inference

adabn_fn(x, domain, models, **kwargs)

AdaBN features and logits.

Parameters:

Name	Type	Description	Default
x		The input to the model	required
domain		0 for source domain, 1 for target domain.	required

Source code in pytorch_adapt\inference\inference.py

def adabn_fn(x, domain, models, **kwargs) -> Dict[str, torch.Tensor]:
    """
    [AdaBN][pytorch_adapt.adapters.AdaBN] features and logits.
    Arguments:
        x: The input to the model
        domain: 0 for source domain, 1 for target domain.
    """
    domain = check_domain(domain, keep_len=True)
    features = models["G"](x, domain)
    logits = models["C"](features, domain)
    return {"features": features, "logits": logits}

adda_fn(x, domain, models, get_all=False, **kwargs)

ADDA features and logits.

Parameters:

Name	Type	Description	Default
x		The input to the model	required
domain		If 0, then features = G(x). Otherwise features = T(x).	required
models		Dictionary of models with keys ["G", "C", "T"].	required
get_all	bool	If True, then return features and logits using both G and T as the feature extractor.	False

Returns:

Type	Description
Dict[str, torch.Tensor]	A dictionary of features and logits. If get_all is False, then the keys are {"features", "logits"}. If get_all is True, then the keys will be {"features", "logits", "other_features", "other_logits"}, where the other_ prefix represents the features and logits obtained using G if domain == 1 and T if domain == 0.

Source code in pytorch_adapt\inference\inference.py

def adda_fn(
    x, domain, models, get_all: bool = False, **kwargs
) -> Dict[str, torch.Tensor]:
    """
    [ADDA][pytorch_adapt.adapters.ADDA] features and logits.
    Arguments:
        x: The input to the model
        domain: If 0, then ```features = G(x)```.
            Otherwise ```features = T(x)```.
        models: Dictionary of models with keys
            ```["G", "C", "T"]```.
        get_all: If ```True```, then return features
            and logits using both ```G``` and ```T```
            as the feature extractor.
    Returns:
        A dictionary of features and logits.

            - If ```get_all``` is ```False```, then the keys are ```{"features", "logits"}```.

            - If ```get_all``` is ```True```,
            then the keys will be ```{"features", "logits", "other_features", "other_logits"}```,
            where the ```other_``` prefix represents the features and logits obtained
            using ```G``` if ```domain == 1``` and ```T``` if ```domain == 0```.
    """
    domain = check_domain(domain)
    fe = "G" if domain == 0 else "T"
    features = models[fe](x)
    logits = models["C"](features)
    output = {"features": features, "logits": logits}
    if get_all:
        fe = "T" if fe == "G" else "G"
        features = models[fe](x)
        logits = models["C"](features)
        output.update({"other_features": features, "other_logits": logits})
    return output

adda_full_fn(x, **kwargs)

ADDA features, logits, discriminator logits, other features, other logits, other discriminator logits. See adda_fn for the input arguments.

Returns:

Type	Description
Dict[str, torch.Tensor]	discriminator logits ("d_logits"), "other" discriminator logits ("other_d_logits") in addition to everything returned by adda_fn with get_all = True.

Source code in pytorch_adapt\inference\inference.py

def adda_full_fn(x, **kwargs) -> Dict[str, torch.Tensor]:
    """
    [ADDA][pytorch_adapt.adapters.ADDA] features, logits, discriminator logits,
    other features, other logits, other discriminator logits.
    See [adda_fn][pytorch_adapt.inference.adda_fn] for the input arguments.
    Returns:
        discriminator logits (```"d_logits"```), "other" discriminator logits (```"other_d_logits"```)
            in addition to everything returned by [adda_fn][pytorch_adapt.inference.adda_fn]
            with ```get_all = True```.
    """
    layer = kwargs.get("layer", "features")
    output = with_d(x=x, fn=adda_fn, get_all=True, **kwargs)
    output2 = d_fn(x=output[f"other_{layer}"], **kwargs)
    output["other_d_logits"] = output2["d_logits"]
    return output

adda_with_d(**kwargs)

ADDA features, logits, and discriminator logits. See adda_fn for the input arguments.

Returns:

Type	Description
Dict[str, torch.Tensor]	discriminator logits as "d_logits", in addition to everything returned by adda_fn.

Source code in pytorch_adapt\inference\inference.py

def adda_with_d(**kwargs) -> Dict[str, torch.Tensor]:
    """
    [ADDA][pytorch_adapt.adapters.ADDA] features, logits, and discriminator logits. See
    [adda_fn][pytorch_adapt.inference.adda_fn] for the input arguments.
    Returns:
        discriminator logits as ```"d_logits"```, in addition to
            everything returned by [adda_fn][pytorch_adapt.inference.adda_fn].
    """
    return with_d(fn=adda_fn, **kwargs)

default_fn(x, models, **kwargs)

The default inference function for BaseAdapter.

Source code in pytorch_adapt\inference\inference.py

def default_fn(x, models, **kwargs) -> Dict[str, torch.Tensor]:
    """
    The default inference function for [BaseAdapter][pytorch_adapt.adapters.BaseAdapter].
    """
    features = models["G"](x)
    logits = models["C"](features)
    return {"features": features, "logits": logits}

mcd_fn(x, models, get_all=False, **kwargs)

Returns:

Type	Description
	Features and logits, where logits = sum(C(features)).

Source code in pytorch_adapt\inference\inference.py

def mcd_fn(x, models, get_all=False, **kwargs):
    """
    Returns:
        Features and logits, where ```logits = sum(C(features))```.
    """
    features = models["G"](x)
    logits_list = models["C"](features)
    logits = sum(logits_list)
    output = {"features": features, "logits": logits}
    if get_all:
        for i, L in enumerate(logits_list):
            output[f"logits{i}"] = L
    return output

rtn_fn(x, domain, models, get_all=False, **kwargs)

RTN features and logits.

Parameters:

Name	Type	Description	Default
x		The input to the model	required
domain		If 0, logits = residual_model(C(G(x))). Otherwise, logits = C(G(x)).	required
models		Dictionary of models with keys ["G", "C", "residual_model"].	required
get_all		If True, then in addition to the regular outputs, it will return the residual_model logits when domain == 1 and the C logits when domain == 0.	False

Returns:

Type	Description
Dict[str, torch.Tensor]	A dictionary of features and logits. If get_all is False, then the keys are {"features", "logits"}. If get_all is True, then the keys will be {"features", "logits", "other_logits"}.

Source code in pytorch_adapt\inference\inference.py

def rtn_fn(x, domain, models, get_all=False, **kwargs) -> Dict[str, torch.Tensor]:
    """
    [RTN][pytorch_adapt.adapters.RTN] features and logits.
    Arguments:
        x: The input to the model
        domain: If 0, ```logits = residual_model(C(G(x)))```.
            Otherwise, ```logits = C(G(x))```.
        models: Dictionary of models with keys
            ```["G", "C", "residual_model"]```.
        get_all: If ```True```, then in addition to the regular outputs,
            it will return the ```residual_model``` logits when
            ```domain == 1``` and the ```C``` logits when ```domain == 0```.
    Returns:
        A dictionary of features and logits.

            - If ```get_all``` is ```False```, then the keys are ```{"features", "logits"}```.

            - If ```get_all``` is ```True```,
            then the keys will be ```{"features", "logits", "other_logits"}```.
    """
    domain = check_domain(domain)
    f_dict = default_fn(x=x, models=models)
    target_logits = f_dict["logits"]
    if get_all or domain == 0:
        src_logits = models["residual_model"](target_logits)
    if domain == 0:
        f_dict["logits"] = src_logits
        if get_all:
            f_dict["other_logits"] = target_logits
    elif get_all and domain == 1:
        f_dict["other_logits"] = src_logits
    return f_dict

symnets_fn(x, domain, models, get_all=False, **kwargs)

Parameters:

Name	Type	Description	Default
x		The input to the model	required
domain		0 for the source domain, 1 for the target domain.	required

Returns:

Type	Description
	Features and logits, where logits = C(features)[domain].

Source code in pytorch_adapt\inference\inference.py

def symnets_fn(x, domain, models, get_all=False, **kwargs):
    """
    Arguments:
        x: The input to the model
        domain: 0 for the source domain, 1 for the target domain.
    Returns:
        Features and logits, where ```logits = C(features)[domain]```.
    """
    domain = check_domain(domain)
    features = models["G"](x)
    logits = models["C"](features)[domain]
    output = {"features": features, "logits": logits}
    if get_all:
        logits = models["C"](features)[int(not domain)]
        output.update({"other_logits": logits})
    return output