saitejdandge / Sentimental_Analysis_LSTM_Conv1D

Sentimental Analysis using DeepLearning

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Sentimental_Analysis of Tweet Emotions

Sentimental Analysis using Long Short Term Memory Recurrent Neural Networks (DeepLearning)
(40000,)
(40000,)

Excluding stopwords ...
Tokenized to Word indices as
(40000,)
After padding data (40000, 20)
Loading Glove Vectors ...
Loaded GloVe Vectors Successfully
Embedding Matrix Generated : (32855, 50)
Label Encoding Classes as { 0: 'anger',
1: 'boredom',
2: 'empty',
3: 'enthusiasm',
4: 'fun',
5: 'happiness',
6: 'hate',
7: 'love',
8: 'neutral',
9: 'relief',
10: 'sadness',
11: 'surprise',
12: 'worry'}

One Hot Encoded class shape :
(40000, 13)

2018-10-09 00:56:56.582717: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2


x shape (4323, 300, 300, 3)
y shape (4323, 5)
Layer(type) Output Shape Param #
embedding_1 (Embedding) (None, 20, 50) 1642750
conv1d_1 (Conv1D) (None, 20, 30) 1530
max_pooling1d_1 (MaxPooling1 (None, 5, 30) 0
lstm_1 (LSTM) (None, 5, 100) 52400
flatten_1 (Flatten) ((None, 500) 0
dense_1 (Dense) (None, 500) 250500
dense_2 (Dense) (None, 300) 150300
dense_3 (Dense) (None, 13) 3913

Total params: 2,101,393
Trainable params: 458,643
Non-trainable params: 1,642,750

Train on 29936 samples, validate on 64 samples

Finished Preprocessing data ...
x_data shape : (40000, 20)
y_data shape : (40000, 13)
spliting data into training, testing set

Yeah I think, I need more data


Epoch 00096: val_acc did not improve from 0.40625 Epoch 97/100 29936/29936 [==============================] - 8s 277us/step - loss: 1.8189 - acc: 0.3791 - val_loss: 1.9288 - val_acc: 0.2969

Epoch 00097: val_acc did not improve from 0.40625 Epoch 98/100 29936/29936 [==============================] - 8s 259us/step - loss: 1.8162 - acc: 0.3803 - val_loss: 1.9488 - val_acc: 0.3281

Epoch 00098: val_acc did not improve from 0.40625 Epoch 99/100 29936/29936 [==============================] - 7s 249us/step - loss: 1.8131 - acc: 0.3824 - val_loss: 1.9878 - val_acc: 0.3438

Epoch 00099: val_acc did not improve from 0.40625 Epoch 100/100 29936/29936 [==============================] - 7s 249us/step - loss: 1.8122 - acc: 0.3822 - val_loss: 1.9144 - val_acc: 0.3281

Epoch 00100: val_acc did not improve from 0.40625

Test accuracy: 0.337



alt text

About

Sentimental Analysis using DeepLearning


Languages

Language:Python 100.0%