README.txt

description of parameters:

--mode:
	test: test the model, need label file.

	train: train the model and save model in pca_model.sav and clf_model.sav. These two files will saved in /app/model, so you can mount local disk folder to /app/model to save it to local disk.. label file is needed.

	test_predict: test the model and the prediction.

	predict: load model from the model files and predict data. Must run after model is generated by train.

--folder_path
	This should be the path of folder in which the images are stored. Default is /app/data. So using mount (-v in docker) to mount images in local disk to /app/data then it can work.

--folder__mode
	This shows whether the program run through all sub-directories under the folder path. 1 is the default, which is run all sub-directories. 0 is only the current folder.

--label
	This should be the file contains label information. The order of label should be same to the order in file_list.txt, in which saves all image files paths and names.

--image_path
	This is only used in predict mode. This is the path of the image which you want to predict.

--model_path
	This is the path to the folder in which saves the model. The default is model/ . You can just mount the model_folder, which contains pca_model.sav and clf_model.sav, to /app/model so that the program can work. This is also the path in which will save the trained model.

--prediction_file
	This is the path to the folder in which will save the prediction result. The default is prediction/prediction.json . So you can just mount the prediction folder in local disk to /app/prediction so the prediction.txt will save to local disk.

--resize_to
	This is the parameter which decide the size of intermediate image. The program will first resize the original image to square image whose size is "resize_to*resize_to", then put it into pca model.

--pca_components
	This is the parameter which decide the final size of features. If this is changed, the model need re-training.