random_main.py

import numpy as np
import os
from prototypes.random_based_mcmc1 import random_MCMC
import sys

if __name__ == '__main__':
	home = os.path.expanduser('~')
	dataset = sys.argv[1]
	place = sys.argv[2]  # Documents/research for beeblebrox; barn for CCNI
	data_home = home + '/' + place + '/EP_project/data/'
	results_home = home + '/' + place + '/stochastic_search/results/'
	num_iters = 51
	start = int(sys.argv[3])
	end = int(sys.argv[4])
	# Gradient calculation
	# Empty features directory
	import glob
	files = glob.glob(data_home + 'features/random1/*.npz')
	for f in files:
		if os.path.exists(f):
			os.remove(f)
	pipeline = {}
	pipeline['feature_extraction'] = ["VGG", "haralick", "inception", "naive_feature_extraction"]
	pipeline['dimensionality_reduction'] = ["PCA", "ISOMAP", "naive_dimensionality_reduction"]
	pipeline['learning_algorithm'] = ["SVM", "RF", "naive_learning_algorithm"]
	pipeline['haralick_distance'] = range(1, 4)
	pipeline['pca_whiten'] = [True, False]
	pipeline['n_neighbors'] = range(3, 8)
	pipeline['n_components'] = range(2, 5)
	pipeline['n_estimators'] = np.round(np.linspace(8, 300, 10))
	pipeline['max_features'] = np.arange(0.3, 0.8, 0.1)
	pipeline['svm_gamma'] = np.linspace(0.01, 8, 10)
	pipeline['svm_C'] = np.linspace(0.1, 100, 10)

	# CONTROL
	type1 = 'random_MCMC'
	for i in range(start, end):
		rm = random_MCMC(data_name=dataset, data_loc=data_home, results_loc=results_home, run=i+1, type1=type1, pipeline=pipeline, iters=num_iters)
		rm.populate_paths()
		rm.randomMcmc()