LAB 3

Team Details:

Name: Aashish Thota

Class Id: 37

Name: Nikhil Sanka

Class Id: 34

Name: Sumanth Sanakkayala

Class Id: 33

Task 1:

Aim: Build a Sequential model using keras to implement Linear Regression

Approach:

• Normalized the data i.e normalized all the features in the dataset
• Then built the basic neural network
• Use keras Wrapper class i.e Kerasregression to apply linear regression and find the error and loss of the model.

After changing the hyper parameters

Task 2:

Aim : Implement Logistic Regression on a Kaggle Dataset

Approach:

• Normalized the data i.e normalized all the features in the dataset
• Built the neural model using softmax activation function
• plotted the loss for the model

After Changing the hyper parameters

Task 3:

Aim: Implement the image classification with CNN model on Kaggle Dataset

Approach :

• Generated the image dataset from the images using ImageGenerator
• ImageGenerator produces iterator object of combined training features and labels
• Applied CNN with SGD optimizer
• fitted the model on the training data

Task 4:

Aim: Implement the text classification with CNN model on movie reviews dataset

Approach:

• Did word Embedding for the phrases
• Then built LSTM neural network
• fitted the data onto the model
• plotted the loss

Task 5:

Aim : Implement the text classification with LSTM model on movie reviews dataset

Appraoch:

• Did word Embedding for the phrases
• Then built CNN 1D neural network
• fitted the data onto the model
• plotted the loss

Task 6:

Compare the results of CNN and LSTM models, for the text classification and describe, which model is best for the text classification based onyour results. Tune the hyperparameters to attain good accuracy and show the results.

As above outputs are showing that CNN has 55% accuracy where as LSTM has 63% accuracy on same dataset for the epochs size 5.Hence LSTM has right hand compared to CNN slightly but not major change .

Video Link: Videos for lab 3

Reference : Keras Models