PolyTorch 0.1.3__py3-none-any.whl

polytorch/__init__.py

@@ -0,0 +1,6 @@
+from .data import BinaryData, CategoricalData, ContinuousData, OrdinalData, HierarchicalData, PolyData
+from .embedding import PolyEmbedding
+from .loss import PolyLoss
+from .enums import ContinuousLossType, CategoricalLossType, BinaryLossType
+from .modules import PolyLinear, PolyLazyLinear
+from .util import split_tensor, total_size

polytorch/data.py

@@ -0,0 +1,160 @@
+from torch import nn
+import abc
+from typing import List, Optional
+from attrs import define, Factory, field, validators
+import torch.nn.functional as F
+from hierarchicalsoftmax import HierarchicalSoftmaxLoss, SoftmaxNode
+
+from .util import permute_feature_axis, squeeze_prediction
+from .enums import ContinuousLossType, BinaryLossType, CategoricalLossType
+
+
+@define(kw_only=True)
+class PolyData(abc.ABC):
+    name: str = Factory(lambda self: self.__class__.__name__, takes_self=True)
+    loss_weighting: float = 1.0
+
+    @abc.abstractmethod
+    def embedding_module(self, embedding_size:int) -> nn.Module:
+        pass
+
+    @abc.abstractmethod
+    def size(self) -> int:
+        """ The prediction size of this data type. """
+        pass
+
+    @abc.abstractmethod
+    def calculate_loss(self, prediction, target, feature_axis:int=-1):
+        pass
+
+
+def binary_default_factory():
+    return ["False", "True"]
+
+
+@define
+class BinaryData(PolyData):
+    loss_type:BinaryLossType = BinaryLossType.CROSS_ENTROPY
+    labels:List[str] = field(factory=binary_default_factory)
+    colors:Optional[List[str]] = None
+
+    def embedding_module(self, embedding_size:int) -> nn.Module:
+        return nn.Embedding(2, embedding_size)
+
+    def size(self) -> int:
+        return 1
+
+    def calculate_loss(self, prediction, target, feature_axis:int=-1):
+        prediction = squeeze_prediction(prediction, target, feature_axis)
+
+        if self.loss_type == BinaryLossType.CROSS_ENTROPY:
+            return F.binary_cross_entropy_with_logits(
+                prediction, 
+                target.float(), 
+            )
+        elif self.loss_type == BinaryLossType.IOU:
+            from .metrics import calc_iou
+            return 1 - calc_iou(
+                prediction.sigmoid(), 
+                target, 
+            )
+        elif self.loss_type == BinaryLossType.DICE:
+            from .metrics import calc_dice_score
+            return 1 - calc_dice_score(
+                prediction.sigmoid(), 
+                target, 
+            )
+        
+        raise NotImplementedError(f"Unknown loss type: {self.loss_type} for {self.__class__.__name__}")
+
+
+@define
+class CategoricalData(PolyData):
+    category_count:int
+    loss_type:CategoricalLossType = CategoricalLossType.CROSS_ENTROPY
+    labels:Optional[List[str]] = None
+    colors:Optional[List[str]] = None
+    label_smoothing:float = field(default=0.0, validator=[validators.ge(0.0), validators.le(1.0)])
+
+    def embedding_module(self, embedding_size:int) -> nn.Module:
+        return nn.Embedding(self.category_count, embedding_size)
+
+    def size(self) -> int:
+        return self.category_count
+
+    def calculate_loss(self, prediction, target, feature_axis:int=-1):
+        if self.loss_type == CategoricalLossType.CROSS_ENTROPY:
+            prediction = permute_feature_axis(prediction, old_axis=feature_axis, new_axis=1)
+            return F.cross_entropy(
+                prediction, 
+                target.long(), 
+                reduction="none", 
+                label_smoothing=self.label_smoothing,
+            )
+        elif self.loss_type == CategoricalLossType.DICE:
+            from .metrics import calc_generalized_dice_score
+            return 1. - calc_generalized_dice_score(
+                prediction.softmax(dim=feature_axis), 
+                target, 
+                feature_axis=feature_axis,
+            )
+        
+        raise NotImplementedError(f"Unknown loss type: {self.loss_type} for {self.__class__.__name__}")
+
+
+@define
+class OrdinalData(CategoricalData):
+    color:str = ""
+    # add in option to estimate distances or to set them?
+
+    def embedding_module(self, embedding_size:int) -> nn.Module:
+        from .embedding import OrdinalEmbedding
+        return OrdinalEmbedding(self.category_count, embedding_size)
+
+
+@define
+class ContinuousData(PolyData):
+    loss_type:ContinuousLossType = ContinuousLossType.SMOOTH_L1_LOSS
+    color:str = ""
+    mean:Optional[float] = None
+    stdev:Optional[float] = None
+
+    def embedding_module(self, embedding_size:int) -> nn.Module:
+        from .embedding import ContinuousEmbedding
+        return ContinuousEmbedding(embedding_size, mean=self.mean, stdev=self.stdev)
+
+    def size(self) -> int:
+        return 1
+    
+    def calculate_loss(self, prediction, target, feature_axis:int=-1):
+        prediction = squeeze_prediction(prediction, target, feature_axis)
+        if self.mean is not None:
+            target = target - self.mean
+        if self.stdev is not None:
+            target = target / self.stdev
+        return self.loss_func(prediction, target, reduction="none")
+
+    @property
+    def loss_func(self):
+        return getattr(F, self.loss_type.name.lower())
+
+
+@define
+class HierarchicalData(PolyData):
+    root:SoftmaxNode
+
+    def embedding_module(self, embedding_size:int) -> nn.Module:
+        raise NotImplementedError("Hierarchical data types are not yet supported for embedding.")
+
+    def size(self) -> int:
+        self.root.set_indexes_if_unset()
+        return self.root.layer_size
+    
+    def calculate_loss(self, prediction, target, feature_axis:int=-1):
+        prediction = permute_feature_axis(prediction, old_axis=feature_axis, new_axis=1)
+        return self.loss_module(prediction, target)
+
+    @property
+    def loss_module(self):
+        return HierarchicalSoftmaxLoss(root=self.root)
+

