def run(run_on_loop, num_iterations):
"""
handles running perceptron in two ways, on a loop or just once
preliminary setup work is done independent of the mode
args: run_on_loop is a boolean indicating whether to loop or not; num_iterations is how many times to train
"""
# create feature space
features = Process.process_features()
# convert train data to vectors
train_data = Process.make_vectors("train.positive", features, 1)
train_data.extend(Process.make_vectors("train.negative", features, -1))
# convert test data to vectors
test_data = Process.make_vectors("test.positive", features, 1)
test_data.extend(Process.make_vectors("test.negative", features, -1))
# train perceptron - returns weights, biases and errors from each iteration
print "Training..."
train_results = Perceptron.train(train_data, num_iterations)
print "End of training."
print "----"
# grab the information returned by the training
weights_arr = train_results.get("weights")
bias_arr = train_results.get("bias")
train_errors = train_results.get("errors")
test_errors = []
print "Testing..."
if (run_on_loop):
# loop and graph mode, runs perceptron until reaches num_iterations
# each iteration uses a different set of weights and bias, corresponding to those produced in training
n = 0
while n < num_iterations:
print "Iteration [{0}]".format(n+1)
# run test perceptron and save results in array for plotting
test_errors.append(Perceptron.test(test_data, weights_arr[n], bias_arr[n]))
n += 1
# plot graph of train and test errors
Process.plot_results(train_errors, test_errors)
else:
# simple mode, runs perceptron once
# uses only the final weight vector and bias
test_errors = Perceptron.test(test_data, weights_arr[-1], bias_arr[-1])
print "End of testing."
print "----"
pass
