Neural Network MLPClassifier API¶

Source code: https://bitbucket.org/kul-reseco/lnns/src.

For issues, bugs, proposals or remarks, visit the issue tracker.

Algorithms¶

class lnns.core.network.Network(number_of_hidden, activation)[source]¶

Bases: object

The Network class create a neural network using the sklearn.neural_network.MLPClassifier. The network can be used to predict classified images using supervised classification.

predict_image(band_data, probability_of_class=None)[source]¶

Use the trained network, it is possible to classify the complete image using the different bands of the image. If probability_of_class is given, than the image will show the probability that this class will occur.

Parameters:	band_data (np.array) – array with all bands probability_of_class (int) – class for which you would like the probability image
Returns:	np.array classified_image: the classified image

train_image(band_data, classes_data, max_iter, no_data_value=0, test_size=0.33, log_function=<built-in function print>)[source]¶

The given network can be trained given an image of different band waves (band_data) and a respectively data set indicating a subset of different classes of the image (class_data).

Parameters:

band_data (np.array) – array with all bands
classes_data (np.array) – an overlap image indicating different classes
max_iter (int) – the number of iterations when training the neural network
no_data_value (int) – value that describes pixels with no data in the classes_data file
test_size (float) – the proportion of the test_size that will be used to evaluate the trained network
log_function – function to log

validate(x_test, y_test, log_function=<built-in function print>)[source]¶

The trained neural network can be evaluated using a test set of the data. The validation results will be saved in the validation_results dictionary with the keys: Average accuracy, a Kappa for each predicted class, an average kappa and a rapport including precision, recall, f1-score and support.

Example output:

Average accuracy: 0.9740708729472775
Kappa class 1: 0.9996048676750681
Kappa class 2: 0.9969686283161031
Kappa class 3: 0.9269671354418852
Kappa class 4: 0.9384942730689072
Kappa class 5: 0.9854574554108237
Kappa class 6: 0.8991142021469177
Average Kappa: 0.9577677603432843

.	precision	recall	f1-score	support
0	1.00	1.00	1.00	1664
1	1.00	1.00	1.00	956
2	0.91	0.96	0.93	597
3	0.96	0.94	0.95	1149
4	1.00	0.98	0.99	2345
5	0.88	0.92	0.90	231
avg / total	0.98	0.98	0.98	6942

Parameters:	x_test (np.array[float]) – test input variables y_test (np.array[float]) – test output values log_function – function to log

Interfaces¶

———————————————————————————————————————-
Date                : August 2018
Copyright           : © 2018 - 2020 by Tinne Cahy (Geo Solutions) and Ann Crabbé (KU Leuven)
Email               : acrabbe.foss@gmail.com
Acknowledgements    : Translated from LNNS 1.0 A neural network simulator [C++ software]
Ghent University, Laboratory of Forest Management and Spatial Information Techniques
Lieven P.C. Verbeke

This file is part of the QGIS Neural Network MLP Classifier plugin and mlp-image-classifier python package.

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public
License as published by the Free Software Foundation, either version 3 of the License, or any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied
warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License (COPYING.txt). If not see www.gnu.org/licenses.
———————————————————————————————————————-

lnns.interfaces.plot(x, y, size=None, title='Neural Network MLPClassifier: Network error decline', testing=False)[source]¶

lnns.interfaces.plot_to_file(x, y, path, title='Neural Network MLPClassifier: Network error decline')[source]¶

lnns.interfaces.run_algorithm_images(image_paths, classified_path, no_data_value, hidden_layer_size, activation, iterations, test_size, probability_of_class, update_process_bar=<built-in function print>, log_function=<built-in function print>, output_path=None, plot_path=None, feedback=None, context=None)[source]¶

Process all input, in order to run the nn script.

Parameters:

image_paths – the absolute path to the input raster files
classified_path – absolute path to the classified raster file
no_data_value – value that describes pixels with no data in the classes_data file
hidden_layer_size – Hidden layer size with length = n_layers - 2, comma separated values.
activation – Activation function for the MLPClassifier, choices=[‘identity’, ‘logistic’, ‘tanh’, ‘relu’]
iterations – Maximum number of iterations
test_size – the proportion of the test_size that will be used to evaluate the trained network
probability_of_class – class for which you would like the probability image
update_process_bar – function to update the progress bar
log_function – function to log
output_path – path for output files (optional)
plot_path – necessary for the processing tool: path for the plot
feedback – necessary for the processing tool
context – necessary for the processing tool

Returns:

output path

lnns.interfaces.run_algorithm_pattern(pattern_train, pattern_predict, no_data_value, hidden_layer_size, activation, iterations, test_size, probability_of_class, log_function=<built-in function print>, output_path=None, plot_path=None, feedback=None, context=None)[source]¶

Process all input, in order to run the nn script.