polytorch/embedding.py

@@ -0,0 +1,116 @@
+from typing import List
+
+import torch
+from torch import nn
+from torch.nn.parameter import Parameter
+import torch.nn.functional as F
+import plotly.graph_objects as go
+
+from .data import PolyData
+from .util import permute_feature_axis
+
+class ContinuousEmbedding(nn.Module):
+    def __init__(
+        self, 
+        embedding_size:int,
+        bias:bool=True,
+        mean:float|None=None,
+        stdev:float|None=None,
+        device=None, 
+        dtype=None,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)    
+        
+        self.embedding_size = embedding_size
+        self.mean = mean
+        self.stdev = stdev
+
+        factory_kwargs = {'device': device, 'dtype': dtype}
+        self.weight = Parameter(torch.empty((embedding_size,), **factory_kwargs), requires_grad=True)
+        if bias:
+            self.bias = Parameter(torch.empty((embedding_size,), **factory_kwargs), requires_grad=True)
+        else:
+            self.bias = Parameter(torch.zeros((embedding_size,), **factory_kwargs), requires_grad=False)
+
+        self.reset_parameters()
+
+    def forward(self, input):
+        x = input.flatten().unsqueeze(1)
+        if self.mean is not None:
+            x = x - self.mean
+        if self.stdev is not None:
+            x = x/self.stdev
+        embedded = self.bias + x * self.weight.unsqueeze(0)
+        embedded = embedded.reshape(input.shape + (-1,))
+
+        return embedded
+
+    def reset_parameters(self) -> None:
+        torch.nn.init.normal_(self.weight)
+        torch.nn.init.constant_(self.bias, 0.0)
+
+
+class OrdinalEmbedding(ContinuousEmbedding):
+    def __init__(
+        self,
+        category_count,
+        embedding_size,
+        bias:bool=True,
+        device=None, 
+        dtype=None,
+        **kwargs,
+    ):
+        super().__init__(
+            embedding_size,
+            bias=bias,
+            device=device, 
+            dtype=dtype,
+            **kwargs,
+        )    
+        factory_kwargs = {'device': device, 'dtype': dtype}
+        self.distance_scores = Parameter(torch.ones((category_count-1,), **factory_kwargs), requires_grad=True)
+
+    def forward(self, x):
+        distances = torch.cumsum(F.softmax(self.distance_scores, dim=0), dim=0)
+        
+        # prepend zero
+        distances = torch.cat([torch.zeros((1,), device=distances.device, dtype=distances.dtype), distances])
+        distance = torch.gather(distances, 0, x.flatten())
+        embedded = self.bias + distance.unsqueeze(1) * self.weight.unsqueeze(0)
+        embedded = embedded.reshape(x.shape + (-1,))
+
+        return embedded
+
+
+class PolyEmbedding(nn.Module):
+    def __init__(
+        self,
+        input_types:List[PolyData],
+        embedding_size:int,  
+        feature_axis:int=-1,      
+        **kwargs,
+    ):
+        super().__init__(**kwargs)    
+        self.input_types = input_types
+        self.embedding_size = embedding_size
+        self.embedding_modules = nn.ModuleList([
+            input.embedding_module(embedding_size) for input in input_types
+        ])
+        self.feature_axis = feature_axis
+
+    def forward(self, *inputs):
+        shape = inputs[0].shape + (self.embedding_size,)
+        embedded = torch.zeros( shape, device=inputs[0].device ) 
+
+        for input, module in zip(inputs, self.embedding_modules):
+            if input.dtype == torch.bool:
+                input = input.int()
+            embedded += module(input)
+
+        return permute_feature_axis(embedded, old_axis=-1, new_axis=self.feature_axis)
+
+    def plot(self, **kwargs) -> go.Figure:
+        from .plots import plot_embedding
+        return plot_embedding(self, **kwargs)
+

