pyphoon.app¶
Content¶
| module | Description |
|---|---|
pyphoon.app.preprocess |
Pre-processing tools for DL/ML applications. |
pyphoon.app.utils |
General tools. |
pyphoon.app.preprocess¶
-
pyphoon.app.preprocess.get_mean_image(X, display=False)¶ Computes the mean image from a list of image batches.
Parameters: - X (list) – List containing image batches. That is, an element of the list
is of shape (N, W, H), where N: #samples, W: image width, H: image
height. Arrays of size (N, W, H, C) with C: #channels are also accepted.
If you only have a single batch (e.g.
B), you just need to encapsulate it in a list (i.e.[B]). - display (bool) – Set to True to display information as function is executed.
Returns: Mean image (size W x H).
Return type: numpy.array
- X (list) – List containing image batches. That is, an element of the list
is of shape (N, W, H), where N: #samples, W: image width, H: image
height. Arrays of size (N, W, H, C) with C: #channels are also accepted.
If you only have a single batch (e.g.
-
pyphoon.app.preprocess.get_mean_pixel(X, display=False)¶ Computes the mean pixel from all samples in the list of image batches X.
Parameters: - X (list) – List containing image batches. That is, an element of the list
is of shape (N, W, H), where N: #samples, W: image width, H: image
height. Arrays of size (N, W, H, C) with C: #channels are also accepted.
If you only have a single batch (e.g.
B), you just need to encapsulate it in a list (i.e.[B]). - display (bool) – Set to True to display information as function is executed.
Returns: Pixel mean (scalar).
Return type: float
- X (list) – List containing image batches. That is, an element of the list
is of shape (N, W, H), where N: #samples, W: image width, H: image
height. Arrays of size (N, W, H, C) with C: #channels are also accepted.
If you only have a single batch (e.g.
-
pyphoon.app.preprocess.get_std_pixel(X, pmean, display=False)¶ Computes the pixel standard deviation from all samples in the list of image batches X.
Parameters: - X (list) – List containing image batches. That is, an element of the list
is of shape (N, W, H), where N: #samples, W: image width, H: image
height. Arrays of size (N, W, H, C) with C: #channels are also accepted.
If you only have a single batch (e.g.
B), you just need to encapsulate it in a list (i.e.[B]). - pmean (float) – Pixel mean of bath list X (see
get_mean_pixel()). - display (bool) – Set to True to display information as function is executed.
Returns: Pixel standard deviation (scalar).
Return type: float
- X (list) – List containing image batches. That is, an element of the list
is of shape (N, W, H), where N: #samples, W: image width, H: image
height. Arrays of size (N, W, H, C) with C: #channels are also accepted.
If you only have a single batch (e.g.
-
pyphoon.app.preprocess.get_max_min(X, display=False)¶ Gets the maximum and minimum pixel values of a list of image batches.
Parameters: - X (list) – List containing image batches. That is, an element of the list
is of shape (N, W, H), where N: #samples, W: image width, H: image
height. Arrays of size (N, W, H, C) with C: #channels are also accepted.
If you only have a single batch (e.g.
B), you just need to encapsulate it in a list (i.e.[B]). - display (bool) – Set to True to display information as function is executed.
Returns: Maximum (first element) and minimum (second element) values.
Return type: tuple
- X (list) – List containing image batches. That is, an element of the list
is of shape (N, W, H), where N: #samples, W: image width, H: image
height. Arrays of size (N, W, H, C) with C: #channels are also accepted.
If you only have a single batch (e.g.
-
pyphoon.app.preprocess.resize(X, size, ignore_last_axis=False)¶ Resizes the image according to size using `cv2.resize`_ with bilinear interpolation.
Parameters: - X (numpy.array) – Image of shape (N, W, H), where N: #samples, W: image width, H: image height.
- size (tuple) – Size to reshape images.
- ignore_last_axis (bool) – Set to True if images have dimensionality (W, H, 1).
Returns: List containing image batches with resized images. E.g. (2,2).
Return type: numpy.array
-
class
pyphoon.app.preprocess.ImagePreprocessor(reshape_mode)¶ Bases:
objectParent class for image preprocessing classes. This class does not implement any method, please refer to its child classes.
Parameters: reshape_mode – Use the mode according to your DL framework. Available modes:
- keras: Reshape to have an extra axis for Keras.
-
apply(X)¶ Applies the preprocessing pipeline to class attribute data.
Parameters: X (numpy array) – Array with images.
-
reshape(X)¶ Reshapes the dimensions of the list so that it is suitable for the specified DL framework. So far, only ‘keras’ option is available.
