Skip to content

ocl.callbacks

FreezeParameters

Bases: Callback

Freeze parameters of model prior to training.

Source code in ocl/callbacks.py 
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
class FreezeParameters(Callback):
    """Freeze parameters of model prior to training."""

    def __init__(self, parameter_groups: List[Dict[str, Any]]):
        """Initialize FreezeParameters callback.

        Args:
            parameter_groups: Parameter groups that should be frozen.
                Uses same syntax as [ocl.optimization.OptimizationWrapper][]
        """
        super().__init__()
        self.parameter_group_specs = parameter_groups
        for idx, param_group_spec in enumerate(self.parameter_group_specs):
            if "params" not in param_group_spec:
                raise ValueError(f'Parameter group {idx + 1} does not contain key "params"')
            param_spec = param_group_spec["params"]
            if isinstance(param_spec, str):
                param_group_spec["params"] = [param_spec]
            elif isinstance(param_spec, Iterable):
                param_group_spec["params"] = list(param_spec)
            else:
                raise ValueError(
                    f'"params" for parameter group {idx + 1} is not of type str or iterable'
                )

            if "predicate" in param_group_spec:
                if not callable(param_group_spec["predicate"]):
                    raise ValueError(f'"predicate" for parameter group {idx + 1} is not a callable')

    def _get_parameters_to_freeze(self, model):
        """Build parameter groups from specification."""
        parameters_to_freeze = []
        for param_group_spec in self.parameter_group_specs:
            for current_params in param_group_spec["params"]:
                param_path = current_params.split(".")
                # Default predicate includes all parameters
                predicate = param_group_spec.get("predicate", lambda name, param: True)
                param = get_tree_element(model, param_path)
                if isinstance(param, nn.Module):
                    parameters_to_freeze.extend(
                        param for name, param in param.named_parameters() if predicate(name, param)
                    )
                elif isinstance(param, nn.Parameter):
                    parameters_to_freeze.append(param)
                else:
                    raise ValueError(
                        "Object at path {'.'.join(param_path)} is neither nn.Module nor nn.Parameter"
                    )
        return parameters_to_freeze

    def on_fit_start(self, trainer, model: nn.Module):
        parameters_to_freeze = self._get_parameters_to_freeze(model)
        for param in parameters_to_freeze:
            param.requires_grad_(False)

__init__

Initialize FreezeParameters callback.

Parameters:

Name Type Description Default
parameter_groups List[Dict[str, Any]]

Parameter groups that should be frozen. Uses same syntax as ocl.optimization.OptimizationWrapper

 required
Source code in ocl/callbacks.py 
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
def __init__(self, parameter_groups: List[Dict[str, Any]]):
    """Initialize FreezeParameters callback.

    Args:
        parameter_groups: Parameter groups that should be frozen.
            Uses same syntax as [ocl.optimization.OptimizationWrapper][]
    """
    super().__init__()
    self.parameter_group_specs = parameter_groups
    for idx, param_group_spec in enumerate(self.parameter_group_specs):
        if "params" not in param_group_spec:
            raise ValueError(f'Parameter group {idx + 1} does not contain key "params"')
        param_spec = param_group_spec["params"]
        if isinstance(param_spec, str):
            param_group_spec["params"] = [param_spec]
        elif isinstance(param_spec, Iterable):
            param_group_spec["params"] = list(param_spec)
        else:
            raise ValueError(
                f'"params" for parameter group {idx + 1} is not of type str or iterable'
            )

        if "predicate" in param_group_spec:
            if not callable(param_group_spec["predicate"]):
                raise ValueError(f'"predicate" for parameter group {idx + 1} is not a callable')

RestoreParameterSubset

Bases: Callback

Restore a subset of parameters using a checkpoint form a different model.

Source code in ocl/callbacks.py 
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
class RestoreParameterSubset(Callback):
    """Restore a subset of parameters using a checkpoint form a different model."""

    def __init__(self, checkpoint_file: str, target_path: str, source_path: Optional[str] = None):
        """Initialize RestoreParameterSubset callback.

        Args:
            checkpoint_file: File from which the model weights should be loaded.
            target_path: The path in the model where the model weights should be
                restored.  This should follow a dot separated syntax, such a `encoder.layer1`.
            source_path: The path in the checkpoint_file that should be used to restore weights.
                If none provided assumes to be the same as `target_path`.

