Coverage for biobb_ml/neural_networks/regression_neural_network.py: 85%
180 statements
« prev ^ index » next coverage.py v7.5.1, created at 2024-05-07 09:39 +0000
« prev ^ index » next coverage.py v7.5.1, created at 2024-05-07 09:39 +0000
1#!/usr/bin/env python3
3"""Module containing the RegressionNeuralNetwork class and the command line interface."""
4import argparse
5import h5py
6import json
7import numpy as np
8import pandas as pd
9from biobb_common.generic.biobb_object import BiobbObject
10from tensorflow.python.keras.saving import hdf5_format
11from sklearn.preprocessing import scale
12from sklearn.model_selection import train_test_split
13from sklearn.metrics import r2_score
14from tensorflow.keras import Sequential
15from tensorflow.keras.layers import Dense
16from tensorflow.keras.callbacks import EarlyStopping
17from biobb_common.configuration import settings
18from biobb_common.tools import file_utils as fu
19from biobb_common.tools.file_utils import launchlogger
20from biobb_ml.neural_networks.common import check_input_path, check_output_path, getHeader, getTargetValue, plotResultsReg, getFeatures, getIndependentVarsList, getTarget, getWeight
23class RegressionNeuralNetwork(BiobbObject):
24 """
25 | biobb_ml RegressionNeuralNetwork
26 | Wrapper of the TensorFlow Keras Sequential method for regression.
27 | Trains and tests a given dataset and save the complete model for a Neural Network Regression. Visit the `Sequential documentation page <https://www.tensorflow.org/api_docs/python/tf/keras/Sequential>`_ in the TensorFlow Keras official website for further information.
29 Args:
30 input_dataset_path (str): Path to the input dataset. File type: input. `Sample file <https://github.com/bioexcel/biobb_ml/raw/master/biobb_ml/test/data/neural_networks/dataset_regression.csv>`_. Accepted formats: csv (edam:format_3752).
31 output_model_path (str): Path to the output model file. File type: output. `Sample file <http://github.com/bioexcel/biobb_ml/raw/master/biobb_ml/test/reference/neural_networks/ref_output_model_regression.h5>`_. Accepted formats: h5 (edam:format_3590).
32 output_test_table_path (str) (Optional): Path to the test table file. File type: output. `Sample file <https://github.com/bioexcel/biobb_ml/raw/master/biobb_ml/test/reference/neural_networks/ref_output_test_regression.csv>`_. Accepted formats: csv (edam:format_3752).
33 output_plot_path (str) (Optional): Loss, MAE and MSE plots. File type: output. `Sample file <https://github.com/bioexcel/biobb_ml/raw/master/biobb_ml/test/reference/neural_networks/ref_output_plot_regression.png>`_. Accepted formats: png (edam:format_3603).
34 properties (dic - Python dictionary object containing the tool parameters, not input/output files):
35 * **features** (*dict*) - ({}) Independent variables or columns from your dataset you want to train. You can specify either a list of columns names from your input dataset, a list of columns indexes or a range of columns indexes. Formats: { "columns": ["column1", "column2"] } or { "indexes": [0, 2, 3, 10, 11, 17] } or { "range": [[0, 20], [50, 102]] }. In case of mulitple formats, the first one will be picked.
36 * **target** (*dict*) - ({}) Dependent variable you want to predict from your dataset. You can specify either a column name or a column index. Formats: { "column": "column3" } or { "index": 21 }. In case of mulitple formats, the first one will be picked.
37 * **weight** (*dict*) - ({}) Weight variable from your dataset. You can specify either a column name or a column index. Formats: { "column": "column3" } or { "index": 21 }. In case of mulitple formats, the first one will be picked.
38 * **validation_size** (*float*) - (0.2) [0~1|0.05] Represents the proportion of the dataset to include in the validation split. It should be between 0.0 and 1.0.
39 * **test_size** (*float*) - (0.1) [0~1|0.05] Represents the proportion of the dataset to include in the test split. It should be between 0.0 and 1.0.
40 * **hidden_layers** (*list*) - (None) List of dictionaries with hidden layers values. Format: [ { 'size': 50, 'activation': 'relu' } ].
