Added code for creating, training and evaluating RNNs.

RNNHandler.py

@@ -0,0 +1,126 @@
+import keras.utils
+from keras.models import Sequential, model_from_json, load_model
+from keras.layers import Dense, Activation, SimpleRNN
+from keras.utils.visualize_util import plot
+import keras.utils.np_utils
+from keras.utils.np_utils import to_categorical
+
+import os
+import numpy as np
+import datetime
+import random
+from reader import Reader
+from sklearn import metrics
+import matplotlib.pyplot as plt
+from sklearn import metrics
+
+def MyMetrics(y_true, y_pred):
+	y_pred[y_pred<0.5] = 0
+	y_pred[y_pred>=0.5] = 1
+	if np.count_nonzero(y_pred == 1) == y_pred.shape[0]:
+		y_pred[0]=0
+	if np.count_nonzero(y_pred == 0) == y_pred.shape[0]:
+		y_pred[0]= 1
+	if np.count_nonzero(y_true == 1) == y_true.shape[0]:
+		y_true[0]=0
+	if np.count_nonzero(y_true == 0) == y_true.shape[0]:
+		y_true[0]= 1
+	confusion = metrics.confusion_matrix(y_true, y_pred)
+	TP = confusion[1, 1]
+	TN = confusion[0, 0]
+	FP = confusion[0, 1]
+	FN = confusion[1, 0]
+	precision = TP / (TP + FP)
+	recall = TP / (TP + FN)
+	fscore = 2 * precision * recall / ( precision + recall)
+	return (precision, recall, fscore)
+
+
+class RNNHandler:
+
+	results_directory = '/home/orestis/net/RNNresults'
+	models_directory = '/home/orestis/net/RNNmodels'
+
+	def __init__(self, model_name, num_categories, loss, optimizer):
+		# GET THE MODEL
+		fp_model = open(os.path.join(self.models_directory, model_name + '.json'), 'r')
+		model_str = fp_model.read()
+		self.model = model_from_json(model_str)
+		self.model.compile(loss=loss, optimizer=optimizer, metrics=['accuracy'])		
+		fp_model.close()
+
+		self.model_name = model_name
+		self.num_categories = num_categories
+		self.loss = loss
+		self.optimizer = optimizer
+
+
+	def fit_and_eval(self, x_train, y_train, x_test, y_test, nb_epoch, dataset_name):	#batch_size is always 1 and shuffle is always False, so we don't pass them as parameters
+		self.results_file = os.path.join(self.results_directory, self.model_name + '.' + dataset_name)
+		self.write_result('Model:' + self.model_name + ' Dataset:' + dataset_name + ' Loss:' + self.loss + ' Optimizer:' + self.optimizer + ' Dropout:No')
+		self.write_result('Epoch|Loss|Accuracy|Precision|Recall|Fscore')		
+		res_loss = []
+		res_accuracy = []
+		res_precision = []		
+		res_recall = []
+		res_fscore = []
+		x_train = x_train.reshape(x_train.shape[0], 1, -1)
+		x_test = x_test.reshape(x_test.shape[0], 1, -1)
+		if self.num_categories > 2:
+			y_train = to_categorical(y_train, self.num_categories)
+			y_test = to_categorical(y_test, self.num_categories)
+		for i in range(1, nb_epoch+1):			
+			self.model.fit(x_train, y_train, batch_size=1, nb_epoch=1, shuffle=False)
+			self.model.reset_states()
+			(loss, accuracy) = self.model.evaluate(x_test, y_test, batch_size=1)
+			self.model.reset_states()
+			res_loss.append(loss)
+			res_accuracy.append(accuracy)
+			(precision, recall, fscore) = (0,0,0)
+			if self.num_categories == 2:			
+				y_pred = self.model.predict(x_test, batch_size=1)
+				self.model.reset_states()
+				(precision, recall, fscore) = MyMetrics(y_test, y_pred)
+				res_precision.append(precision)
+				res_recall.append(recall)
+				res_fscore.append(fscore)
+			self.write_result(str(i) +'|'+ str(loss) +'|'+ str(accuracy) +'|'+ str(precision) +'|'+ str(recall) +'|'+ str(fscore))
+
+		return (res_loss, res_accuracy, res_precision, res_recall, res_fscore)
+
+
+	def write_result(self, text):
+		fp = open(self.results_file, 'a')
+		fp.write(text + '\n')
+		fp.close()
+
+#	def save_weights():
+
+
+
+	@staticmethod
+	def plot_results(title, metric, results):
+		lns = []
+		for k in results.keys():
+			result = results[k]
+			myplot = plt.subplot()
+			myplot.grid(True)
+			myplot.set_xlabel("Epoch Number")
+			myplot.set_ylabel(metric)
+			x_Axis = np.arange(1, len(result)+1)
+			#myplot.xaxis.set_ticks(x_Axis)#np.arange(	1, len(x_Axis)+1, 1))
+			#myplot.set_xticklabels(x_Axis, rotation=0)
+			tokens = k.split('|')
+			loss = tokens[0]
+			if loss=='categorical_crossentropy' or loss=='binary_crossentropy':
+				loss = 'crossentropy'
+			optimizer = tokens[1]
+			line = myplot.plot(x_Axis, result, label = 'loss:' + loss + ' opt:' + optimizer)		
+			lns = lns + line
+		box = myplot.get_position()
+		myplot.set_position([box.x0, box.y0 + box.height * 0.30, box.width, box.height * 0.70])
+		labs = [l.get_label() for l in lns]
+		plt.title(title)
+		lgd = plt.legend(lns, labs, loc='upper center', bbox_to_anchor=(0.5, -0.12), fancybox=True, shadow=True, ncol=2)
+		plt.savefig('/home/orestis/net/RNNresults/' + title + '.png')
+		plt.clf()