Parameters: X (numpy.ndarray) – List of images. Returns: List of reshaped images Return type: list
-
class
pyphoon.app.preprocess.DefaultImagePreprocessor(mean, std, reshape_mode, resize_factor=None)¶ Bases:
pyphoon.app.preprocess.ImagePreprocessorChild of of
ImagePreprocessor. Assuming an input image \(X\), this preprocessor first centres and normalises it as\[\frac{X-\mu}{\sigma}\]where \(\mu\) and \(\sigma\) denote the pixel mean and standard deviation, respectively. Next, it resizes the image using the method
resize().Variables: - resize_factor – To resize the image. For instance, half the dimensions by setting this parameter equal to 2.
- mean – Used to centre the data.
- std – Used to normalise the data.
- reshape_mode – Used to normalise the data. See
ImagePreprocessor
-
apply(X)¶ Processes an array of images, scaling and normalising them as required and, eventually, reshapes the list to be suitable for a specific DL framework.
Parameters: X (numpy.ndarray) – List with image data (images as numpy arrays). Returns: Updated, preprocessed list of images.
-
class
pyphoon.app.preprocess.MeanImagePreprocessor(mean_image, scale_factor, reshape_mode, resize_factor=None)¶ Bases:
pyphoon.app.preprocess.ImagePreprocessorChild of
ImagePreprocessor. Assuming an input image \(X\), this preprocessor first centres and normalises it as\[\frac{X \ominus \hat{X}}{s}\]where \(\hat{X}\) denotes the image mean (same size as :math: X), \(s\) is the scale factor (scalar) and \(\ominus\) is pixel-wise subtraction operations. Next, it resizes the image using the method
resize().Variables: - mean_image – Mean image (2D matrix)
- scale_factor – Used to normalise the data.
- resize_factor – To resize the image. Define the new size of the images.
- reshape_mode – Used to normalise the data. See
ImagePreprocessor
-
apply(X)¶ Processes an array of images, scaling and normalising them as required and, eventually, reshapes the list to be suitable for a specific DL framework.
Parameters: X (numpy.ndarray) – List with image data (images as numpy arrays). Returns: Updated, preprocessed list of images.
pyphoon.app.utils¶
-
pyphoon.app.utils.read_h5datachunk_old(path_to_file, shuffle=False)¶ Reads a chunk of data stored as h5.
Parameters: - path_to_file (str) – Path name to the H5 file to read
- shuffle (bool) – Set to true if data should be shuffled.
Returns: Array of images (X), array of labels (Y)
Return type: list
-
pyphoon.app.utils.load_h5datachunks(dataset_dir, chunk_filenames, features, ignore_classes=None, display=False)¶ Loads a set of h5 files as individual arrays in a list.
Parameters: - dataset_dir (str) – Directory containing the chunk files.
- chunk_filenames (list) – Filenames of the data chunks.
- features (list) – Features to retrieve from the h5 data chunks as string names.
- ignore_classes (list, default None) – List of class labels to consider. Labels as ints. By default it considers all classes.
- display (bool, default False) – Set to True to have some informative messages printed out.
Returns: List with the data chunks as numpy.arrays.
Return type: list
-
pyphoon.app.utils.data_generator_from_chunklist(X, Y, batch_sz)¶ Generates batches of data from samples X and labels Y.
Parameters: - X (list) – Sample data.
- Y (list) – Label data.
- batch_sz (int) – Batch size.
Returns: Generator of batches of samples, labels and weights (importance of samples).
Return type: tuple
pyphoon.app.callbacks¶
-
class
pyphoon.app.callbacks.StoreModelWeights(filepath, monitor='val_loss', verbose=0, save_best_only=True, save_weights_only=False, mode='auto', period=1)¶ Bases:
keras.callbacks.ModelCheckpointKeras callback used to store the model weights after each epoch. More details in keras documentation.
Parameters: - filepath (str) – Folder to store the weights HDF5 files.
- monitor (str, default 'val_loss') – Model metric to rely on when storing the weights.
- verbose (int) – Set to 1 to display execution information
- save_best_only – Set to True to only store weights when model performance (according to metric monitor) improves.
- save_weights_only (bool, default False) – Set to True to only store the weights.
- mode (str, default 'auto') –
- period (int, default 1) –
-
class
pyphoon.app.callbacks.PlotRegressionValidation(X, Y)¶ Bases:
keras.callbacks.CallbackPlot a scatterplot illustrating the ground truth values and the network predictions after each epoch.
Parameters: X – Batch of samples. Shape (N, W, H, C), where N: #samples, W: width, H: height and C: #channels. :type X: numpy.array :param Y: Target values. :type Y: numpy.array
-
plot_regression(y_true, y_pred, show=False, save=False, filename='none')¶ Plot scatterplot with network estimation and ground truth values.
Parameters: - y_true (numpy.array) – Ground truth values.
- y_pred (numpy.array) – Network estimation values.
- show (bool) – Set to True to plot a figure.
- save (bool) – Set to True to store the image.
- filename (str) – Name of the file if image is stored.
-
on_epoch_end(epoch, logs={})¶ Executed after each epoch.
Parameters: - epoch – Current epoch.
- logs –
Returns:
-