41 * **output_layer_activation** (*string*) - ("softmax") Activation function to use in the output layer. Values: sigmoid (Sigmoid activation function: sigmoid[x] = 1 / [1 + exp[-x]]), tanh (Hyperbolic tangent activation function), relu (Applies the rectified linear unit activation function), softmax(Softmax converts a real vector to a vector of categorical probabilities).
42 * **optimizer** (*string*) - ("Adam") Name of optimizer instance. Values: Adadelta (Adadelta optimization is a stochastic gradient descent method that is based on adaptive learning rate per dimension to address two drawbacks: the continual decay of learning rates throughout training and the need for a manually selected global learning rate), Adagrad (Adagrad is an optimizer with parameter-specific learning rates; which are adapted relative to how frequently a parameter gets updated during training. The more updates a parameter receives; the smaller the updates), Adam (Adam optimization is a stochastic gradient descent method that is based on adaptive estimation of first-order and second-order moments), Adamax (It is a variant of Adam based on the infinity norm. Default parameters follow those provided in the paper. Adamax is sometimes superior to adam; specially in models with embeddings), Ftrl (Optimizer that implements the FTRL algorithm), Nadam (Much like Adam is essentially RMSprop with momentum; Nadam is Adam with Nesterov momentum), RMSprop (Optimizer that implements the RMSprop algorithm), SGD (Gradient descent -with momentum- optimizer).
43 * **learning_rate** (*float*) - (0.02) [0~100|0.01] Determines the step size at each iteration while moving toward a minimum of a loss function
44 * **batch_size** (*int*) - (100) [0~1000|1] Number of samples per gradient update.
45 * **max_epochs** (*int*) - (100) [0~1000|1] Number of epochs to train the model. As the early stopping is enabled, this is a maximum.
46 * **random_state** (*int*) - (5) [1~1000|1] Controls the shuffling applied to the data before applying the split. .
47 * **scale** (*bool*) - (False) Whether or not to scale the input dataset.
48 * **remove_tmp** (*bool*) - (True) [WF property] Remove temporal files.
49 * **restart** (*bool*) - (False) [WF property] Do not execute if output files exist.
51 Examples:
52 This is a use example of how to use the building block from Python::
54 from biobb_ml.neural_networks.regression_neural_network import regression_neural_network
55 prop = {
56 'features': {
57 'columns': [ 'column1', 'column2', 'column3' ]
58 },
59 'target': {
60 'column': 'target'
61 },
62 'validation_size': 0.2,
63 'test_size': .33,
64 'hidden_layers': [
65 {
66 'size': 10,
67 'activation': 'relu'
68 },
69 {
70 'size': 8,
71 'activation': 'relu'
72 }
73 ],
74 'optimizer': 'Adam',
75 'learning_rate': 0.01,
76 'batch_size': 32,
77 'max_epochs': 150
78 }
79 regression_neural_network(input_dataset_path='/path/to/myDataset.csv',
80 output_model_path='/path/to/newModel.h5',
81 output_test_table_path='/path/to/newTable.csv',
82 output_plot_path='/path/to/newPlot.png',
83 properties=prop)
85 Info:
86 * wrapped_software:
87 * name: TensorFlow Keras Sequential
88 * version: >2.1.0
89 * license: MIT
90 * ontology:
91 * name: EDAM
92 * schema: http://edamontology.org/EDAM.owl
94 """
96 def __init__(self, input_dataset_path,
97 output_model_path, output_test_table_path=None, output_plot_path=None, properties=None, **kwargs) -> None:
98 properties = properties or {}
100 # Call parent class constructor
101 super().__init__(properties)
102 self.locals_var_dict = locals().copy()
104 # Input/Output files
105 self.io_dict = {
106 "in": {"input_dataset_path": input_dataset_path},
107 "out": {"output_model_path": output_model_path, "output_test_table_path": output_test_table_path, "output_plot_path": output_plot_path}
108 }
110 # Properties specific for BB
111 self.features = properties.get('features', {})
112 self.target = properties.get('target', {})
113 self.weight = properties.get('weight', {})