RNN_1A.py

@@ -0,0 +1,26 @@
+
+from keras.models import Sequential
+from keras.layers import Dense, Activation, SimpleRNN
+from keras.utils.visualize_util import plot
+import os
+import numpy as np
+from reader import Reader
+
+
+length = Reader.getInputShape()
+
+model = Sequential()
+
+#EXPECTS INPUT AS (nb_sample, timesteps, nb_features), where nb_sample=1 (batch_size = 1), timesteps = 1 and nb_features = length
+
+#model.add(Dense(40, input_dim = 12, init='uniform', activation='relu'))
+model.add(SimpleRNN(output_dim=50, input_shape=(1,length), batch_input_shape=(1,1,length), init='uniform', inner_init='uniform', activation='relu', stateful=True))
+model.add(Dense(1, init='uniform', activation = 'sigmoid'))
+
+
+model.summary()
+plot(model, to_file='/home/orestis/net/RNNmodels/RNN_1A.png')
+fp = open('/home/orestis/net/RNNmodels/RNN_1A.json', 'w')
+fp.write(model.to_json())
+fp.close()
+

RNN_1B.py

@@ -0,0 +1,26 @@
+
+from keras.models import Sequential
+from keras.layers import Dense, Activation, SimpleRNN
+from keras.utils.visualize_util import plot
+import os
+import numpy as np
+from reader import Reader
+
+
+length = Reader.getInputShape()
+
+model = Sequential()
+
+#EXPECTS INPUT AS (nb_sample, timesteps, nb_features), where nb_sample=1 (batch_size = 1), timesteps = 1 and nb_features = length
+
+#model.add(Dense(40, input_dim = 12, init='uniform', activation='relu'))
+model.add(SimpleRNN(output_dim=50, input_shape=(1,length), batch_input_shape=(1,1,length), init='uniform', inner_init='uniform', activation='relu', stateful=True))
+model.add(Dense(3, init='uniform', activation = 'softmax'))
+
+
+model.summary()
+plot(model, to_file='/home/orestis/net/RNNmodels/RNN_1B.png')
+fp = open('/home/orestis/net/RNNmodels/RNN_1B.json', 'w')
+fp.write(model.to_json())
+fp.close()
+

RNN_1C.py

@@ -0,0 +1,26 @@
+
+from keras.models import Sequential
+from keras.layers import Dense, Activation, SimpleRNN
+from keras.utils.visualize_util import plot
+import os
+import numpy as np
+from reader import Reader
+
+
+length = Reader.getInputShape()
+
+model = Sequential()
+
+#EXPECTS INPUT AS (nb_sample, timesteps, nb_features), where nb_sample=1 (batch_size = 1), timesteps = 1 and nb_features = length
+
+#model.add(Dense(40, input_dim = 12, init='uniform', activation='relu'))
+model.add(SimpleRNN(output_dim=50, input_shape=(1,length), batch_input_shape=(1,1,length), init='uniform', inner_init='uniform', activation='relu', stateful=True))
+model.add(Dense(5, init='uniform', activation = 'softmax'))
+
+
+model.summary()
+plot(model, to_file='/home/orestis/net/RNNmodels/RNN_1C.png')
+fp = open('/home/orestis/net/RNNmodels/RNN_1C.json', 'w')
+fp.write(model.to_json())
+fp.close()
+

