Deep Learning

Deep Learning Assignment 

In this assignment, you’ll need amazon_review_500.csv for this assignment. This csv file has two columns as follows. The label column provides polarity sentiment, either positive or negative

Train a CNN classification model

Create a function sentiment_cnn( ) to detect sentiment as follows:

the input parameter is the full filename path to amazon_review_500.csv convert the text into padded sequences of numbers (see Exercise 5.2)

hold 20% of the data for testing

carefully select hyperparameters: max number of words for embedding layer, inputsentence length, filters, the number of filters, batch size, and epoch etc. create a CNN model with the training data

print out accuracy, precision, recall calculated from testing data.

Your precision_macro, recall_macro, and accurracy should be all about 70%.

If your result is much lower than that (e.g. below 67%), you need to tune the hyperparameters

Also note that the label in the dataset is either 1 or 2. Your binary prediction out of CNN is either 0 or 1. Conversion is needed in order to compare predictions with actual labels

This function has no return. Besides your code, also provide a pdf document showing the following

How you choose the hyperparameters

Screenshots of model trainning history

Testing accuracy, precision, recall

Improve the performance of CNN model

Create a function improved_sentiment_cnn( ) to detect sentiment with improved accuracy

You still need to train a CNN model

You can apply different techniques, e.g.

map words to pretrained word vectors

e.g. from Google

(https://drive.google.com/file/d/0B7XkCwpI5KDYNlNUTTlSS21pQmM/edit?

usp=sharing

(https://drive.google.com/file/d/0B7XkCwpI5KDYNlNUTTlSS21pQmM/edit?

usp=sharing)) or

e.g. from spacy package (https://spacy.io/usage/vectors-similarity(https://spacy.io/usage/vectors-similarity))

e.g. create your own pretrained word vectors using other review documents you can find

add additional features etc.

Your taraget is to improve the accuracy by about 5% from the model you created in Q1.

For fair comparison, make sure you use the same datasets for training/testing.

This function has no return. Please provide a pdf document showing the following

Screenshots of model trainning history

Testing accuracy, precision, recall

Your analysis about

what technique contributes to the performance improvement why this technique is useful

In [ ]:

Last Updated on February 11, 2019