figures.py

#!/bin/env python

'''
generates figures used in the paper
'''
import matplotlib.pyplot as plt
import numpy as np
from os.path import join
import ipdb as pdb

def plot_val_loss(data_dir):
  # single test
  tmp = np.load(join(data_dir,'e2c-val-loss.npz'))
  L_mean=tmp['mean'] # mean validation loss
  L_std=tmp['std'] # std validation loss
  T = range(L_mean.size)
  plt.errorbar(T[::4], L_mean[::4], L_std[::4],label='validation loss')
  L_train=np.load(join(data_dir,'l_hist.npy'))
  B=100
  plt.plot(range(2,L_mean.size),L_train[2*B::B],label='train loss') # minibatch length
  plt.ylim([0,650])
  plt.xlabel('Episode')
  plt.legend()
  return

if __name__ == '__main__':
  # data needs to be generated by plane_analyze.py first
  data_dir='/ltmp/e2c-plane2-adaptive'
  #data_dir='/ltmp/e2c-plane2-rand/'
  plot_val_loss(data_dir)
  #plt.title('Random Policy')
  #plt.savefig('rand_post.png')
  plt.show()