for each experiment.num_trials (
int
) – The number of trials to run. Defaults to1
.direction (
str
) – The direction to optimize. Defaults to"maximize"
.
- test(data=None, model=None)
Test the model.
- Parameters
data (
dict
, optional) – The input data if set toNone
, the specifieddata
in the intialization of the experiments will be used. Defaults toNone
.model (
nn.Module
, optional) – The model if set toNone
, the trained best model will be used. Defaults toNone
.
- abstract train(data, model, optimizer, criterion)[source]
Train model for one epoch.
- Parameters
data (
dict
) – The input data.model (
nn.Module
) – The model.optimizer (
torch.optim.Optimizer
) – The model optimizer.criterion (
nn.Module
) – The loss function.
- validate(data, model)
Validate the model.
- Parameters
data (
dict
) – The input data.model (
nn.Module
) – The model.
Vertex Classification Task
- class dhg.experiments.VertexClassificationTask(work_root, data, model_builder, train_builder, evaluator, device, structure_builder=None, study_name=None, overwrite=True)[source]
Bases:
dhg.experiments.base.BaseTask
The auto-experiment class for the vertex classification task.
- Parameters
work_root (
Optional[Union[str, Path]]
) – User’s work root to store all studies.data (
dict
) – The dictionary to store input data that used in the experiment.model_builder (
Callable
) – The function to build a model with a fixed parametertrial
.train_builder (
Callable
) – The function to build a training configuration with two fixed parameterstrial
andmodel
.evaluator (
dhg.metrics.BaseEvaluator
) – The DHG evaluator object to evaluate performance of the model in the experiment.device (
torch.device
) – The target device to run the experiment.structure_builder (
Optional[Callable]
) – The function to build a structure with a fixed parametertrial
. The structure can bedhg.Graph
,dhg.DiGraph
,dhg.BiGraph
, anddhg.Hypergraph
.study_name (
Optional[str]
) – The name of this study. If set toNone
, the study name will be generated automatically according to current time. Defaults toNone
.overwrite (
bool
) – The flag that whether to overwrite the existing study. Different studies are identified by thestudy_name
. Defaults toTrue
.
- experiment(trial)[source]
Run the experiment for a given trial.
- Parameters
trial (
optuna.Trial
) – Theoptuna.Trial
object.
- run(max_epoch, num_trials=1, direction='maximize')[source]
Run experiments with automatically hyper-parameter tuning.
- Parameters
max_epoch (
int
) – The maximum number of epochs to train for each experiment.num_trials (
int
) – The number of trials to run. Defaults to1
.direction (
str
) – The direction to optimize. Defaults to"maximize"
.
- test(data=None, model=None)
Test the model.
- Parameters
data (
dict
, optional) – The input data if set toNone
, the specifieddata
in the intialization of the experiments will be used. Defaults toNone
.model (
nn.Module
, optional) – The model if set toNone
, the trained best model will be used. Defaults toNone
.
- to(device)[source]
Move the input data to the target device.
- Parameters
device (
torch.device
) – The specified target device to store the input data.
- train(data, model, optimizer, criterion)[source]
Train model for one epoch.
- Parameters
data (
dict
) – The input data.model (
nn.Module
) – The model.optimizer (
torch.optim.Optimizer
) – The model optimizer.criterion (
nn.Module
) – The loss function.
- validate(data, model)
Validate the model.
- Parameters
data (
dict
) – The input data.model (
nn.Module
) – The model.
- property vars_for_DL
Return a name list for available variables for deep learning in the vertex classification task. The name list includes
features
,structure
,labels
,train_mask
,val_mask
, andtest_mask
.
On Graph
- class dhg.experiments.GraphVertexClassificationTask(work_root, data, model_builder, train_builder, evaluator, device, study_name=None, overwrite=True)[source]
Bases:
dhg.experiments.vertex_classification.VertexClassificationTask
The auto-experiment class for the vertex classification task on graph.
