Python and Deep Learning: Special Topics

Rajeshwari Sai Aishwarya Puppala

Student ID: 16298162

Class ID: 35

Deep Learning-In class programming:5

Objectives

1.Save the model and use the saved model to predict on new text data (ex, “A lot of good things are happening. We are respected again throughout the world, and that's a great thing.@realDonaldTrump”)

2.Apply GridSearchCV on the source code provided in the class

3.Apply the code on spam data set available in the source code (text classification on the spam.csv data set)

Import Data

• Import the necessary Packages required
• Import the sentiment dataset and load all of the train and test data
• Make the text to lower and remove the text which is other than letters and numbers.

Tokenization

• Take the text and tokenize the words.
• We use fit_on_text which means that it will make a dictionary of words will a number as a value

Model

• Create a model with and embedded layer with embed dimension =128, max features=2000 and input shape
• Add an LSTM layer with dropout -0.2 and recurrent dropout=0.2
• Add a dense layer with activation function "Softmax"(This is the output layer)

Encoding

As the sentiment column is categorical, It has to be encoded so label encoder is used

Encoded Values

• The encoded values are 0- negative
• 1- neutral
• 2- positive

Accuracy and Loss

The accuracy and loss of the model are 68.1% and 79.3%

Save and Load Model and Prediction on text

Now Save the model into the system, with name final.DLICP5 and load the mode

Encoded

• The encoded values are 0- negative
• 1- neutral
• 2- positive

With the model loaded predict the text to which class it belongs From the following, we can say that it belongs to the neutral class

GridSearchCv

• It has the provision of providing different hyperparameters
• It auto-tunes the parameters, takes the best hyperparameters and gives the best result
• The best accuracy is 68% with batch size=64, epochs=2

Model2

Load the Spam data set and create the same model which is done above

Accuracy and loss of the model are 98% and 91%