for Sentiment Classification and Language Modeling
Build and train a feed forward neural network (FFNN) with 2 layers with hidden vector size 20.
Initalized weights: random weights.
Loss function: mean squared error.
Activation function: sigmoid.
Learning rate: 0.25-0.5.
Train/valid rate: 4:1.
Emoticon tokenizer: TweetTokenizer.
Minibatch: Yes.