Parameters:

pattern_train – the absolute path to the pattern file with training data
pattern_predict – absolute path to the pattern file for predictions
no_data_value – value that describes pixels with no data in the classes_data file
hidden_layer_size – Hidden layer size with length = n_layers - 2, comma separated values.
activation – Activation function for the MLPClassifier, choices=[‘identity’, ‘logistic’, ‘tanh’, ‘relu’]
iterations – Maximum number of iterations
test_size – the proportion of the test_size that will be used to evaluate the trained network
probability_of_class – class for which you would like the probability image
log_function – function to log
output_path – path for output files (optional)
plot_path – necessary for the processing tool: path for the plot
feedback – necessary for the processing tool
context – necessary for the processing tool

Returns:

output path

lnns.interfaces.tuple_int(s)[source]¶

lnns.interfaces.tuple_int_cli(s)[source]¶

lnns.interfaces.tuple_string(s)[source]¶

class lnns.interfaces.imports.PatternFile(number_of_patterns, input_neurons, output_neurons, x, y)[source]¶: Bases: object

lnns.interfaces.imports.check_path(path)[source]¶

Check if path exists. Skipp path which are in memory

Parameters:	path – the absolute path to the input file

lnns.interfaces.imports.import_image(path, reflectance=False)[source]¶

Browse for an image.

Parameters:	path – the absolute path to the image reflectance – return reflectance values instead of DN between 0 and 254
Returns:	float32 numpy array [#good bands x #rows x #columns]

lnns.interfaces.imports.read_pattern(path_prn)[source]¶

lnns.interfaces.exports.write_classified_image(file_path, image, geo_transform=None, projection=None, gdal_type=<Mock name='mock.gdal.GDT_Int16' id='140574537438160'>)[source]¶

lnns.interfaces.exports.write_pattern(output_file, number_of_patterns, input_neurons, output_neurons, array, fmt)[source]¶

CLI: mlpclassifier-image¶

The Network class create a neural network using the sklearn.neural_network.MLPClassifier. The network can be used to predict classified images using supervised classification.

usage: mlpclassifier-image [-h] [-n NO_DATA_VALUE] [-l HIDDEN_LAYER_SIZE]
                           [-a {identity,logistic,tanh,relu}] [-i ITERATIONS]
                           [-t TEST_SIZE] [-p PROBABILITY_OF_CLASS]
                           [-o OUTPUT] [-g OUTPUT_GRAPH]
                           images_folder image_names classes_data_name

Positional Arguments¶

`images_folder`	Path to the input image data and the classes data.
`image_names`	Name(s) of different image bands.
`classes_data_name`
	Name of an overlap image indicating different classes

Named Arguments¶

`-n, --no_data_value`
	Value that describes pixels with no data in the classes_data file (default: -1). Default: -1
`-l, --hidden_layer_size`
	Hidden layer size with length = n_layers - 2, comma separated values. The ith element represents the number of neurons in the ith hidden layer. (default: 10,) Default: (10,)
`-a, --activation`
	Possible choices: identity, logistic, tanh, relu Activation function for the MLPClassifier (default: logistic). Default: “logistic”
`-i, --iterations`
	Maximum number of iterations (default: 200). Default: 200
`-t, --test_size`
	Portion of test pixels used to evaluate the trained network (default: 0.33). Default: 0.33
`-p, --probability_of_class`
	Class for which you would like the probability image (default: None).
`-o, --output`	Output predicted file (default: in same folder with name ‘output_classified.tif’
`-g, --output_graph`
	Output error graph (default: in same folder with name ‘output_error.PNG’

CLI: mlpclassifier-pattern¶

The Network class create a neural network using the sklearn.neural_network.MLPClassifier. The network can be used to predict classified images using supervised classification.

usage: mlpclassifier-pattern [-h] [-n NO_DATA_VALUE] [-l HIDDEN_LAYER_SIZE]
                             [-a {identity,logistic,tanh,relu}]
                             [-i ITERATIONS] [-t TEST_SIZE]
                             [-p PROBABILITY_OF_CLASS] [-o OUTPUT]
                             [-g OUTPUT_GRAPH]
                             pattern_predict_path pattern_train_path

Positional Arguments¶

`pattern_predict_path`
	Pattern text file that includes the values to predict,’number_of_patterns’, ‘number_of_inputs’ and ‘number_of_outputs.’
`pattern_train_path`
	Pattern text file that includes the network training values, ‘number_of_patterns’, ‘number_of_inputs’ and ‘number_of_outputs.’

Named Arguments¶

`-n, --no_data_value`
	Value that describes pixels with no data in the classes_data file (default: -1). Default: -1
`-l, --hidden_layer_size`
	Hidden layer size with length = n_layers - 2, comma separated values. The ith element represents the number of neurons in the ith hidden layer. (default: 10,) Default: (10,)
`-a, --activation`
	Possible choices: identity, logistic, tanh, relu Activation function for the MLPClassifier (default: logistic). Default: “logistic”
`-i, --iterations`
	Maximum number of iterations (default: 200). Default: 200
`-t, --test_size`
	Portion of test pixels used to evaluate the trained network (default: 0.33). Default: 0.33
`-p, --probability_of_class`
	class for which you would like the probability image (default: None).
`-o, --output`	Output predicted file (default: in same folder with extension ‘_predict.prn’
`-g, --output_graph`
	Output error graph (default: in same folder with extension ‘_error.PNG’