DL Lab 1 - SASLEENREZA/Python_DeepLearning GitHub Wiki

Team ID: 15

Name : Sasleen Reza Shaik, Class ID : 27

Name : Navya Ramya Sirisha, Class ID : 24

Youtube link : Click me!

Introduction :

The main aim of the lab assignment is to create an exposure to Deep Learning Concepts like:

  • Linear Regression
  • Logistic Regression

Objective :

The assignment is divided into two tasks which focuses to make one familiar with deep learning concepts listed above:

  • Plotting the graphs in tensor board.
  • Plotting the losses
  • Changing the hyperparameters like learning rate, batch size, activation function, optimizer and finding the change in results.

Program-1 : Linear Regression

The main objective is to analyse the Linear Regression Model on a dataset. A graph is plotted in the Tensor Board and also losses are calculated. The variation in results can be viewed by changing learning rate, batch size, optimizer, activation function in each case. To implement the Linear Regression model, we have used mnist dataset. The model is trained by varying epochs, batch size and optimizer. We have used the Deep Learning libraries and concepts like Keras,Tensorboard, optimizer and Activation functions to evaluate the model.

Case 1: batch_size = 128, nb_classes = 20, nb_epoch = 200, optimizer=adam learning rate=0.002, activation function=sigmoid, Results: Loss- 17% Case 2: batch_size = 140, nb_classes = 10, nb_epoch = 50, optimizer=RMSprop, learning rate=0.001, activation function=tanh,sigmoid ,Results: Loss- 12%

Program-2 : Logistic Regression

To implement logistic regression, we took MNIST dataset of handwritten digits. We built a Sequential model and tuned the model by changing the hyperparameters and calculated which suited the best while calculating the loss and accuracy and also plotting these values in tensorboard.

Case 1: Batch size: 150, classes= 30, epoch = 400 optimizer=adam We got the results as Accuracy = 92% loss= 27% Case 2: Batch size: 130, classes= 20, epoch = 200 optimizer=relu We got the results as Accuracy = 92% loss= 27% Case 3: Batch size: 180, classes= 10, epoch = 200 optimizer=adam We got the results as Accuracy = 95% loss= 22%

Conclusion:

Thus the linear regression and logistic regression have been worked out by varying the hyperparameter values and the accuracy and losses are calculated.