ggt.utils
=========

.. py:module:: ggt.utils


Submodules
----------

.. toctree::
   :maxdepth: 1

   /autoapi/ggt/utils/data_utils/index
   /autoapi/ggt/utils/device_utils/index
   /autoapi/ggt/utils/model_utils/index
   /autoapi/ggt/utils/tensor_utils/index


Functions
---------

.. autoapisummary::

   ggt.utils.discover_devices
   ggt.utils.tensor_to_numpy
   ggt.utils.arsinh_normalize
   ggt.utils.load_tensor
   ggt.utils.standardize_labels
   ggt.utils.metric_output_transform_al_loss
   ggt.utils.metric_output_transform_al_cov_loss
   ggt.utils.load_cat
   ggt.utils.get_output_shape
   ggt.utils.enable_dropout
   ggt.utils.specify_dropout_rate


Package Contents
----------------

.. py:function:: discover_devices()

   Check for available devices.


.. py:function:: tensor_to_numpy(x)

   Convert a torch tensor to NumPy for plotting.


.. py:function:: arsinh_normalize(X)

   Normalize a Torch tensor with arsinh.


.. py:function:: load_tensor(filename, tensors_path, as_numpy=True)

   Load a Torch tensor from disk.


.. py:function:: standardize_labels(input, data_dir, split, slug, label_col, scaling, invert=False)

   Standardizes data. During training, input should
   be the labels, and during inference, input should be the
   predictions.


.. py:function:: metric_output_transform_al_loss(output)

   Transforms the output of the model, when using
   aleatoric loss, to a form which can be used by the
   ignote metric calculators


.. py:function:: metric_output_transform_al_cov_loss(output)

   Transforms the output of the model, when using
   aleatoric covariance loss, to a form which can be used by the
   ignote metric calculators


.. py:function:: load_cat(data_dir, slug, split)

   Loads and returns pandas dataframe


.. py:function:: get_output_shape(model, image_dim)

   Get output shape of a PyTorch model or layer


.. py:function:: enable_dropout(model)

   Enable random dropout during inference. From StackOverflow #63397197


.. py:function:: specify_dropout_rate(model, rate)

   Specify the dropout rate of all layers