# Setting up python

module avail python
module load version


# virtualenv

source ENV/bin/activate
deactivate

# June 14th  

The new batch are in png form, and they are in the following location

data/newRescored/pngsNeedConvert

The fully sorted and rescored data is here:

	data/newRescored
with the train and test sets in that dir. 

Everything has been fully renamed correctly as in everything in a pos directory is n1,
 and everything in a neg is n0.

* convert pngs add them to the full dataset, and then re-run the model. 
You should have now 9K images.


# June 15th 

Data is now fully converted over to jpgs in a dir location:
data/newRescored/pngsNeedConvert/raw-data


# June 15th 

renamed some directories, everything is basically lower case now.

* wrote a new image conversion script image2square which pads a image
up to a square based on largest dimension.
* ran the conversion on all the png
* split the converted images into pos and neg, and 
* the new directory to be trained from can be found in.

	data/new-rescored/train

* new images were only added to the training directory not test.

## June 16th 

All images less than 200x200 have been removed, keep this in mind for future. 

## June 24th

The script datasetup takes two parameters now.

usage:

./datasetup n1 .93, 

Means to seperate positive examples that have n1 and negative examples that 
don't. .93 means to use 93% of the data as training.


## July 12th 

data-06-29-18 batch added, needs to be processed. 

## July 17th. 


Old images (unpadded but wrongly labelled) are in data/unrescored/raw-data 

New Images (padded but correctly labelled) are in images 

Problem, there are 100 more in images, that are not in data/unrescored/raw-data. delete these. 

# September 14th

Do a per class precision, recall, and *accuracy* <- doctors on a
per class basis.

Example of what is needed

_| C | N | H
P|   |   |   <- Positive ACC
N|   |   |   <- Negative ACC
T|   |   |   <- Total ACC
N|   |   |   <- Number of images