Function reconstruction

refann.refann

class refann.refann.ANN(data, hidden_layer=1, mid_node=4096, hp_model='rec_1', loss_func='L1')

Bases: refann.train.Train

Reconstruct functions with Artificial Neural Network.

Parameters:

• data (array-like) – An array with shape of (N, 3), where N is the number of data points, each column represents \(X, Y, \sigma_Y\).
• hidden_layer (int, optional) – The number of hidden layers. Default: 1
• mid_node (int, optional) – The number of nodes (or neurons) of the middle layer. Default: 4096
• hp_model (int, optional) – The hyperparameter models, ‘rec_1’ (no batch normalization) or ‘rec_2’ (with batch normalization). Default: ‘rec_1’
• loss_func (str, optional) – The loss function, there are three loss functions in this code, L1Loss (‘L1’), MSELoss (‘MSE’), and SmoothL1Loss (‘SmoothL1’). Default: ‘L1’

Variables:

• lr (float, optional) – The learning rate. Default: 1e-1
• lr_min (float, optional) – The minimum of the learning rate. Default: 1e-8
• iteration (int, optional) – The number of iterations. Default: 30000
• batch_size_max (int, optional) – The maximum of the batch size. Defalt: 300
• scale_inputs (bool, optional) – If True, the inputs data will be normalized, otherwise, do nothing. Defalt: True
• scale_target (bool, optional) – If True, the outputs (or target) data will be normalized, otherwise, do nothing. Defalt: True
• scale_type (str, optional) – The normalization method, ‘minmax’, ‘mean’, or ‘z_score’. Default: ‘z_score’
• fix_initialize (bool, optional) – If True, the network will be initialized from a specific seed. Default: True
• print_info (bool, optional) – If True, some information about the training process will be printed. Default: True

Note

Hyperparameters of the ANN, such as the number of hidden layers (hidden_layer), the number of neurons (mid_node), hyperparameter model (hp_model), should be optimized before reconstructing functions from data. See https://doi.org/10.3847/1538-4365/ab620b for details.

plot_func()

plot_loss()

predict(xpoint=None, xspace=None)

Prediction

Parameters:

• xpoint (array-like or None, optional) – An array of \(X\) points. Default: None
• xspace (tuple, list, or None, optional) – If not None, xpoint will be ignored, and it should be (xmin, xmax, npoint) or [xmin, xmax, npoint]. Default: None

Returns:

The reconstructed function.

Return type:

array-like

save_func(path='func', obsName='Hz', file_type='npy')

save_net(path='func', obsName='Hz')

statistic()

train()

class refann.refann.OptimizeANN(truth, hidden_layers=[1, 2, 3], mid_nodes=[128, 256])

Bases: object

get_optimal()

get_risk(path='', obsName='Hz', file_type='npy')

plot_risk()

risk(predict, error)

rss(predict)

class refann.refann.RePredictANN(data, hidden_layer=1, mid_node=4096)

Bases: refann.refann.ANN

Reconstruct function using the saved well-trained networks

Parameters:

• data (array-like) – An array with shape of (N, 3), where N is the number of data points, each column represents \(X, Y, \sigma_Y\).
• hidden_layer (int, optional) – The number of hidden layers. Default: 1
• mid_node (int, optional) – The number of nodes (or neurons) of the middle layer. Default: 4096

Variables:

• scale_inputs (bool, optional) – If True, the inputs data will be normalized, otherwise, do nothing. Defalt: True
• scale_target (bool, optional) – If True, the outputs (or target) data will be normalized, otherwise, do nothing. Defalt: True
• scale_type (str, optional) – The normalization method, ‘minmax’, ‘mean’, or ‘z_score’. Default: ‘z_score’

load_net(path='func', obsName='Hz')

refann.evaluate

refann.evaluate.plot_loss(loss)

refann.evaluate.predict(net, inputs, use_GPU=False, in_type='numpy')

Make predictions using a well-trained network.

Parameters:

• inputs (numpy array or torch tensor) – The inputs of the network.
• use_GPU (bool) – If True, calculate using GPU, otherwise, calculate using CPU.
• in_type (str) – The data type of the inputs, it can be ‘numpy’ or ‘torch’.

refann.save

refann.save.mkdir(path)

Make a directory in a particular location if it is not exists, otherwise, do nothing.

Parameters:path (str) – The path of a file.

Examples

>>> mkdir('/home/UserName/test')
>>> mkdir('test/one')
>>> mkdir('../test/one')

refann.save.savenpy(path, FileName, File, dtype=<type 'numpy.float32'>)

Save an array to a binary file in .npy format using numpy.save() function.

Parameters:

• path (str) – The path of the file to be saved.
• FileName (str) – The name of the file to be saved.
• File (object) – The file to be saved.
• dtype (str or object) – The type of the data to be saved. Default: numpy.float32.

refann.save.savetxt(path, FileName, File)

Save the .txt files using numpy.savetxt() funtion.

Parameters:

• path (str) – The path of the file to be saved.
• FileName (str) – The name of the file to be saved.
• File (object) – The file to be saved.