Deep Learning - Word Embedding

Published: May 01, 2018 by Shaofei Li

• Categories:
• DataScience 4

• Tags:
• Deep Learning 1
• Python 4

Deep Learning

Word Embedding

• Produced word embeddings by using Neural Network;
• Trained deep neural network autoencoder for feature extraction to processing the data efficiently;
• Compared the performance and code readability of TensorFlow and PyTorch with the CUDA platform.

TensorFlow

Autoencoder

Results

• The initial accuracy is 97.9600% after 45 epoch.

   Extracting MNIST_data/train-images-idx3-ubyte.gz
   Extracting MNIST_data/train-labels-idx1-ubyte.gz
   Extracting MNIST_data/t10k-images-idx3-ubyte.gz
   Extracting MNIST_data/t10k-labels-idx1-ubyte.gz
   The accuracy is 0.9796000123023987 after 45 epoch with learning rate 0.001 and batch size 100.

• The final accuracy is 99.0100% after 20 epoch.

Extracting MNIST_data/train-images-idx3-ubyte.gz
Extracting MNIST_data/train-labels-idx1-ubyte.gz
Extracting MNIST_data/t10k-images-idx3-ubyte.gz
Extracting MNIST_data/t10k-labels-idx1-ubyte.gz
The final accuracy on test set is 99.0100%.