#Document Classifier Using Vanilla Recurrent Neural Network

The Dataset consists of Textual Data that belong to one of the 8 Categories

Data Preprocessing Steps

1. Tokenize The Sentences
2. Remove Tokens That Are Stop Words
3. If the number of words in the Document is Greater Than 20 then retain only the First 20 Words of the sentence.
4. Convert Every Word to its Unique Identifcation Number
5. Now for the sentences with number of words Less Than 20, pad the sentence with 0's

Had to Fix the Sequence Length to 20 and Pad with 0's to make Training Faster

RNN Model Configuration

1. Learning Rate = 1e-2
2. Epochs = 4050
3. Word Embedding Dimension = 100
4. Hidden State Dimension = 128
5. Truncated Backpropagation Length = 4
6. Training Sequence Length = 20
7. Batch Size = 1000
8. Weight Initialization was done from a Gaussian Distribution with mean=0.0 and std=1
9. Bias were Zero Initialized

Model Performance

Test Set Accuracy = 74.22%
The training time for the model was about 6 hours

Loss-Iteration Curve

Downloading Model

The Model can be downloaded from here