kgcnn.utils package

Submodules

kgcnn.utils.devices module

kgcnn.utils.devices.check_device()[source]

Simple function to check for available devices for computing models. Mostly GPUs.

kgcnn.utils.devices.set_cuda_device(device_id: Union[int, list])[source]

Set the cuda device by ID.

Better use cuda environment variable to do this instead of this function:

import os
os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"]="1"  # specify which GPU(s) to be used
Parameters

device_id (int) – ID of the GPU to set.

kgcnn.utils.plots module

kgcnn.utils.plots.plot_predict_true(y_predict, y_true, data_unit: Optional[list] = None, model_name: str = '', filepath: Optional[str] = None, file_name: str = '', dataset_name: str = '', target_names: Optional[list] = None, figsize: Optional[list] = None, dpi: Optional[float] = None, show_fig: bool = False, scaled_predictions: bool = False)[source]

Make a scatter plot of predicted versus actual targets. Not for k-splits.

Parameters

  • y_predict (np.ndarray) – Numpy array of shape (N_samples, n_targets) or (N_samples, ).

  • y_true (np.ndarray) – Numpy array of shape (N_samples, n_targets) or (N_samples, ).

  • data_unit (list) – String or list of string that matches n_targets. Name of the data’s unit.

  • model_name (str) – Name of the model. Default is “”.

  • filepath (str) – Full path where to save plot to, without the name of the file. Default is “”.

  • file_name (str) – File name base. Model name and dataset will be added to the name. Default is “”.

  • dataset_name (str) – Name of the dataset which was fitted to. Default is “”.

  • target_names (list) – String or list of string that matches n_targets. Name of the targets.

  • figsize (list) – Size of the figure. Default is None.

  • dpi (float) – The resolution of the figure in dots-per-inch. Default is None.

  • show_fig (bool) – Whether to show figure. Default is True.

  • scaled_predictions (bool) – Whether predictions had been standardized. Default is False.

Returns

Figure of the scatter plot.

Return type

matplotlib.pyplot.figure

kgcnn.utils.plots.plot_train_test_loss(histories: list, loss_name: Optional[str] = None, val_loss_name: Optional[str] = None, data_unit: str = '', model_name: str = '', filepath: Optional[str] = None, file_name: str = '', dataset_name: str = '', figsize: Optional[list] = None, dpi: Optional[float] = None, show_fig: bool = True)[source]

Plot training curves for a list of fit results in form of keras history objects. This means, training- and test-loss is plotted vs. epochs for all splits.

Parameters

  • histories (list) – List of tf.keras.callbacks.History() objects.

  • loss_name (str) – Which loss or metric to pick from history for plotting. Default is “loss”.

  • val_loss_name (str) – Which validation loss or metric to pick from history for plotting. Default is “val_loss”.

  • data_unit (str) – Unit of the loss. Default is “”.

  • model_name (str) – Name of the model. Default is “”.

  • filepath (str) – Full path where to save plot to, without the name of the file. Default is “”.

  • file_name (str) – File name base. Model name and dataset will be added to the name. Default is “”.

  • dataset_name (str) – Name of the dataset which was fitted to. Default is “”.

  • figsize (list) – Size of the figure. Default is None.

  • dpi (float) – The resolution of the figure in dots-per-inch. Default is None.

  • show_fig (bool) – Whether to show figure. Default is True.

Returns

Figure of the training curves.

Return type

matplotlib.pyplot.figure

kgcnn.utils.serial module

kgcnn.utils.serial.deserialize(obj_dict: dict) → Any[source]

General deserialization scheme for objects. Requires module information. For each submodule there might be a separate deserialization scheme, that works e.g. with name only.

Parameters

obj_dict – Serialized object.

Returns

Class or object from obj_dict.

Return type

Any

kgcnn.utils.serial.serialize(obj) → dict[source]

General serialization scheme for objects. Requires module information. For each submodule there might be a separate serialization scheme, that works e.g. with name only.

Parameters

obj – Object to serialize.

Returns

Serialized object.

Return type

dict

kgcnn.utils.tests module

class kgcnn.utils.tests.TestCase(*args, **kwargs)[source]

Bases: unittest.case.TestCase

static assertAllClose(x1, x2, atol=1e-06, rtol=1e-06, msg=None)[source]
assertAllEqual(x1, x2, msg=None)[source]
assertLen(iterable, expected_len, msg=None)[source]
assertNotAllClose(x1, x2, atol=1e-06, rtol=1e-06, msg=None)[source]
kgcnn.utils.tests.compare_static_shapes(found: Optional[tuple], expected: Optional[tuple])[source]

Module contents