Project Exam 2 - Murarishetti-Shiva-Kumar/Python-Deep-Learning-Programming GitHub Wiki

Team 3:

Name	Class ID	Email
Jagruthi Bobbala	06	[email protected]
Lavanya Gadde	12	[email protected]
Sravani Garikapati	13	[email protected]
Shiva Kumar Murarishetti	31	[email protected]

1. Implement text classification on the review’s sentiment dataset using CNN model.

Imported the test.tsv, train.tsv dataset from above link using pandas library
Checked for the null values
Created Feature matrix and target Matrix
Removed the punctuation marks, capital letters by using regular expressions
Tokenized the data
Applied CNN with the embedding layer
Used Label encoder to convert target’s categorical data to numerical data
Fitting the model

a. Include Embedding layer in the design of your models and report if that leads to a better performance

Predicted the model

b. Plot loss of the model and report if you see any overfitting problem

We have found overfitting

c. What techniques you can apply to fix overfitting model

To overcome overfitting we are using early stop method

2. Implement text classification on the 20news_group dataset using LSTM model.

fetched 2 categories from 20news_group_data Dataset
Created Feature matrix and target Matrix
Tokenization of the data
Used Label encoder to convert target’s categorical data to numerical data
Split the data into train and test datasets

a. Include Embedding layer in the design of your models and report if that leads to a better performance

b. Plot loss of the model and report if you see any overfitting problem

We have seen the over fitting, to overcome overfitting we are using early stop method

3. Implement image classification with CNN model, using one of the below datasets:

10 monkey species

Uploaded datasets to drive and mounted to google colab.
Assigned train and test directories
Determined the label info
Checking the image in directory

a. Report your classification result with and without doing scaling

With Scaling

Without Scaling

b. Save the model and then predict on one of the test data. Report the prediction and check if it has been predicted correctly or not

4. The purpose of this question is to learn about text generation. Use New York Times Comments and Headlines to train a text generation language model which can be used to generate News Headlines

Necessary packages are imported and required data is loaded.
Loaded the dataset

Dataset cleaning

Generating sequence of n_gram tokens

Padding the Sequences and obtain Variables : Predictors and Target

LSTMs for Text Generation

a. Pass a sample headline or word to the model and predict the generated headline

5. Apply Autoencoder on the Cifar_10 dataset and then pass the result of Autoencoder to CNN or LSTM or three layers model to classify data

Loaded dataset into x_train, y_train,x_test,y_test
Reshaped the dataframes
Used Onehot encoder to convert target’s categorical data to numerical data

a. Repeat the same thing with PCA (apply PCA on the dataset and then pass the result to CNN or LSTM or three layers model)

b. Compare the result of both approach