polytorch/enums.py

@@ -0,0 +1,22 @@
+from enum import Enum
+import torch.nn.functional as F
+
+
+class ContinuousLossType(Enum):
+    L1_LOSS = 0
+    MSE_LOSS = 1
+    SMOOTH_L1_LOSS = 2
+
+
+class BinaryLossType(Enum):
+    CROSS_ENTROPY = 0
+    IOU = 1
+    DICE = 2
+    
+
+class CategoricalLossType(Enum):
+    CROSS_ENTROPY = 0
+    DICE = 1
+    # TODO: Focal Loss
+    # TODO Wasserstein distance (Earth mover loss)
+

polytorch/loss.py

@@ -0,0 +1,35 @@
+from typing import List
+from torch import nn
+
+from .data import PolyData
+from .util import split_tensor
+
+class PolyLoss(nn.Module):
+    def __init__(
+        self,
+        data_types:List[PolyData],
+        feature_axis:int=-1,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.data_types = data_types   
+        self.feature_axis = feature_axis 
+
+    def forward(self, predictions, *targets):
+        
+
+        if not isinstance(predictions, (tuple, list)):
+            predictions = split_tensor(predictions, self.data_types, feature_axis=self.feature_axis)
+
+        assert len(predictions) == len(targets) == len(self.data_types)
+
+        loss = 0.0
+        for prediction, target, data_type in zip(predictions, targets, self.data_types):
+            feature_axis = self.feature_axis % len(prediction.shape)
+
+            if not hasattr(data_type, "calculate_loss"):
+                raise ValueError(f"Data type {data_type} does not have a calculate_loss method")
+            
+            loss += data_type.loss_weighting * data_type.calculate_loss(prediction, target, feature_axis=feature_axis).mean()
+
+        return loss

