astronomaly.feature_extraction package¶

Submodules¶

astronomaly.feature_extraction.autoencoder module¶

class astronomaly.feature_extraction.autoencoder.Autoencoder(model_file='')¶

Bases: object

compile_autoencoder_model(input_image_shape)¶

Compiles the default autoencoder model. Note this model is designed to operate on 128x128 images. While it can run on different size images this can dramatically change the size of the final feature space.

Parameters:	input_image_shape (tuple) – The expected shape of the input images. Can either be length 2 or 3 (to include number of channels).

encode(images)¶

Returns the deep encoded features for an array of images.

Parameters:	images (np.ndarray) – Input images (nobjects x image_shape). For a single image, provide [image] as an array is expected.
Returns:	Deep features (nobjects x nfeatures)
Return type:	np.ndarray

fit(training_data, batch_size=32, epochs=10)¶

Actually train the autoencoder.

Parameters:

training_data (np.ndarray, list) – Either array or list of images. It’s recommended that this data be augmented with translation or rotation (or both).
batch_size (int, optional) – Number of samples used to update weights in each iteration. A larger batch size can be more accurate but requires more memory and is slower to train.
epochs (int, optional) – Number of full passes through the entire training set.

save(filename)¶

Saves Keras model in HDF5 format

Parameters:	filename (string) – Location for saved model

shape_check(images)¶

Convenience function to reshape images appropriate for deep learning.

Parameters:	images (np.ndarray, list) – Array of list of images
Returns:	Converted array compliant with CNN
Return type:	np.ndarray

class astronomaly.feature_extraction.autoencoder.AutoencoderFeatures(training_dataset=None, retrain=False, **kwargs)¶: Bases: astronomaly.base.base_pipeline.PipelineStage

astronomaly.feature_extraction.feets_features module¶

astronomaly.feature_extraction.flatten_features module¶

class astronomaly.feature_extraction.flatten_features.Flatten_Features(**kwargs)¶: Bases: astronomaly.base.base_pipeline.PipelineStage

astronomaly.feature_extraction.flux_histogram module¶

astronomaly.feature_extraction.photutils_features module¶

astronomaly.feature_extraction.power_spectrum module¶

class astronomaly.feature_extraction.power_spectrum.PSD_Features(nbins='auto', **kwargs)¶: Bases: astronomaly.base.base_pipeline.PipelineStage

astronomaly.feature_extraction.power_spectrum.psd_2d(img, nbins)¶

Computes the power spectral density for an input image. Translation and rotation invariate features.

Parameters:	img (np.ndarray) – Input image nbins (int) – Number of frequency bins to use. Frequency will range from 1 pixel to the largest axis of the input image, measured in pixels.
Returns:	Power spectral density at each frequency
Return type:	np.ndarray

astronomaly.feature_extraction.shape_features module¶

astronomaly.feature_extraction.wavelet_features module¶

class astronomaly.feature_extraction.wavelet_features.WaveletFeatures(level=2, wavelet_family='sym2', **kwargs)¶: Bases: astronomaly.base.base_pipeline.PipelineStage

astronomaly.feature_extraction.wavelet_features.flatten_swt2_coefficients(wavelet_coeffs)¶

A standardised way of flattening the swt2d coefficients They are stored as n_levels -> (cA, (cH, cV, cD)) where each of the sets of coeffcients has a list of [npixels, npixels]

Parameters:	wavelet_coeffs (list) – Exactly as output by pywt
Returns:	np.ndarray – Flattened coefficients labels – The labels of the coefficients

astronomaly.feature_extraction.wavelet_features.generate_labels(wavelet_coeffs)¶: Because the number of features may not be known till runtime, we can only create the labels of these features at runtime.

astronomaly.feature_extraction.wavelet_features.reshape_swt2_coefficients(flat_coeffs, nlev, image_shape)¶

Inverse function to restore a flattened array to pywt structure.

Parameters:	flat_coeffs (np.ndarray) – Flattened array of coefficients nlev (int) – Number of levels wavelet decomposition was performed with image_shape (tuple) – Shape of original images
Returns:	pywt compatible coefficient structure
Return type:	list

astronomaly.feature_extraction.wavelet_features.wavelet_decomposition(img, level=2, wavelet_family='sym2')¶

Perform wavelet decomposition on single image