Udacity_Natural_Language_Processing

Codes learn from Udacity Course (Natural Language Processing)

Project 1 - POS Tagger

Part of Speech Tagger using Hidden Markov Model. pomegranate is used for build a model. Simple HMM model with add-one Laplace Smoothing is applied

Project 1 (Optional) - IBM Bookworm

By using IBM Watson's Cloud-based NLP services, build a simple QA model. QA model with assistant workspace. By sending a query to system, simple matched answer can be returned.

Project 2 - Machine Translator

Implement various RNN mdoels with keras to make machine traslator (Eng->Fr)

• Understand how to construct RNN model using keras
• 4 versions are available
  1. simple RNN (GRU) model which uses word number itself as input
  2. RNN model with embedding layer
  3. Bidiretional RNN model
  4. Encoder-Decoder model
  5. (final) Encoder-Decoder model with Embedding layer

Exercise 1 - LDA topic Modeling

• Understand Latent Dirichlet Allocation
• Build LDA model with BOW and TFIDF
• Topic classification

Exercise 2 - Sentimental Analysis

• Can draw WordCloud with IMDB image

1. classic ML classification models

• Make textdata into vector
• Sentimental analysis with GradientBoostingClassifier, GaussianNB

2. RNN classification model

• use LSTM to build sentimental anaylsis model

3. NN classification model

• Implement NN model from the base
• Parameter adjusting is contained

Exercise 3 - Keras

• Can build NN models using Keras
  • MLP, RNN (LSTM)

Exercise 4 - Attension Basics

• Reproduce and visualize a process of how attention is applied on single cell

Exercise 5 - Deciphering Code with character level RNNs

• using many-to-many GRU cell to build code deciphering model
• Basic generation model to translate code to plain text