polytorch/metrics.py

@@ -0,0 +1,203 @@
+from attrs import define, Factory
+import torch
+import torch.nn.functional as F
+from typing import Optional
+from hierarchicalsoftmax import SoftmaxNode, greedy_accuracy
+
+from .util import squeeze_prediction
+
+
+def get_predictions_target_for_index(predictions, *targets, data_index=None, feature_axis=-1):
+    if not isinstance(predictions, (tuple, list)):
+        predictions = (predictions,)
+
+    assert len(predictions) == len(targets)
+    assert data_index is not None, "data_index must be specified"
+    assert data_index >= 0, "data_index must be non-negative"
+
+    my_predictions = predictions[data_index]
+    my_targets = targets[data_index]
+
+    return my_predictions, my_targets
+
+
+def function_metric(predictions, *targets, data_index=None, feature_axis=-1, function=None) -> torch.Tensor:
+    my_predictions, my_targets = get_predictions_target_for_index(predictions, *targets, data_index=data_index, feature_axis=feature_axis)
+    my_predictions = squeeze_prediction(my_predictions, my_targets, feature_axis)
+    return function(my_predictions, my_targets)
+
+
+def categorical_accuracy(predictions, *targets, data_index=None, feature_axis=-1) -> torch.Tensor:
+    my_predictions, my_targets = get_predictions_target_for_index(predictions, *targets, data_index=data_index, feature_axis=feature_axis)
+    my_predictions = torch.argmax(my_predictions, dim=feature_axis)
+
+    accuracy = (my_predictions == my_targets).float().mean()
+    return accuracy
+
+
+def binary_accuracy(predictions, *targets, data_index=None, feature_axis=-1) -> torch.Tensor:
+    my_predictions, my_targets = get_predictions_target_for_index(predictions, *targets, data_index=data_index, feature_axis=feature_axis)
+    my_predictions = my_predictions >= 0.0
+    my_predictions = squeeze_prediction(my_predictions, my_targets, feature_axis)
+
+    accuracy = (my_predictions == my_targets).float().mean()
+    return accuracy
+
+
+def mse(predictions, *targets, data_index=None, feature_axis=-1) -> torch.Tensor:
+    return function_metric(predictions, *targets, data_index=data_index, feature_axis=feature_axis, function=F.mse_loss)
+
+
+def l1(predictions, *targets, data_index=None, feature_axis=-1) -> torch.Tensor:
+    return function_metric(predictions, *targets, data_index=data_index, feature_axis=feature_axis, function=F.l1_loss)
+
+
+@define(kw_only=True)
+class PolyMetric():
+    data_index:int
+    feature_axis:int = -1
+    function:Optional[callable] = None
+    name:str = Factory(lambda self:self.default_name(), takes_self=True)
+
+    def default_name(self):
+        if self.function:
+            return self.function.__name__
+        return self.__class__.__name__
+
+    @property
+    def __name__(self):
+        """ For using as a FastAI metric. """
+        return self.name
+    
+    def __call__(self, predictions, *targets):
+        my_predictions, my_targets = get_predictions_target_for_index(predictions, *targets, data_index=self.data_index, feature_axis=self.feature_axis)
+        my_predictions = squeeze_prediction(my_predictions, my_targets, self.feature_axis)
+
+        return self.calc(my_predictions, my_targets)
+    
+    def calc(self, predictions, targets):
+        if self.function:
+            return self.function(predictions, targets)
+        
+        raise NotImplementedError(f"No function given. Either implement calc or provide a function to {self.name}")
+
+
+@define
+class HierarchicalGreedyAccuracy(PolyMetric):
+    root:SoftmaxNode
+    max_depth:Optional[int]=None
+
+    def calc(self, predictions, targets):
+        return greedy_accuracy(predictions, targets, self.root, max_depth=self.max_depth)
+
+
+class CategoricalAccuracy(PolyMetric):
+    def calc(self, predictions, targets):
+        predictions = torch.max(predictions, dim=self.feature_axis).indices # should be argmax but there is an issue using MPS
+        return (predictions == targets).float().mean()
+    
+
+def smooth_l1(predictions, *targets, data_index=None, feature_axis=-1) -> torch.Tensor:
+    return function_metric(predictions, *targets, data_index=data_index, feature_axis=feature_axis, function=F.smooth_l1_loss)
+
+
+def calc_dice_score(predictions, target, smooth:float=1.) -> torch.Tensor:
+    predictions = predictions.view(-1)
+    target = target.view(-1)
+    intersection = (predictions * target).sum()
+    
+    return ((2. * intersection + smooth) /
+              (predictions.sum() + target.sum() + smooth)
+    )
+
+
+def calc_generalized_dice_score(predictions, target, power:float=2.0, smooth:float=1.0, feature_axis:int=-1) -> torch.Tensor:
+    """
+    A generalized Dice score for multi-class segmentation.
+
+    If power=0.0, this is equivalent to normal Dice score (i.e. volume "implicit" weighting)
+    If power=1.0, this is equivalent to 'equal' weighting.
+    If power=2.0, this is equivalent to 'inverse volume' weighting.
+
+    See:
+        - https://www.sciencedirect.com/science/article/pii/S2590005619300049#bib73
+        - https://arxiv.org/pdf/1707.03237.pdf
+        - https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=1717643
+    """
+    n_classes = predictions.shape[feature_axis]
+    numerator = 0.0
+    denominator = 0.0
+    slice_indices = [slice(0, None)] * len(predictions.shape)
+    for i in range(n_classes):
+        my_target = (target == i)
+        my_target_sum = my_target.sum()
+        weight = 1/(my_target_sum**power + smooth)
+
+        slice_indices[feature_axis] = i
+        my_predictions = predictions[slice_indices]
+
+        numerator += weight * (my_predictions*my_target).sum()
+        denominator += weight * (my_predictions.sum() + my_target_sum)
+    
+    return 2. * numerator / denominator
+
+
+def calc_iou(predictions, target, smooth:float=1.):
+    predictions = predictions.view(-1)
+    target = target.view(-1)
+    intersection = (predictions * target).sum()
+    union = predictions.sum() + target.sum() - intersection
+    
+    return (intersection + smooth) / (union + smooth)
+
+
+def binary_dice(predictions, *targets, data_index=None, feature_axis=-1):
+    my_predictions, my_targets = get_predictions_target_for_index(predictions, *targets, data_index=data_index, feature_axis=feature_axis)
+    my_predictions = my_predictions >= 0.0
+    my_predictions = squeeze_prediction(my_predictions, my_targets, feature_axis)
+
+    return calc_dice_score(my_predictions, my_targets)
+
+
+def binary_iou(predictions, *targets, data_index=None, feature_axis=-1):
+    my_predictions, my_targets = get_predictions_target_for_index(predictions, *targets, data_index=data_index, feature_axis=feature_axis)
+    my_predictions = my_predictions >= 0.0
+    my_predictions = squeeze_prediction(my_predictions, my_targets, feature_axis)
+
+    return calc_iou(my_predictions, my_targets)
+
+
+def generalized_dice(predictions, *targets, data_index=None, feature_axis=-1, perform_softmax:bool=True, power:float=2.0) -> torch.Tensor:
+    """
+    Calculate the generalized dice score for a single data index. Used for for multi-class segmentation.
+
+    See:
+        - https://www.sciencedirect.com/science/article/pii/S2590005619300049#bib73
+        - https://arxiv.org/pdf/1707.03237.pdf
+        - https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=1717643
+
+    Args:
+        predictions (_type_): Logits or probabilities. If probabilities, perform_softmax must be False.
+        data_index (_type_, optional): The index of the target to . Defaults to None.
+        feature_axis (int, optional): _description_. Defaults to -1.
+        perform_softmax (bool, optional): Whether or not to normalize the predictions using the softmax function. Defaults to True.
+        power (float, optional): The power to use for the generalized dice score. Defaults to 2.0 (i.e. inverse volume weighting)
+            If power=0.0, this is equivalent to normal Dice score (i.e. volume "implicit" weighting)
+            If power=1.0, this is equivalent to 'equal' weighting.
+            If power=2.0, this is equivalent to 'inverse volume' weighting.
+
+    Returns:
+        torch.Tensor: The generalized dice score for the given data index.
+    """
+    my_predictions, my_targets = get_predictions_target_for_index(predictions, *targets, data_index=data_index, feature_axis=feature_axis)
+    if perform_softmax:
+        my_predictions = my_predictions.softmax(dim=feature_axis)
+    
+    score = calc_generalized_dice_score(
+        my_predictions, 
+        my_targets, 
+        feature_axis=feature_axis,
+        power=power,
+    )
+    return score
+    

