Ubuntu 16.04
Python 2
matplotlib pickle numpy opencv (This is optional if need to construct the dataset)
To install libraries if missing,I use pip
sudo apt-get install python-pip
Numpy
sudo pip install numpy
Matplotlib
sudo pip install matplotlib
pickle
sudo pip install opencv
I have converted the dataset of images (read opencv library) to randomized data set and pickled it which is a 549 mb file You may download it from: https://drive.google.com/file/d/15jpYFUBDXY9sOXI_mRkIIpQk6RYtMJmy/view?usp=sharing and paste it in the folder of code
To reconstruct the dataset from scratch:
uncomment the lines:11-13 in main.py
1 in Utility.py
I have stored the outputs of all experiments as pickled file in a folder stored_experiments
python graph.py <experiment_no.>
e.g
python graph.py 1
python graph.py 2
python graph.py 3
python graph.py 4
python main.py <iteration> <batch_size> <learning_rate>(optional)
python main.py 5000 64 0.01
python main.py 1000 32 0.01
python main.py 1000 64 0.01
python main.py 1000 128 0.01
python main.py 1000 64 0.001 0.5
python main.py 1000 64 0.05
python main.py 1000 64 0.005
python main.py 1000 64 0.001
Note: All these experiments take time
When done on Lab computers(with 8 cores) each epoch took aroudn 30 seconds
Each experiment when ends creates two pickled file(of small kbs) with name format
"Training_Epoch"+str(iteration)+"_batch_size"+str(mini_batch_size)+"_alpha"+str(learning_rate)
"Test_Epoch"+str(iteration)+"_batch_size"+str(mini_batch_size)+"_alpha"+str(learning_rate)
Paste the files in the stored_experiments folder before running graph.py
Contact:
[email protected] (Valid till 2020)
[email protected]
github.com/adityamatt