        """
        self.checkpoint_file = checkpoint_file
        self.target_path = target_path
        self.source_path = source_path if source_path else self.target_path

    def on_fit_start(self, trainer, model: nn.Module):
        if model.global_step != 0:
            # Don't restore when we are resuming training.
            rank_zero_warn("Not restoring parameter subset as training is being resumed")
            return
        # Get parameters from state dict
        state_dict = torch.load(self.checkpoint_file, map_location=model.device)["state_dict"]
        # Add offset of 1 to remove potential dot.
        offset_keys = len(self.source_path) + 1
        state_dict = {
            key[offset_keys:]: value
            for key, value in state_dict.items()
            if key.startswith(self.source_path)
        }

        # Get module from model
        model_component: nn.Module = get_tree_element(model, self.target_path.split("."))
        result = model_component.load_state_dict(state_dict)
        if len(result.missing_keys):
            rank_zero_warn(
                f"Mismatch between state dict and model. Missing keys: {result.missing_keys}"
            )
        if len(result.unexpected_keys):
            rank_zero_warn(
                f"Mismatch between state dict and model. Unexpected keys: {result.missing_keys}"
            )
        rank_zero_info(f"Restored subset of model parameters from {self.checkpoint_file}")

__init__

Initialize RestoreParameterSubset callback.

Parameters:

Name Type Description Default
checkpoint_file str

File from which the model weights should be loaded.

 required
target_path str

The path in the model where the model weights should be restored. This should follow a dot separated syntax, such a encoder.layer1.

 required
source_path Optional[str]

The path in the checkpoint_file that should be used to restore weights. If none provided assumes to be the same as target_path.

 None
Source code in ocl/callbacks.py 
72
73
74
75
76
77
78
79
80
81
82
83
84
85
def __init__(self, checkpoint_file: str, target_path: str, source_path: Optional[str] = None):
    """Initialize RestoreParameterSubset callback.

    Args:
        checkpoint_file: File from which the model weights should be loaded.
        target_path: The path in the model where the model weights should be
            restored.  This should follow a dot separated syntax, such a `encoder.layer1`.
        source_path: The path in the checkpoint_file that should be used to restore weights.
            If none provided assumes to be the same as `target_path`.

    """
    self.checkpoint_file = checkpoint_file
    self.target_path = target_path
    self.source_path = source_path if source_path else self.target_path

UpdateHyperparameterScheduling

Bases: Callback

Callback to update hyperparameter schedulers found ocl.scheduling.

Source code in ocl/callbacks.py 
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
class UpdateHyperparameterScheduling(Callback):
    """Callback to update hyperparameter schedulers found `ocl.scheduling`."""

    def __init__(self):
        self._hyperparameter_schedulers: List[HPScheduler] = []

    def on_fit_start(self, trainer, pl_module):
        del trainer
        self._hyperparameter_schedulers = list(
            map(
                lambda a: a[1],
                walk_tree_with_paths(pl_module, instance_check=lambda t: isinstance(t, HPScheduler)),
            )
        )
        # Set global step to 0 for pretraining evaluation routine.
        self._update_schedulers(0)

    def _update_schedulers(self, step):
        if len(self._hyperparameter_schedulers) == 0:
            rank_zero_warn(
                "UpdateHyperparameterScheduling: "
                "No schedulable hyperparameters where found in model."
            )
        for hparam in self._hyperparameter_schedulers:
            hparam.update_global_step(step)

    def on_train_batch_start(self, trainer, pl_module, batch, batch_idx):
        del trainer, batch, batch_idx
        global_step = pl_module.global_step
        self._update_schedulers(global_step)

SetEpochEnvironmentVariable

Bases: Callback

Sets environment variable EPOCH which is used by ocl.transforms.SampleSlices.

Source code in ocl/callbacks.py 
148
149
150
151
152
class SetEpochEnvironmentVariable(Callback):
    """Sets environment variable `EPOCH` which is used by [ocl.transforms.SampleSlices][]."""

    def on_train_epoch_start(self, trainer, pl_module):
        os.environ["EPOCH"] = str(pl_module.current_epoch)