polytorch/modules.py

@@ -0,0 +1,41 @@
+from torch import nn
+from typing import List
+
+from .data import PolyData
+from .util import total_size, split_tensor
+
+
+class PolyLayerError(RuntimeError):
+    pass
+
+
+class PolyLayerMixin():
+    def __init__(self, output_types:List[PolyData], out_features=None, **kwargs):
+        self.output_types = output_types
+
+        if out_features is not None:
+            raise PolyLayerError(
+                "Trying to create a PolyLinear Layer by explicitly setting `out_features`. "
+                "This value should be determined from the list of output types and not the `out_features` argument."
+            )
+
+        super().__init__(out_features=total_size(self.output_types), **kwargs)
+
+    def forward(self, *inputs):
+        outputs = super().forward(*inputs)
+        return split_tensor(outputs, self.output_types, feature_axis=-1)
+
+
+class PolyLinear(PolyLayerMixin, nn.Linear):
+    """
+    Creates a linear layer designed to be the final layer in a neural network model that produces unnormalized scores given to PolyLoss.
+
+    The `out_features` value is set internally from root.layer_size and cannot be given as an argument.
+    """
+
+class PolyLazyLinear(PolyLayerMixin, nn.LazyLinear):
+    """
+    Creates a lazy linear layer designed to be the final layer in a neural network model that produces unnormalized scores given to PolyLoss.
+
+    The `out_features` value is set internally from root.layer_size and cannot be given as an argument.
+    """

