Lab 2 Wiki Report - adtmv7/CS5590-490-Python-Deep-Learning GitHub Wiki

Report for Lab Assignment 2

Class ID - 24

I. Introduction

The Module-2 of CS5590/490 class provided a good introduction to the foundations of Deep Learning and helped us learn about how to build neural networks, learned about Convolutional networks, Recurrent Neural Networks, LSTM, different activation, optimizers and loss functions. We had an exposure to use and explore datasets from healthcare, real estate, image and text classification. I have used the learnt concept from Module-2 to develop my class Project and solutions for given problem in Lab and ICP’s. The Lab 2 assignment focus on exploring and developing Deep learning models to analyze diverse range of datasets which includes image, text using Keras and Python.

II. Objectives

  1. Create a linear regression sequential model on a data set of our choice, display the results obtained using Tensor board analysis and document any changes impacted from variations in learning rate, batch size, optimizer and activation function.
  2. Implementation of logistic regression analysis on heart disease related data, normalize the data before feeding to the model, show graphical representation of loss associated to the model using tensor flow analysis and document any changes impacted from variations in learning rate, batch size, optimizer and activation function.
  3. Develop a CNN model to perform image classification on natural images data set.
  4. Perform a text classification leading to sentiment analysis on IMDB movie review data using CNN model along with evaluation of the difference in the results based on the different model approaches.
  5. Perform a text classification leading to sentiment analysis on IMDB movie review data using LSTM model and evaluate difference compared to CNN model.
  6. Perform auto-encoding techniques using the MNIST data set, encode and decode images.

III. Tools/Software:

  • PyCharm
  • Python3 Interpreter
  • Anaconda

IV. Datasets Used:

  • Heat Disease and Death rate
  • Natural Image classification
  • IMDB movie reviews
  • MNIST dataset

V. Workflow/Approaches:

  • Identify the specific data set to be used for the deep learning model for given problem.
  • Load the appropriate data set as needed
  • Process, filter and select only the required information from the data.
  • Split the data set into training, validation and test data sets to be used on the model.
  • Identify the deep learning models to be used for given problem.
  • Identify the input, hidden (one or more as considered appropriate) and output layers to be applied to the model defined
  • Compile the model using appropriate optimizer, loss and metrics functions
  • Using the training data set, fit the model defined
  • Evaluate the model by predicting the output for test data and cross-comparing the prediction with actual result of the test data.
  • Identify the accuracy and loss associated with the model obtained using the test data
  • Visualize the results (accuracy, loss) using graphical means.
  • Define the tensor call back model using Tensor Board function
  • Using the training data set, fit the model with tensor call back
  • Evaluate the model by predicting the output for test data and cross-comparing the prediction with actual result of the test data.
  • Identify the accuracy and loss associated with the model obtained from tensor flow
  • Visualize the results (accuracy, loss) obtained from the tensor flow as graphical presentation

VI. Concepts:

  • Linear regression sequential data analysis
  • Logistic regression data analysis
  • Image classification using CNN model
  • CNN model-based text classification
  • LSTM model-based text classification
  • Autoencoding resulting in encoding and decoding of images

VII. Source Code & Report

VIII. Video

IX. References: