loss curve

source/loss_curve/loss_curve.py

@@ -0,0 +1,59 @@
+from keras.models import Model
+from keras.preprocessing import sequence
+from keras.layers import Input
+from keras.layers import Dense, Activation
+from keras.layers import Embedding
+from keras.layers import Conv1D, GlobalMaxPooling1D
+from keras.datasets import imdb
+from keras import optimizers
+import matplotlib
+matplotlib.use('agg')
+import matplotlib.pyplot as plt
+
+# set parameters:
+batch_size = 32
+embedding_dims = 200
+filters = 250
+kernel_size = 3
+
+
+# load the dataset but only keep the top n words, zero the rest
+top_words = 5000
+(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=top_words)
+# pad dataset to a maximum review length in words
+max_words = 500
+x_train = sequence.pad_sequences(x_train, maxlen=max_words)
+x_test = sequence.pad_sequences(x_test, maxlen=max_words)
+
+# Build model
+sentence = Input(batch_shape=(None, max_words), dtype='int32', name='sentence')
+embedding_layer = Embedding(top_words, embedding_dims, input_length=max_words)
+sent_embed = embedding_layer(sentence)
+conv_layer = Conv1D(filters, kernel_size, padding='valid', activation='relu')
+sent_conv = conv_layer(sent_embed)
+sent_pooling = GlobalMaxPooling1D()(sent_conv)
+sent_repre = Dense(250)(sent_pooling)
+sent_repre = Activation('relu')(sent_repre)
+sent_repre = Dense(1)(sent_repre)
+pred = Activation('sigmoid')(sent_repre)
+model = Model(inputs=sentence, outputs=pred)
+
+#underfit
+epochs = 10
+sgd = optimizers.sgd(lr=0.01)
+history = model.compile(loss='binary_crossentropy', optimizer=sgd, metrics=['accuracy'])
+
+#overfit
+# epochs = 30
+# sgd = optimizers.sgd(lr=0.01)
+# history = model.compile(loss='binary_crossentropy', optimizer=sgd, metrics=['accuracy'])
+
+loss = history.history['loss']
+val_loss = history.history['val_loss']
+plt.plot(loss, label='loss')
+plt.plot(val_loss, label='val_loss')
+plt.title('model loss')
+plt.ylabel('loss')
+plt.xlabel('epoch')
+plt.legend(['train', 'valid'], loc='upper left')
+plt.savefig('./loss.png')