polytorch/plots.py

@@ -0,0 +1,137 @@
+from pathlib import Path
+import plotly.graph_objects as go
+from sklearn.decomposition import PCA
+from itertools import cycle
+import torch
+
+from .embedding import PolyEmbedding
+from .data import CategoricalData, OrdinalData
+
+
+def format_fig(fig) -> go.Figure:
+    """Formats a plotly figure in a nicer way."""
+    fig.update_layout(
+        width=1000,
+        height=800,
+        plot_bgcolor="white",
+        title_font_color="black",
+        font=dict(
+            family="Linux Libertine Display O",
+            size=18,
+            color="black",
+        ),
+    )
+    gridcolor = "#dddddd"
+    fig.update_xaxes(gridcolor=gridcolor)
+    fig.update_yaxes(gridcolor=gridcolor)
+
+    fig.update_xaxes(showline=True, linewidth=1, linecolor='black', mirror=True, ticks='outside', zeroline=True, zerolinewidth=1, zerolinecolor='black')
+    fig.update_yaxes(showline=True, linewidth=1, linecolor='black', mirror=True, ticks='outside', zeroline=True, zerolinewidth=1, zerolinecolor='black')
+
+    return fig
+
+
+def plot_embedding(embedding:PolyEmbedding, n_components:int=2, show:bool=False, output_path:str|Path|None=None) -> go.Figure:
+    """
+    Plots the embedding in 2D or 3D.
+
+    Args:
+        embedding (PolyEmbedding): The embedding to plot.
+        n_components (int, optional): The number of principal components to plot. Can be 2 or 3. Defaults to 2.
+        show (bool, optional): Whether to show the plot. Defaults to False.
+        output_path (str|Path|None, optional): The path to save the plot to. 
+            Can be in HTML, PNG, JPEG, SVG or PDF. Defaults to None.
+
+    Returns:
+        go.Figure: The plotly figure
+    """
+    if n_components not in [2, 3]:
+        raise ValueError(f"n_components must be 2 or 3, not {n_components}")
+    
+    # get embedding weights
+    weights = []
+    labels = []
+    colors = []
+
+    # The default colors are same as px.colors.qualitative.Plotly
+    # I'm not using that directly because that requires plotly express
+    # which requires pandas to be installed
+    cmap = cycle(['#636EFA', '#EF553B', '#00CC96', '#AB63FA', '#FFA15A', '#19D3F3', '#FF6692', '#B6E880', '#FF97FF', '#FECB52'])
+
+    for input_type, module in zip(embedding.input_types, embedding.embedding_modules):
+        weight = module.weight
+        if len(weight.shape) == 1:
+            weight = weight.unsqueeze(0)
+        
+        weights.append(weight)
+        
+        if isinstance(input_type, CategoricalData) and not isinstance(input_type, OrdinalData):
+            my_labels = (
+                input_type.labels if input_type.labels is not None 
+                else [f"{input_type.name}_{i}" for i in range(input_type.category_count)]
+            )
+            labels.extend(my_labels)
+
+            my_colors = (
+                input_type.colors if input_type.colors is not None 
+                else [next(cmap) for _ in range(input_type.category_count)]
+            )
+            colors.extend(my_colors)
+        else:
+            labels.append(input_type.name)
+            colors.append(getattr(input_type, "color", '') or next(cmap))
+
+    weights = torch.cat(weights, dim=0).detach()
+
+    # Perform a principal component analysis
+    pca = PCA(n_components=n_components)
+    weights_reduced = pca.fit_transform(weights)
+
+    # plot
+    fig = go.Figure()
+
+    # This is done as a loop so that the legend has all the different categorical labels separate
+    # This will be a large list in the legend potentially
+    # This could be an option in the future
+    for vector, label, color in zip(weights_reduced, labels, colors):
+        if n_components == 2:
+            fig.add_trace(go.Scatter(
+                x=[vector[0]],
+                y=[vector[1]],
+                mode='markers',
+                name=label,
+                marker_color=color,
+            ))
+            fig.update_xaxes(title_text="Component 1")
+            fig.update_yaxes(title_text="Component 2")
+        elif n_components == 3:
+            fig.add_trace(go.Scatter3d(
+                x=[vector[0]],
+                y=[vector[1]],
+                z=[vector[2]],
+                mode='markers',
+                name=label,
+                marker_color=color,
+            ))
+            fig.update_layout(scene = dict(
+                xaxis_title='Component 1',
+                yaxis_title='Component 2',
+                zaxis_title='Component 3',
+            ))
+        
+    fig.update_layout(showlegend=True)
+    format_fig(fig)
+
+    # output image as file
+    if output_path:
+        output_path = Path(output_path)
+        output_path.parent.mkdir(parents=True, exist_ok=True)
+        if output_path.suffix == '.html':
+            fig.write_html(str(output_path))
+        else:
+            fig.write_image(str(output_path))
+
+    if show:
+        fig.show()
+
+    return fig

polytorch/util.py

@@ -0,0 +1,73 @@
+from typing import List, Tuple
+from torch import Tensor
+
+
+def total_size(data_types) -> int:
+    """
+    Calculates the total number of features required to predict a list of output types.
+
+    Args:
+        data_types (List[PolyData]): The data types to predict.
+
+    Returns:
+        int: The number of features required to predict the given data types.
+    """
+    return sum(data_type.size() for data_type in data_types)
+
+
+def split_tensor(tensor:Tensor, data_types, feature_axis:int=-1) -> Tuple[Tensor, ...]:
+    """
+    Splits a tensor into a tuple of tensors, one for each data type.
+
+    Args:
+        tensor (Tensor): The predictions tensor.
+        data_types (List[PolyData]): The data types to predict.
+        feature_axis (int, optional): The axis which has the features to predict. Defaults to last axis.
+
+    Returns:
+        Tuple[Tensor, ...]: A tuple of tensors, one for each data type.
+    """
+    current_index = 0
+    split_tensors = []
+    slice_indices = [slice(0, None)] * len(tensor.shape)
+    for data_type in data_types:
+        size = data_type.size()
+        slice_indices[feature_axis] = slice(current_index,current_index+size)
+        split_tensors.append(tensor[slice_indices])
+        current_index += size
+    
+    assert current_index == tensor.shape[feature_axis]
+
+    return tuple(split_tensors)
+
+
+def permute_feature_axis(tensor:Tensor, old_axis:int, new_axis:int) -> Tensor:
+    """
+    Changes the shape of a tensor so that the feature axis is in a new axis.
+
+    Args:
+        tensor (torch.Tensor): The tensor to permute.
+        new_axis (int): The desired index of the feature axis.
+
+    Returns:
+        torch.Tensor: The predictions tensor with the feature axis at the specified index.
+    """
+    axes_count = len(tensor.shape)
+    if old_axis % axes_count != new_axis % axes_count:
+        axes = list(range(axes_count))
+        axes.insert(new_axis, axes.pop(old_axis))
+        return tensor.permute(*axes)
+    return tensor
+
+
+def squeeze_prediction(prediction:Tensor, target:Tensor, feature_axis:int):
+    """
+    Squeeze feature axis if necessary
+    """
+    feature_axis = feature_axis % len(prediction.shape)
+    if (
+        len(prediction.shape) == len(target.shape) + 1 and 
+        prediction.shape[:feature_axis] + prediction.shape[feature_axis+1:] == target.shape
+    ):
+        prediction = prediction.squeeze(feature_axis)
+    return prediction

polytorch-0.1.3.dist-info/LICENSE

@@ -0,0 +1,192 @@
+
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   Copyright 2023 Robert Turnbull
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

polytorch-0.1.3.dist-info/METADATA

@@ -0,0 +1,240 @@
+Metadata-Version: 2.3
+Name: PolyTorch
+Version: 0.1.3
+Summary: Embeddings and loss functions for different data types.
+License: Apache-2.0
+Keywords: pytorch,deep learning,command-line interface
+Author: Robert Turnbull
+Author-email: robert.turnbull@unimelb.edu.au
+Requires-Python: >=3.10,<4.0
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Requires-Dist: attrs (>=19)
+Requires-Dist: hierarchicalsoftmax (>=1.4.1)
+Requires-Dist: kaleido (==0.2.1)
+Requires-Dist: plotly (>=5.15.0)
+Requires-Dist: scikit-learn (>=1.0.2)
+Requires-Dist: torch (>=1.10.2)
+Project-URL: Documentation, https://rbturnbull.github.io/polytorch
+Project-URL: Homepage, https://github.com/rbturnbull/polytorch/
+Project-URL: Repository, https://github.com/rbturnbull/polytorch/
+Description-Content-Type: text/x-rst
+
+================================================================
+polytorch
+================================================================
+
+.. start-badges
+
+|pypi badge| |testing badge| |coverage badge| |docs badge| |black badge| 
+
+.. |pypi badge| image:: https://img.shields.io/pypi/v/polytorch?color=blue
+   :alt: PyPI - Version
+   :target: https://pypi.org/project/polytorch/
+
+.. |testing badge| image:: https://github.com/rbturnbull/polytorch/actions/workflows/testing.yml/badge.svg
+    :target: https://github.com/rbturnbull/polytorch/actions
+
+.. |docs badge| image:: https://github.com/rbturnbull/polytorch/actions/workflows/docs.yml/badge.svg
+    :target: https://rbturnbull.github.io/polytorch
+    
+.. |black badge| image:: https://img.shields.io/badge/code%20style-black-000000.svg
+    :target: https://github.com/psf/black
+    
+.. |coverage badge| image:: https://img.shields.io/endpoint?url=https://gist.githubusercontent.com/rbturnbull/f84ac74436887cd42d77fbe2246d1f57/raw/coverage-badge.json
+    :target: https://rbturnbull.github.io/polytorch/coverage/
+    
+.. end-badges
+
+.. start-quickstart
+
+Embeddings and loss functions for different data types.
+
+Installation
+==================================
+
+Install using pip:
+
+.. code-block:: bash
+
+    pip install polytorch
+
+Or install the latest version from GitHub:
+
+.. code-block:: bash
+
+    pip install git+https://github.com/rbturnbull/polytorch.git
+
+
+Data Types
+==================================
+
+This package allow you to input and output different data types in PyTorch models.
+
+Binary Data
+----------------------------------------
+
+.. code-block:: python
+
+    from polytorch import BinaryData
+
+    binary_data = BinaryData()
+
+    # Or with labels and colors
+    binary_data = BinaryData(labels=["no_feature", "with_feature", colors=["red", "blue"])
+
+Categorical Data
+----------------------------------------
+
+.. code-block:: python
+
+    from polytorch import CategoricalData
+
+    category_count = 5  # Number of categories
+    categorical_data = CategoricalData(category_count)
+
+    # Or with labels, colors and label smoothing
+    categorical_data = CategoricalData(
+        category_count=category_count,
+        labels=["cat", "dog", "fish", "bird", "reptile"],
+        colors=["red", "blue", "green", "yellow", "purple"],
+        label_smoothing=0.1,
+    )
+
+Ordinal Data
+----------------------------------------
+
+.. code-block:: python
+
+    from polytorch import OrdinalData
+
+    ordinal_data = OrdinalData()
+    
+    # Or with color
+    ordinal_data = OrdinalData(color="pink")
+    
+Continuous Data
+----------------------------------------
+
+.. code-block:: python
+
+    from polytorch import ContinuousData
+
+    continuous_data = ContinuousData()
+
+    # Or with color
+    continuous_data = ContinuousData(color="orange")
+
+
+Hierarchical Data
+----------------------------------------
+
+
+.. code-block:: python
+
+    from polytorch import HierarchicalData
+    from hiearchicalsoftmax import SoftmaxNode
+
+    root = SoftmaxNode("root")
+    child1 = SoftmaxNode("child1", parent=root)
+    child2 = SoftmaxNode("child2", parent=root)
+    tip1 = SoftmaxNode("tip1", parent=child1)
+    tip2 = SoftmaxNode("tip2", parent=child1)
+    tip3 = SoftmaxNode("tip3", parent=child2)
+    tip4 = SoftmaxNode("tip4", parent=child2)
+    
+    
+    hierarchical_data = HierarchicalData(root)
+    
+Embedding your data
+==================================
+
+.. code-block:: python
+
+    from torch import nn
+    from polytorch import Embedding
+    
+    class MyModule(nn.Module):
+        def __init__(self, embedding_size:int=128):
+            super(MyModule, self).__init__()
+
+            input_types = [binary_data, categorical_data] # for example. Could be other data types as well.
+            self.embedding = PolyEmbedding( input_types=input_types, embedding_size=embedding_size)
+            
+            # Other modules
+            ...
+
+        def forward(self, x_binary, x_categorical):
+            
+            embedded = self.embedding( x_binary, x_categorical )
+
+            # Use the embedded features in your model
+            ...            
+
+.. warning:: 
+
+    The ``HierarchicalData`` type is not yet supported by the ``PolyEmbedding`` class.
+
+Outputting your data
+==================================
+
+You can also get your model to output to different data types.
+
+.. code-block:: python
+
+    from torch import nn
+    from polytorch import PolyLazyLinear
+
+    output_types = [
+        CategoricalData(category_count=5, loss_weighting=0.5),  # For example, a categorical output with 5 categories
+        BinaryData(loss_weighting=1.0),                        # A binary output
+        ContinuousData(loss_weighting=0.1)                     # A continuous output
+    ]
+
+    class MyModule(nn.Module):
+        def __init__(self, output_types):
+            super(MyModule, self).__init__()
+
+            self.output = PolyLazyLinear(output_types=output_types)
+
+        def forward(self, x):
+            # Your model logic
+            ...
+
+            # Output to different data types
+            return self.output(x)
+
+Then add set this as the loss:
+
+.. code-block:: python
+
+    from polytorch import PolyLoss
+
+    loss_module = PolyLoss(output_types=output_types)
+
+    # In your training loop
+    loss = loss_module(predictions, categorical_target, binary_target, continuous_target)
+
+.. end-quickstart
+
+
+Credits
+==================================
+
+.. start-credits
+
+Robert Turnbull
+For more information contact: <robert.turnbull@unimelb.edu.au>
+
+Created using torchapp (https://github.com/rbturnbull/torchapp).
+
+.. end-credits
+
+