114 self.validation_size = properties.get('validation_size', 0.1)
115 self.test_size = properties.get('test_size', 0.1)
116 self.hidden_layers = properties.get('hidden_layers', [])
117 self.output_layer_activation = properties.get('output_layer_activation', 'softmax')
118 self.optimizer = properties.get('optimizer', 'Adam')
119 self.learning_rate = properties.get('learning_rate', 0.02)
120 self.batch_size = properties.get('batch_size', 100)
121 self.max_epochs = properties.get('max_epochs', 100)
122 self.random_state = properties.get('random_state', 5)
123 self.scale = properties.get('scale', False)
124 self.properties = properties
126 # Check the properties
127 self.check_properties(properties)
128 self.check_arguments()
130 def check_data_params(self, out_log, err_log):
131 """ Checks all the input/output paths and parameters """
132 self.io_dict["in"]["input_dataset_path"] = check_input_path(self.io_dict["in"]["input_dataset_path"], "input_dataset_path", False, out_log, self.__class__.__name__)
133 self.io_dict["out"]["output_model_path"] = check_output_path(self.io_dict["out"]["output_model_path"], "output_model_path", False, out_log, self.__class__.__name__)
134 self.io_dict["out"]["output_test_table_path"] = check_output_path(self.io_dict["out"]["output_test_table_path"], "output_test_table_path", True, out_log, self.__class__.__name__)
135 self.io_dict["out"]["output_plot_path"] = check_output_path(self.io_dict["out"]["output_plot_path"], "output_plot_path", True, out_log, self.__class__.__name__)
137 def build_model(self, input_shape):
138 """ Builds Neural network according to hidden_layers property """
140 # create model
141 model = Sequential([])
143 # if no hidden_layers provided, create manually a hidden layer with default values
144 if not self.hidden_layers:
145 self.hidden_layers = [{'size': 50, 'activation': 'relu'}]
147 # generate hidden_layers
148 for i, layer in enumerate(self.hidden_layers):
149 if i == 0:
150 model.add(Dense(layer['size'], activation=layer['activation'], kernel_initializer='he_normal', input_shape=input_shape)) # 1st hidden layer
151 else:
152 model.add(Dense(layer['size'], activation=layer['activation'], kernel_initializer='he_normal'))
154 model.add(Dense(1)) # output layer
156 return model
158 @launchlogger
159 def launch(self) -> int:
160 """Execute the :class:`RegressionNeuralNetwork <neural_networks.regression_neural_network.RegressionNeuralNetwork>` neural_networks.regression_neural_network.RegressionNeuralNetwork object."""
162 # check input/output paths and parameters
163 self.check_data_params(self.out_log, self.err_log)
165 # Setup Biobb
166 if self.check_restart():
167 return 0
168 self.stage_files()
170 # load dataset
171 fu.log('Getting dataset from %s' % self.io_dict["in"]["input_dataset_path"], self.out_log, self.global_log)
172 if 'columns' in self.features:
173 labels = getHeader(self.io_dict["in"]["input_dataset_path"])
174 skiprows = 1
175 else:
176 labels = None
177 skiprows = None
178 data = pd.read_csv(self.io_dict["in"]["input_dataset_path"], header=None, sep="\\s+|;|:|,|\t", engine="python", skiprows=skiprows, names=labels)
180 X = getFeatures(self.features, data, self.out_log, self.__class__.__name__)
181 fu.log('Features: [%s]' % (getIndependentVarsList(self.features)), self.out_log, self.global_log)
182 # target
183 y = getTarget(self.target, data, self.out_log, self.__class__.__name__)
184 fu.log('Target: %s' % (str(getTargetValue(self.target))), self.out_log, self.global_log)
185 # weights
186 if self.weight:
187 w = getWeight(self.weight, data, self.out_log, self.__class__.__name__)
189 # shuffle dataset
190 fu.log('Shuffling dataset', self.out_log, self.global_log)
191 shuffled_indices = np.arange(X.shape[0])
192 np.random.shuffle(shuffled_indices)
193 np_X = X.to_numpy()
194 shuffled_X = np_X[shuffled_indices]
195 shuffled_y = y[shuffled_indices]
196 if self.weight:
197 shuffled_w = w[shuffled_indices]
199 # train / test split
200 fu.log('Creating train and test sets', self.out_log, self.global_log)
