Lab Assignment 4 - sirisha1206/Python GitHub Wiki
Python LAB Assignment - 4
Name: Vinay Santhosham Class ID: 28 Team : 4 Tech Partner: Naga Sirisha Sunkara Class ID:34
CNN Datasets:
- Eco hotel data
- Sentimental label data
1. Implement the text classification with CNN model, with a new dataset which is not used in the class
Hyper parameters:
Filter Size : 3,4,5
Optimizer : RMS Prop Optimizer
Number of Filters : 32
Dropout Probability : 0.25
Batch size : 64
Number of epochs : 100
Output:
2018-07-27T11:07:08.333553: step 300, loss 0.0975096, acc 0.941176
Filter Size : 3,4,5
Optimizer : Adam Optimizer
Number of Filters : 32
Dropout Probability : 0.25
Batch size : 64
Number of epochs : 100
Output:
2018-07-27T11:33:00.207615: step 300, loss 0.34742, acc 0.823529
Filter Size : 3,4,5
Optimizer : Adagrad Optimizer
Number of Filters : 32
Dropout Probability : 0.25
Batch size : 64
Number of epochs : 100
Output: 2018-07-27T11:36:44.425907: step 300, loss 0.541423, acc 0.764706
Filter Size : 3,4,5
Optimizer : Gradient descent Optimizer
Number of Filters : 32
Dropout Probability : 0.25
Batch size : 64
Number of epochs : 100
Output:
Evaluation:
2018-07-27T11:40:51.604835: step 300, loss 0.471642, acc 0.823529
Filter Size : 1,2,3
Optimizer : Gradient descent Optimizer
Number of Filters : 64
Dropout Probability : 0.125
Batch size : 32
Number of epochs : 50
Output:
Evaluation:
2018-07-27T11:44:25.826457: step 300, loss 0.555298, acc 0.705882
Filter Size : 1,2,3
Optimizer : Adam Optimizer
Number of Filters : 64
Dropout Probability : 0.125
Batch size : 32
Number of epochs : 50
Output:
Evaluation:
2018-07-27T11:47:46.011735: step 300, loss 0.179279, acc 0.882353
Filter Size : 1,2,3
Optimizer : Adagrad Optimizer
Number of Filters : 64
Dropout Probability : 0.125
Batch size : 32
Number of epochs : 50
Output:
Evaluation:
2018-07-27T11:50:11.530652: step 300, loss 0.782413, acc 0.529412
Filter Size : 1,2,3
Optimizer : RMS Prop Optimizer
Number of Filters : 64
Dropout Probability : 0.125
Batch size : 32
Number of epochs : 50
Output:
Evaluation:
2018-07-27T11:53:29.406375: step 300, loss 0.0986688, acc 1
Tensor Board Graph and Summaries:
RNN
Datasets:
- Eco hotel data
- Sentimental label data
2. Implement the text classification with RNN/LSTM model, with a new dataset which is not used in the class
Hyper parameters:
Drop Out Probability : 0.25
Batch size: 64
Number of epochs: 100
Optimizer: RMS Prop Optimizer
Output:
Evaluation:
2018-07-27T19:50:37.805353: step 200, loss 0.37526, acc 0.909091
Drop Out Probability : 0.25
Batch size: 64
Number of epochs: 100
Optimizer: Adam Optimizer
Output:
Evaluation: 2018-07-27T19:51:58.919084: step 200, loss 0.561634, acc 0.909091
Drop Out Probability : 0.25
Batch size: 64
Number of epochs: 100
Optimizer: Adagrad Optimizer
Output:
Evaluation: 2018-07-27T19:52:14.095679: step 200, loss 4.38106, acc 0.727273
Drop Out Probability : 0.25
Batch size: 64
Number of epochs: 100
Optimizer: Gradient descent Optimizer
Output:
Evaluation: 2018-07-27T19:52:29.014433: step 200, loss 2.02515, acc 0.818182
Drop Out Probability : 0.125
Batch size: 32
Number of epochs: 50
Optimizer: RMS Prop Optimizer
Output:
Drop Out Probability : 0.125
Batch size: 32
Number of epochs: 50
Optimizer: Adam Optimizer
Output:
Drop Out Probability : 0.125
Batch size: 32
Number of epochs: 50
Optimizer: Ada grad
Output:
Drop Out Probability : 0.125
Batch size: 32
Number of epochs: 50
Optimizer: Gradient descent
Output:
Tensorboard scalar and graphs:
3. Compare the results of CNN and RNN/LSTM models, for the text classification (same dataset for 2 models to compare) and describe, which model is best for the text classification based on your results
Datasets used:
- Eco hotel data
- Sentimental label data
CNN parameters:
Filter Size : 3,4,5
Number of Filters : 32
Dropout Probability : 0.25
Batch size : 64
Number of epochs : 100
RNN parameters:
Drop Out Probability : 0.25
Batch size: 64
Number of epochs: 100
Optimizer
RMS Prop
Output:
CNN : 2018-07-27T11:07:08.333553: step 300, loss 0.0975096, acc 0.941176
Evaluation: RNN : 2018-07-27T19:50:37.805353: step 200, loss 0.37526, acc 0.909091
Adam
CNN: Output:
2018-07-27T11:33:00.207615: step 300, loss 0.34742, acc 0.823529
RNN
Evaluation:
2018-07-27T19:51:58.919084: step 200, loss 0.561634, acc 0.909091
Ada grad
CNN
Output:
2018-07-27T11:36:44.425907: step 300, loss 0.541423, acc 0.764706
RNN
Evaluation:
2018-07-27T19:52:14.095679: step 200, loss 4.38106, acc 0.727273
Gradient descent
CNN
Evaluation:
2018-07-27T11:40:51.604835: step 300, loss 0.471642, acc 0.823529
RNN Evaluation: 2018-07-27T19:52:29.014433: step 200, loss 2.02515, acc 0.818182
From the table above CNN model is best for text classification. In CNN, RMS Prop optimizer is best for text classification.
4. Implement the image classification with CNN model, with a new dataset which is not used in the class
(E.g. CIFAR 10 dataset)
Code for CNN model:
Optimizer Used: RMS Prop
Output for CNN Model:
Ada grad optimizer
Output:
step 0, training accuracy 0.06
step 100, training accuracy 0.4
step 200, training accuracy 0.48
step 300, training accuracy 0.64
step 400, training accuracy 0.64
test accuracy 0.6137
Time for building convnet:
103051
Adam optimizer
step 0, training accuracy 0.22
step 100, training accuracy 0.8
step 200, training accuracy 0.82
step 300, training accuracy 0.84
step 400, training accuracy 0.76
test accuracy 0.8989
Time for building convnet: 95280
TensorBoard Graph:
Youtube Links: Part 1: https://youtu.be/lceLXh3cTis Part 2: https://youtu.be/YCA4Bx5PCSc