- Parameters
work_root (
Optional[Union[str, Path]]
) – User’s work root to store all studies.data (
dict
) – The dictionary to store input data that used in the experiment.model_builder (
Callable
) – The function to build a model with a fixed parametertrial
.train_builder (
Callable
) – The function to build a training configuration with two fixed parameterstrial
andmodel
.evaluator (
dhg.metrics.BaseEvaluator
) – The DHG evaluator object to evaluate performance of the model in the experiment.device (
torch.device
) – The target device to run the experiment.structure_builder (
Optional[Callable]
) – The function to build a structure with a fixed parametertrial
. The structure should bedhg.Graph
.study_name (
Optional[str]
) – The name of this study. If set toNone
, the study name will be generated automatically according to current time. Defaults toNone
.overwrite (
bool
) – The flag that whether to overwrite the existing study. Different studies are identified by thestudy_name
. Defaults toTrue
.
- experiment(trial)[source]
Run the experiment for a given trial.
- Parameters
trial (
optuna.Trial
) – Theoptuna.Trial
object.
- run(max_epoch, num_trials=1, direction='maximize')[source]
Run experiments with automatically hyper-parameter tuning.
- Parameters
max_epoch (
int
) – The maximum number of epochs to train for each experiment.num_trials (
int
) – The number of trials to run. Defaults to1
.direction (
str
) – The direction to optimize. Defaults to"maximize"
.
- test(data=None, model=None)
Test the model.
- Parameters
data (
dict
, optional) – The input data if set toNone
, the specifieddata
in the intialization of the experiments will be used. Defaults toNone
.model (
nn.Module
, optional) – The model if set toNone
, the trained best model will be used. Defaults toNone
.
- to(device)[source]
Move the input data to the target device.
- Parameters
device (
torch.device
) – The specified target device to store the input data.
- train(data, model, optimizer, criterion)[source]
Train model for one epoch.
- Parameters
data (
dict
) – The input data.model (
nn.Module
) – The model.optimizer (
torch.optim.Optimizer
) – The model optimizer.criterion (
nn.Module
) – The loss function.
- validate(data, model)
Validate the model.
- Parameters
data (
dict
) – The input data.model (
nn.Module
) – The model.
- property vars_for_DL
Return a name list for available variables for deep learning in the vertex classification on graph. The name list includes
features
,structure
,labels
,train_mask
,val_mask
, andtest_mask
.
On Hypergraph
- class dhg.experiments.HypergraphVertexClassificationTask(work_root, data, model_builder, train_builder, evaluator, device, structure_builder=None, study_name=None, overwrite=True)[source]
Bases:
dhg.experiments.vertex_classification.VertexClassificationTask
The auto-experiment class for the vertex classification task on hypergraph.
- Parameters
work_root (
Optional[Union[str, Path]]
) – User’s work root to store all studies.data (
dict
) – The dictionary to store input data that used in the experiment.model_builder (
Callable
) – The function to build a model with a fixed parametertrial
.train_builder (
Callable
) – The function to build a training configuration with two fixed parameterstrial
andmodel
.evaluator (
dhg.metrics.BaseEvaluator
) – The DHG evaluator object to evaluate performance of the model in the experiment.device (
torch.device
) – The target device to run the experiment.structure_builder (
Optional[Callable]
) – The function to build a structure with a fixed parametertrial
. The structure should bedhg.Hypergraph
.study_name (
Optional[str]
) – The name of this study. If set toNone
, the study name will be generated automatically according to current time. Defaults toNone
.overwrite (
bool
) – The flag that whether to overwrite the existing study. Different studies are identified by thestudy_name
. Defaults toTrue
.
- experiment(trial)[source]
Run the experiment for a given trial.
- Parameters
trial (
optuna.Trial
) – Theoptuna.Trial
object.
- run(max_epoch, num_trials=1, direction='maximize')[source]
Run experiments with automatically hyper-parameter tuning.
- Parameters
max_epoch (
int
) – The maximum number of epochs to train for each experiment.num_trials (
int
) – The number of trials to run. Defaults to1
.direction (
str
) – The direction to optimize. Defaults to"maximize"
.
- test(data=None, model=None)
Test the model.
- Parameters
data (
dict
, optional) – The input data if set toNone
, the specifieddata
in the intialization of the experiments will be used. Defaults toNone
.model (
nn.Module
, optional) – The model if set toNone
, the trained best model will be used. Defaults toNone
.
- to(device)[source]
Move the input data to the target device.
- Parameters
device (
torch.device
) – The specified target device to store the input data.
- train(data, model, optimizer, criterion)[source]
Train model for one epoch.
- Parameters
data (
dict
) – The input data.model (
nn.Module
) – The model.optimizer (
torch.optim.Optimizer
) – The model optimizer.criterion (
nn.Module
) – The loss function.
- validate(data, model)
Validate the model.
- Parameters
data (
dict
) – The input data.model (
nn.Module
) – The model.
- property vars_for_DL
Return a name list for available variables for deep learning in the vertex classification on hypergraph. The name list includes
features
,structure
,labels
,train_mask
,val_mask
, andtest_mask
.
Recommender Task
On User-Item Bipartite Graph
- class dhg.experiments.UserItemRecommenderTask(work_root, data, model_builder, train_builder, evaluator, device, structure_builder=None, study_name=None, overwrite=True)[source]
Bases:
dhg.experiments.base.BaseTask
The auto-experiment class for the recommender task on user-item bipartite graph.
- Parameters
work_root (
Optional[Union[str, Path]]
) – User’s work root to store all studies.data (
dict
) – The dictionary to store input data that used in the experiment.model_builder (
Callable
) – The function to build a model with a fixed parametertrial
.train_builder (
Callable
) – The function to build a training configuration with two fixed parameterstrial
andmodel
.evaluator (
dhg.metrics.BaseEvaluator
) – The DHG evaluator object to evaluate performance of the model in the experiment.device (
torch.device
) – The target device to run the experiment.structure_builder (
Optional[Callable]
) – The function to build a structure with a fixed parametertrial
. The structure should bedhg.DiGraph
.study_name (
Optional[str]
) – The name of this study. If set toNone
, the study name will be generated automatically according to current time. Defaults toNone
.overwrite (
bool
) – The flag that whether to overwrite the existing study. Different studies are identified by thestudy_name
. Defaults toTrue
.
- experiment(trial)[source]
Run the experiment for a given trial.
- Parameters
trial (
optuna.Trial
) – Theoptuna.Trial
object.
- run(max_epoch, num_trials=1, direction='maximize')[source]
Run experiments with automatically hyper-parameter tuning.
- Parameters
max_epoch (
int
) – The maximum number of epochs to train for each experiment.num_trials (
int
) – The number of trials to run. Defaults to1
.direction (
str
) – The direction to optimize. Defaults to"maximize"
.
- test(data=None, model=None)
Test the model.
- Parameters
data (
dict
, optional) – The input data if set toNone
, the specifieddata
in the intialization of the experiments will be used. Defaults toNone
.model (
nn.Module
, optional) – The model if set toNone
, the trained best model will be used. Defaults toNone
.
- to(device)[source]
Move the input data to the target device.
- Parameters
device (
torch.device
) – The specified target device to store the input data.
- train(data, model, optimizer, criterion)[source]
Train model for one epoch.
- Parameters
data (
dict
) – The input data.model (
nn.Module
) – The model.optimizer (
torch.optim.Optimizer
) – The model optimizer.criterion (
nn.Module
) – The loss function.
- validate(data, model)
Validate the model.
- Parameters
data (
dict
) – The input data.model (
nn.Module
) – The model.
- property vars_for_DL
Return a name list for available deep learning variables for the recommender task on user-item bipartite graph. The name list includes
structure
.