201 arrays_sets = (shuffled_X, shuffled_y)
202 # if user provide weights
203 if self.weight:
204 arrays_sets = arrays_sets + (shuffled_w,)
205 X_train, X_test, y_train, y_test, w_train, w_test = train_test_split(*arrays_sets, test_size=self.test_size, random_state=self.random_state)
206 else:
207 X_train, X_test, y_train, y_test = train_test_split(*arrays_sets, test_size=self.test_size, random_state=self.random_state)
209 # scale dataset
210 if self.scale:
211 fu.log('Scaling dataset', self.out_log, self.global_log)
212 X_train = scale(X_train)
214 # build model
215 fu.log('Building model', self.out_log, self.global_log)
216 model = self.build_model((X_train.shape[1],))
218 # model summary
219 stringlist = []
220 model.summary(print_fn=lambda x: stringlist.append(x))
221 model_summary = "\n".join(stringlist)
222 fu.log('Model summary:\n\n%s\n' % model_summary, self.out_log, self.global_log)
224 # get optimizer
225 mod = __import__('tensorflow.keras.optimizers', fromlist=[self.optimizer])
226 opt_class = getattr(mod, self.optimizer)
227 opt = opt_class(lr=self.learning_rate)
228 # compile model
229 model.compile(optimizer=opt, loss='mse', metrics=['mae', 'mse'], sample_weight_mode='samplewise')
231 # fitting
232 fu.log('Training model', self.out_log, self.global_log)
233 # set an early stopping mechanism
234 # set patience=2, to be a bit tolerant against random validation loss increases
235 early_stopping = EarlyStopping(patience=2)
237 if self.weight:
238 sample_weight = w_train
239 class_weight = []
240 else:
241 # TODO: class_weight not working since TF 2.4.1 update
242 # fu.log('No weight provided, class_weight will be estimated from the target data', self.out_log, self.global_log)
243 sample_weight = None
244 class_weight = [] # compute_class_weight('balanced', np.unique(y_train), y_train)
246 # fit the model
247 mf = model.fit(X_train,
248 y_train,
249 class_weight=class_weight,
250 sample_weight=sample_weight,
251 batch_size=self.batch_size,
252 epochs=self.max_epochs,
253 callbacks=[early_stopping],
254 validation_split=self.validation_size,
255 verbose=1)
257 fu.log('Total epochs performed: %s' % len(mf.history['loss']), self.out_log, self.global_log)
259 # predict data from X_train
260 train_predictions = model.predict(X_train)
261 train_predictions = np.around(train_predictions, decimals=2)
263 score_train_inputs = (y_train, train_predictions)
264 if self.weight:
265 score_train_inputs = score_train_inputs + (w_train,)
266 train_score = r2_score(*score_train_inputs)
268 train_metrics = pd.DataFrame()
269 train_metrics['metric'] = ['Train loss', 'Train MAE', 'Train MSE', 'Train R2', 'Validation loss', 'Validation MAE', 'Validation MSE']
270 train_metrics['coefficient'] = [mf.history['loss'][-1], mf.history['mae'][-1], mf.history['mse'][-1], train_score, mf.history['val_loss'][-1], mf.history['val_mae'][-1], mf.history['val_mse'][-1]]
272 fu.log('Training metrics\n\nTRAINING METRICS TABLE\n\n%s\n' % train_metrics, self.out_log, self.global_log)
274 # testing
275 if self.scale:
276 X_test = scale(X_test)
277 fu.log('Testing model', self.out_log, self.global_log)
278 test_loss, test_mae, test_mse = model.evaluate(X_test, y_test)
280 # predict data from X_test
281 test_predictions = model.predict(X_test)
282 test_predictions = np.around(test_predictions, decimals=2)
283 tpr = np.squeeze(np.asarray(test_predictions))
285 score_test_inputs = (y_test, test_predictions)
286 if self.weight:
287 score_test_inputs = score_test_inputs + (w_test,)
288 score = r2_score(*score_test_inputs)
290 test_metrics = pd.DataFrame()
291 test_metrics['metric'] = ['Test loss', 'Test MAE', 'Test MSE', 'Test R2']
292 test_metrics['coefficient'] = [test_loss, test_mae, test_mse, score]
294 fu.log('Testing metrics\n\nTESTING METRICS TABLE\n\n%s\n' % test_metrics, self.out_log, self.global_log)
296 test_table = pd.DataFrame()
297 test_table['prediction'] = tpr
298 test_table['target'] = y_test
299 test_table['residual'] = test_table['target'] - test_table['prediction']