RNN_plot_all.py

@@ -0,0 +1,85 @@
+import keras.utils
+from keras.models import Sequential, model_from_json, load_model
+from keras.layers import Dense, Activation, SimpleRNN
+from keras.utils.visualize_util import plot
+import keras.utils.np_utils
+from keras.utils.np_utils import to_categorical
+
+import os
+import numpy as np
+import datetime
+import random
+from reader import Reader
+from sklearn import metrics
+import matplotlib.pyplot as plt
+from sklearn import metrics
+import RNNHandler
+
+
+#(1, 'Dataset1', 2, 'RNN_1A')
+#(2, 'Dataset2', 2, 'RNN_1A')
+#(3, 'Dataset3', 2, 'RNN_1A') !! grafiki
+#(4, 'Dataset4', 2, 'RNN_1A') !!! mixed type binary and continous
+#(5, 'Dataset5', 3, 'RNN_1B') +categorical
+dataset_id = 5
+dataset_name = 'Dataset5'
+num_classes = 3
+RNN_name = 'RNN_1B'
+
+num_epochs = 10
+
+(x_train, y_train), (x_test, y_test) = Reader.getDataset(dataset_id)
+x_train = x_train[0:1000,:]
+y_train = y_train[0:1000]
+x_test = x_test[0:1000,:]
+y_test = y_test[0:1000]
+
+results = {}
+for loss in ['mae', 'mse', 'categorical_crossentropy']:	#categorical_crossentropy
+	for optimizer in ['sgd','adagrad', 'rmsprop']:
+		RNNmodel = RNNHandler.RNNHandler(RNN_name, num_classes, loss, optimizer)
+		(res_loss, res_accuracy, res_precision, res_recall, res_fscore) = RNNmodel.fit_and_eval(x_train, y_train, x_test, y_test, num_epochs, dataset_name)
+		if num_classes == 2:		
+			results[loss + '|' + optimizer] = res_fscore
+		else:
+			results[loss + '|' + optimizer] = res_accuracy
+
+title = 'Dataset ' + dataset_name + ', ' + 'Model ' + RNN_name
+metric = 'accuracy'
+if num_classes == 2:
+	metric = 'fscore'
+RNNHandler.RNNHandler.plot_results(title, metric, results)
+
+
+
+#(6, 'Dataset0', 5, 'RNN_1C') +categorical
+dataset_id = 6
+dataset_name = 'Dataset0'
+num_classes = 5
+RNN_name = 'RNN_1C'
+
+num_epochs = 10
+
+fp_logfile = open('/home/orestis/net/debug/logfile', "a")
+reader = Reader(fp_logfile, False)
+(x_train, y_train), (x_test, y_test) = reader.getDataNormalized()
+x_train = x_train[0:1000,:]
+y_train = y_train[0:1000]
+x_test = x_test[0:1000,:]
+y_test = y_test[0:1000]
+results = {}
+for loss in ['mae', 'mse', 'categorical_crossentropy']: 
+	for optimizer in ['sgd','adagrad', 'rmsprop']:
+		RNNmodel = RNNHandler.RNNHandler(RNN_name, num_classes, loss, optimizer)
+		(res_loss, res_accuracy, res_precision, res_recall, res_fscore) = RNNmodel.fit_and_eval(x_train, y_train, x_test, y_test, num_epochs, dataset_name)			
+		if num_classes == 2:		
+			results[loss + '|' + optimizer] = res_fscore
+		else:
+			results[loss + '|' + optimizer] = res_accuracy
+
+title = 'Dataset: ' + dataset_name + ', ' + 'Model: ' + RNN_name
+metric = 'accuracy'
+if num_classes == 2:
+	metric = 'fscore'
+RNNHandler.RNNHandler.plot_results(title, metric, results)
+

RNNmodels/RNN_1.json

@@ -0,0 +1 @@
+{"class_name": "Sequential", "config": [{"class_name": "SimpleRNN", "config": {"b_regularizer": null, "U_regularizer": null, "W_regularizer": null, "activation": "relu", "go_backwards": false, "init": "uniform", "inner_init": "uniform", "dropout_U": 0.0, "input_dtype": "float32", "batch_input_shape": [1, 1, 79], "name": "simplernn_1", "return_sequences": false, "dropout_W": 0.0, "unroll": false, "trainable": true, "stateful": true, "consume_less": "cpu", "output_dim": 50}}, {"class_name": "Dense", "config": {"b_regularizer": null, "W_regularizer": null, "activation": "softmax", "activity_regularizer": null, "b_constraint": null, "init": "uniform", "bias": true, "W_constraint": null, "input_dim": null, "output_dim": 3, "trainable": true, "name": "dense_1"}}], "keras_version": "1.1.1"}