polytorch-0.1.3.dist-info/RECORD

@@ -0,0 +1,13 @@
+polytorch/__init__.py,sha256=QMfk8tmvhhjAS6YCWpAFgBqIPImiMiBR7ohMAXo_76g,332
+polytorch/data.py,sha256=Lfb08Ln6FRo9Zn9MtJK9bKQ1w-Ihx9cZBJv2GtnkJik,5351
+polytorch/embedding.py,sha256=T5o3WFeGjVfjOWQA14ex43Mg6nnwJXWd6UptVBCmSRA,3614
+polytorch/enums.py,sha256=Xq0DbL6ek0SHhd_rjWGf_ZI_fJ28iJXf5HPa8yqGH98,370
+polytorch/loss.py,sha256=0qrByugj4kkD9a_2X4bOL8VKwnXBTpzC3okh7JCauj0,1148
+polytorch/metrics.py,sha256=qXVnU0PR_fjL8LexOpaaq7q_vY01wHRIMfi59zjJx5Q,8431
+polytorch/modules.py,sha256=MKmB9R4Qy9ZJ8pOT3SLrxLHaOE8nUO6Srf2MjI2zYRI,1459
+polytorch/plots.py,sha256=AQY8QJrf76UVdzDft44aMZvMSFXbov8YYdTwgJI32XU,4790
+polytorch/util.py,sha256=AYElA3agqtuxTODIQoVhz32jpGMryzt-pXTBhGaBYjU,2408
+polytorch-0.1.3.dist-info/LICENSE,sha256=bGHVFIy-mmbSifFidZ-YRE1Rq1xVlKv2KSsoQzJjcEM,10761
+polytorch-0.1.3.dist-info/METADATA,sha256=pmQMso0g1cMfQhnvtP363ALnU1R-CXAeswf61CWNugM,6828
+polytorch-0.1.3.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
+polytorch-0.1.3.dist-info/RECORD,,

polytorch-0.1.3.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: poetry-core 2.1.1
+Root-Is-Purelib: true
+Tag: py3-none-any