300 test_table['difference %'] = np.absolute(test_table['residual']/test_table['target']*100)
301 pd.set_option('display.float_format', lambda x: '%.2f' % x)
302 # sort by difference in %
303 test_table = test_table.sort_values(by=['difference %'])
304 test_table = test_table.reset_index(drop=True)
305 fu.log('TEST DATA\n\n%s\n' % test_table, self.out_log, self.global_log)
307 # save test data
308 if (self.io_dict["out"]["output_test_table_path"]):
309 fu.log('Saving testing data to %s' % self.io_dict["out"]["output_test_table_path"], self.out_log, self.global_log)
310 test_table.to_csv(self.io_dict["out"]["output_test_table_path"], index=False, header=True)
312 # create test plot
313 if (self.io_dict["out"]["output_plot_path"]):
314 fu.log('Saving plot to %s' % self.io_dict["out"]["output_plot_path"], self.out_log, self.global_log)
315 test_predictions = test_predictions.flatten()
316 train_predictions = model.predict(X_train).flatten()
317 plot = plotResultsReg(mf.history, y_test, test_predictions, y_train, train_predictions)
318 plot.savefig(self.io_dict["out"]["output_plot_path"], dpi=150)
320 # save model and parameters
321 vars_obj = {
322 'features': self.features,
323 'target': self.target,
324 'scale': self.scale,
325 'type': 'regression'
326 }
327 variables = json.dumps(vars_obj)
328 fu.log('Saving model to %s' % self.io_dict["out"]["output_model_path"], self.out_log, self.global_log)
329 with h5py.File(self.io_dict["out"]["output_model_path"], mode='w') as f:
330 hdf5_format.save_model_to_hdf5(model, f)
331 f.attrs['variables'] = variables
333 # Copy files to host
334 self.copy_to_host()
336 self.tmp_files.extend([
337 self.stage_io_dict.get("unique_dir")
338 ])
339 self.remove_tmp_files()
341 self.check_arguments(output_files_created=True, raise_exception=False)
343 return 0
346def regression_neural_network(input_dataset_path: str, output_model_path: str, output_test_table_path: str = None, output_plot_path: str = None, properties: dict = None, **kwargs) -> int:
347 """Execute the :class:`RegressionNeuralNetwork <neural_networks.regression_neural_network.RegressionNeuralNetwork>` class and
348 execute the :meth:`launch() <neural_networks.regression_neural_network.RegressionNeuralNetwork.launch>` method."""
350 return RegressionNeuralNetwork(input_dataset_path=input_dataset_path,
351 output_model_path=output_model_path,
352 output_test_table_path=output_test_table_path,
353 output_plot_path=output_plot_path,
354 properties=properties, **kwargs).launch()
357def main():
358 """Command line execution of this building block. Please check the command line documentation."""
359 parser = argparse.ArgumentParser(description="Wrapper of the TensorFlow Keras Sequential method.", formatter_class=lambda prog: argparse.RawTextHelpFormatter(prog, width=99999))
360 parser.add_argument('--config', required=False, help='Configuration file')
362 # Specific args of each building block
363 required_args = parser.add_argument_group('required arguments')
364 required_args.add_argument('--input_dataset_path', required=True, help='Path to the input dataset. Accepted formats: csv.')
365 required_args.add_argument('--output_model_path', required=True, help='Path to the output model file. Accepted formats: h5.')
366 parser.add_argument('--output_test_table_path', required=False, help='Path to the test table file. Accepted formats: csv.')
367 parser.add_argument('--output_plot_path', required=False, help='Loss, MAE and MSE plots. Accepted formats: png.')
369 args = parser.parse_args()
370 args.config = args.config or "{}"
371 properties = settings.ConfReader(config=args.config).get_prop_dic()
373 # Specific call of each building block
374 regression_neural_network(input_dataset_path=args.input_dataset_path,
375 output_model_path=args.output_model_path,
376 output_test_table_path=args.output_test_table_path,
377 output_plot_path=args.output_plot_path,
378 properties=properties)
381if __name__ == '__main__':
382 main()