responses

Baseline results:

Number of epochs to reach a validation accuracy of 0.4: Epoch 3
Validation accuracy after 100 epochs (batch_size = 100): 0.645

Add your names (all your names), a brief description of the
modification you made (what's on the assignment is fine) and the summary results:

A. Names: Nirman Dave, Mei Zhou, Alina Dao
Description: run two different learning rates, one larger and one smaller
Summary of Results:
Learning Rate = 0.01
Number of epochs to reach a validation accuracy of 0.4: Epoch 7
Validation accuracy after 100 epochs (batch_size = 100): 0.655531

Learning Rate = 0.3
Number of epochs to reach a validation accuracy of 0.4: Epoch 5
Validation accuracy after 100 epochs (batch_size = 100): 0.492098

B: Change batch size from 100 to 1000
Validation accuracy after 10 epochs (batch size = 1000): 0.3809
First time validation accuracy > 0.4 is after 12 epochs

After 102 epochs, validation accuracy is 0.6

C: John Festa, Max Roche, John Cross
Description: run a batch size of 20
Summary Results:
Number of epochs to reach a validation accuracy of 0.4: Never
Validation accuracy after 100 epochs (batch_size = 100) : 0.10482097

D: Ian Nanez, Obi Daniel, Hutomo Limanto
We changed the size of the fully connected hidden layer to 512.
It took us 2 epochs to reach a validation accuracy > 0.4. 
After 10000 epochs * batch size, our validation accuracy is 0.641.

E. Names: Saharsha Karki, Evelyn Ting, Zach Bessette
Description: Change the size of the fully connected hidden layer to 256
Summary Results:
Number of epochs to reach a validation accuracy of 0.4: Epoch 2
Validation accuracy after 100 epochs (batch_size = 100): 0.643529

F. Names: Ella Yarmo-Gray, Lehua Matsumoto, Sarah Teichman
Description: Changed the size of the fully connected hidden layer to 2048
Summary results:
- It took 4 epochs to reach validation accuracy of 0.4
- Validation accuracy after 100 epochs with 100 batch size = 0.58191639

G. Names: Mahesh Dhulipala, Ian McClaugherty, Bishesh Yadav
Description: Remove Dropout
Summary Results:
Number of epochs to reach a validation accuracy of 0.4: Epoch 6
Validation accuracy after 100 epochs (batch_size = 100): 0.50790161

H. Names: Kevin Goldberg, Robbie Zielinski, Andy Tang

Description: Change the keep percentage to 0.25

Summary Results:
Number of epochs to reach a validation accuracy of 0.4: 3
Validation accuracy after 10000 epochs*batch_size: 0.66493297

I. Names: Sean Fitzgerald, Reynaldo Pena, Juan Guaracao
Description: Change the keep percentage to 0.75
Summary Results:
Number of epochs to reach a validation accuracy of 0.4: Never
Validation accuracy after 100 epochs (batch_size = 100): 0.10482097
(Stuck)

With different params 3 epochs,  validation .57 after 9

J. Names: Nicholas Lafky, Connor Haley, Phong Hoang
Description: Use momentum optimization with a momentum of 0.5

**Note: we were supposed to use a momentum of 0.9 but neither the validation
accuracy or the test accuracy were changing at all. Thus we opted for a 
momentum of .5.
Summary Results:
Number of epochs to reach a validation accuracy of 0.4: Epoch 12
Validation accuracy after 100 epochs (batch_size = 100): 0.611322

K. Names: Mahima Ghale, Alex Watson, Allison Lee
Description: Use Adam Optimizer
Summary Results:
- Number of epochs to reach validation accuracy of 0.4 (with learning rate 0.00001) = Epoch 20
- Validation accuracy after 100 epochs and 100 batch size = 57111424

L. Names: Jeff Ewing, Sam Procter, Jake Meyer
Description: Change the number of features of the first conv layer to 16
Summary of Results:
Number of epochs to reach a validation accuracy of 0.4: Epoch 2
Validation accuracy after 100 epochs (batch_size = 100): 0.57431489

M. Names: Daniel Mariselli, Alex Chou, Carina Corbin
Description: Change the number of features in the first convolution layer to 8.
Summary Results:
Number of epochs to reach a validation accuracy of 0.4: Epoch 3
Validation accuracy after 100 epochs (batch_size = 100): 0.56391281

N. Names: Andrew Kim, Emily Masten
Description: Change the number of features in the first convolution layer to 64.
Summary Results:
Number of epochs to reach a validation accuracy of 0.4: Epoch 5
Validation accuracy after 100 epochs (batch_size = 100): 0.65273052

O. Names: Pierre-Alexander Low, Gabriel Young
Description: Change the number of features in the second convolution layer to 32.
Summary Results:
Number of epochs to reach a validation accuracy of 0.4: Epoch 3
Validation accuracy after 100 epochs (batch_size = 100): 0.64112824

P. Names: Zeina Amhaz, Patrick Liu
Description: Change the number of features in the second conv layer to 16
Summary Results:
Number of epochs to reach a validation accuracy of 0.4: 3
Validation accuracy after 100 epochs (batch_size=100): 0.60932189

Q: Change the size of the convolutions to 3x3

Number of epochs to reach a validation accuracy of 0.4: Epoch 4
Validation accuracy after 100 epochs (batch_size = 100): 0.647129

R. Names: Dorit Song, Harrison Marick, Megan Root
Description: 7x7 Convolutions

Number of epochs to reach a validation accuracy of 0.4: 11
Validation accuracy after 100 epochs (batch_size = 100): 0.575515

S. Shu (Roy) Liu, Sam Stallop and Michael Shi.
Description: change the maxpool layers so they have a stride of 1 (so the size of the image does not decrease

Reached validation accuracy of 0.4 after 2 epochs
+Program never completed more than 5 epochs in one run without crashing


T. Names: Tiffany Kha, Jordan Browning, Azka Javaid
Description: add another conv/relu/maxpool layer so that the input to the fully connected layer is 3x3. 
Summary results: 
- took 8 epochs to get to a validation accuracy of .4
validation accuracy after 100 epochs w/ batch size 100 = .56631327

U. Names: Oluwatobi Oni-Orisan, Amal Buford, Mohammed Ibrahim
Description: Add another conv/relu/maxpool layer while maintaining the
    fully connected layer at 6x6
Summary Results:
    Number of epochs to reach a validation accuracy of 0.4: Epoch 5
    Validation accuracy after 100 epochs (batch_size = 100): 0.6365273

V. Names: Evan Wolf, Brendan Routh, Gab Gonclaves
Description: add a second fully connected layer of size 1024.
Summary Results:
- Took 11 epochs to get a validation accuracy of 0.4
- Validation accuracy after epoch 100: 0.593619 (batch size = 100)

W. Names: Jackson Herrick, Drew Kelleher
Description: Add a second fully connected layer of size 128.
Summary Results:
Number of epochs to reach a validation accuracy of 0.4: Epoch 8
Validation accuracy after 100 epochs (batch_size = 100): 0.65893179

X. Names: Ehsan Mojahedi, Jack Muller, Olivia Pinney
Description: Add a second fully connected layer but make the size of both 128.
Summary Results:
Number of epochs to reach a validation accuracy of 0.4: Epoch 3
Validation accuracy after 100 epochs (batch_size = 100): 0.62292457

Y. Names: Michael Barnett, Denison Marsland-Rello, Miriam De La Riva Ovalle
Description: Eliminate the conv layers, leaving an ordinary NN with 1024 hidden units
Did not reach a validation accuracy of 0.4 after 1000 epochs.
Validation accuracy after epoch 100: 0.29105821