RNNmodels/RNN_1A.json

@@ -0,0 +1 @@
+{"keras_version": "1.1.1", "config": [{"config": {"dropout_U": 0.0, "name": "simplernn_1", "stateful": true, "input_dtype": "float32", "unroll": false, "U_regularizer": null, "init": "uniform", "inner_init": "uniform", "dropout_W": 0.0, "output_dim": 50, "batch_input_shape": [1, 1, 79], "return_sequences": false, "trainable": true, "W_regularizer": null, "go_backwards": false, "b_regularizer": null, "consume_less": "cpu", "activation": "relu"}, "class_name": "SimpleRNN"}, {"config": {"name": "dense_1", "activity_regularizer": null, "trainable": true, "output_dim": 1, "b_constraint": null, "bias": true, "activation": "sigmoid", "input_dim": null, "init": "uniform", "W_regularizer": null, "W_constraint": null, "b_regularizer": null}, "class_name": "Dense"}], "class_name": "Sequential"}

RNNmodels/RNN_1B.json

@@ -0,0 +1 @@
+{"class_name": "Sequential", "keras_version": "1.1.1", "config": [{"class_name": "SimpleRNN", "config": {"consume_less": "cpu", "inner_init": "uniform", "return_sequences": false, "output_dim": 50, "W_regularizer": null, "go_backwards": false, "batch_input_shape": [1, 1, 79], "trainable": true, "b_regularizer": null, "stateful": true, "unroll": false, "U_regularizer": null, "activation": "relu", "input_dtype": "float32", "name": "simplernn_1", "dropout_U": 0.0, "dropout_W": 0.0, "init": "uniform"}}, {"class_name": "Dense", "config": {"activity_regularizer": null, "output_dim": 3, "W_regularizer": null, "W_constraint": null, "b_constraint": null, "activation": "softmax", "trainable": true, "b_regularizer": null, "bias": true, "input_dim": null, "name": "dense_1", "init": "uniform"}}]}

RNNmodels/RNN_1C.json

@@ -0,0 +1 @@
+{"keras_version": "1.1.1", "class_name": "Sequential", "config": [{"class_name": "SimpleRNN", "config": {"trainable": true, "go_backwards": false, "input_dtype": "float32", "activation": "relu", "unroll": false, "U_regularizer": null, "init": "uniform", "b_regularizer": null, "batch_input_shape": [1, 1, 79], "name": "simplernn_1", "consume_less": "cpu", "inner_init": "uniform", "stateful": true, "output_dim": 50, "dropout_U": 0.0, "return_sequences": false, "dropout_W": 0.0, "W_regularizer": null}}, {"class_name": "Dense", "config": {"trainable": true, "input_dim": null, "activation": "softmax", "W_constraint": null, "activity_regularizer": null, "b_constraint": null, "init": "uniform", "b_regularizer": null, "name": "dense_1", "output_dim": 5, "bias": true, "W_regularizer": null}}]}

