COMP 551 Project 3 - Image Analysis
Objective: Devise a machine learning algorithm to automatically compute mathematical functions from images.
Note: Please retrieve the relevant .csv files from https://www.kaggle.com/c/comp551-modified-mnist/data and place them in the data directory.
_HOW TO RUN
1)Pickle the data
a) Retrieve the relevant .csv files from https://www.kaggle.com/c/comp551-modified-mnist/data and place them in the data directory.
b) Write the following in a command line prompt
cd ~/ARBITRARY_PATH/comp551-project3-master
python preprocess/pickle_data.py
-
Run the CNN predictor
a) make sure to run in an (virtual) environment that supports tensorflow. Information regarding this is found at https://www.tensorflow.org/install/
b) if using a virtual environment, activate it.
c) The model number of the CNN must be changed every time it is re-trained. line 117: model_fn=cnn_model_fn, model_dir="/tmp/comp551_convnet_model212")
- change "/tmp/comp551_convnet_model212" to something else such as "/tmp/comp551_convnet_model213"
d) Write the following in a command line prompt
cd ~/ARBITRARY_PATH/comp551-project3-master python Convolutional-Neural-Net/predictCNN.py -
Run the NN predictor
cd ~/ARBITRARY_PATH/comp551-project3-master python neural-network/nn.py
DESCRIPTION OF DIRECTORIES AND FILES_______
Directory: preprocess
File: pickle_data.py
This script reads in the data in csv form, preprocesses it, and saves it as a .pkl file. These are larger, but are in the numpy array form so they are faster to read in. It saves the pkl files in the 'data' directory.
File: image_processing.py
This contains the methods for preprocessing the images used in pickle_data.py
Directory: Convolutional-Neural-Net
File: predictCNN.py
This script reads in the pkl files, and trains a CNN. When it has finished training it will output a result file
to the directory Convolutional-Neural-Net