RNNresults/OLD/RNNresultsDataset2

@@ -0,0 +1,45 @@
+RNN:: Loss Function:binary_crossentropy,   Optimizer:adagrad,   Batch Size:1,   Dropout Rate:No
+Loss Function: binary_crossentropy   Epoch: 0   Prec: 0.995253584705   Rec: 0.9995   F: 0.997372272485
+Loss Function: binary_crossentropy   Epoch: 1   Prec: 0.995452181514   Rec: 0.999577777778   F: 0.99751071391
+Loss Function: binary_crossentropy   Epoch: 2   Prec: 0.9956066089   Rec: 0.999622222222   F: 0.997610374633
+Loss Function: binary_crossentropy   Epoch: 3   Prec: 0.995760976636   Rec: 0.999644444444   F: 0.997698931528
+Loss Function: binary_crossentropy   Epoch: 4   Prec: 0.995838175881   Rec: 0.999655555556   F: 0.997743214395
+Loss Function: binary_crossentropy   Epoch: 5   Prec: 0.995926454798   Rec: 0.999677777778   F: 0.997798590432
+Loss Function: binary_crossentropy   Epoch: 6   Prec: 0.99594877301   Rec: 0.999744444444   F: 0.997842999174
+Loss Function: binary_crossentropy   Epoch: 7   Prec: 0.99601487779   Rec: 0.999733333333   F: 0.997870641469
+Loss Function: binary_crossentropy   Epoch: 8   Prec: 0.9960700962   Rec: 0.999755555556   F: 0.997909423123
+Loss Function: binary_crossentropy   Epoch: 9   Prec: 0.99611420474   Rec: 0.999755555556   F: 0.997931558429
+Loss Function: binary_crossentropy   Epoch: 10   Prec: 0.996136388797   Rec: 0.999788888889   F: 0.997959296845
+Loss Function: binary_crossentropy   Epoch: 11   Prec: 0.996191530584   Rec: 0.999788888889   F: 0.997986967974
+Loss Function: binary_crossentropy   Epoch: 12   Prec: 0.99623564841   Rec: 0.999788888889   F: 0.998009105983
+Loss Function: binary_crossentropy   Epoch: 13   Prec: 0.996323936488   Rec: 0.9998   F: 0.998058941624
+Loss Function: binary_crossentropy   Epoch: 14   Prec: 0.996357073257   Rec: 0.999811111111   F: 0.998081103864
+Loss Function: binary_crossentropy   Epoch: 15   Prec: 0.996379058334   Rec: 0.999788888889   F: 0.998081061295
+Loss Function: binary_crossentropy   Epoch: 16   Prec: 0.996434226991   Rec: 0.999788888889   F: 0.998108739178
+Loss Function: binary_crossentropy   Epoch: 17   Prec: 0.996467331118   Rec: 0.999788888889   F: 0.998125346644
+Loss Function: binary_crossentropy   Epoch: 18   Prec: 0.996478288315   Rec: 0.999766666667   F: 0.998119769048
+Loss Function: binary_crossentropy   Epoch: 19   Prec: 0.996522432525   Rec: 0.999766666667   F: 0.998141913441
+Loss Function: binary_crossentropy   Epoch: 20   Prec: 0.99655554325   Rec: 0.999766666667   F: 0.998158522381
+RNN:: Loss Function:mse,   Optimizer:adagrad,   Batch Size:1,   Dropout Rate:No
+Loss Function: mse   Epoch: 0   Prec: 0.996677004874   Rec: 0.999777777778   F: 0.998224983359
+Loss Function: mse   Epoch: 1   Prec: 0.996732098547   Rec: 0.999744444444   F: 0.998235998935
+Loss Function: mse   Epoch: 2   Prec: 0.996765188107   Rec: 0.999733333333   F: 0.998247054386
+Loss Function: mse   Epoch: 3   Prec: 0.996798245128   Rec: 0.999711111111   F: 0.998252553214
+Loss Function: mse   Epoch: 4   Prec: 0.996809182464   Rec: 0.999677777778   F: 0.998241419291
+Loss Function: mse   Epoch: 5   Prec: 0.996809111759   Rec: 0.999655555556   F: 0.99823030451
+Loss Function: mse   Epoch: 6   Prec: 0.996819979834   Rec: 0.9996   F: 0.998208054324
+Loss Function: mse   Epoch: 7   Prec: 0.996842000288   Rec: 0.999577777778   F: 0.998208014558
+Loss Function: mse   Epoch: 8   Prec: 0.996864057444   Rec: 0.999566666667   F: 0.998213532767
+Loss Function: mse   Epoch: 9   Prec: 0.996886047053   Rec: 0.999533333333   F: 0.99820793502
+Loss Function: mse   Epoch: 10   Prec: 0.99689699119   Rec: 0.9995   F: 0.998196798624
+Loss Function: mse   Epoch: 11   Prec: 0.996885909015   Rec: 0.999488888889   F: 0.998185702
+Loss Function: mse   Epoch: 12   Prec: 0.996918916523   Rec: 0.999444444444   F: 0.998180083006
+Loss Function: mse   Epoch: 13   Prec: 0.996940777442   Rec: 0.999366666667   F: 0.998152248098
+Loss Function: mse   Epoch: 14   Prec: 0.9969406418   Rec: 0.999322222222   F: 0.998130011375
+Loss Function: mse   Epoch: 15   Prec: 0.996940540061   Rec: 0.999288888889   F: 0.998113333185
+Loss Function: mse   Epoch: 16   Prec: 0.996929318907   Rec: 0.999233333333   F: 0.998079996449
+Loss Function: mse   Epoch: 17   Prec: 0.996940336563   Rec: 0.999222222222   F: 0.99807997514
+Loss Function: mse   Epoch: 18   Prec: 0.996929216784   Rec: 0.9992   F: 0.998063316778
+Loss Function: mse   Epoch: 19   Prec: 0.996918062592   Rec: 0.999166666667   F: 0.998041098095
+Loss Function: mse   Epoch: 20   Prec: 0.996929046563   Rec: 0.999144444444   F: 0.998035516093
+