plot.py

#!/usr/bin/python

import numpy as np
import matplotlib.pyplot as plt
import numpy as np

def learning_curve():
    Q = np.load("Data/2016_07_05-2_Q.npy")
    s = np.load("Data/2016_07_05-2_s.npy")
    
    # Average Q values
    qs = []
    for q in Q:
        qs.append(np.mean(q))
    
    fig, ax1 = plt.subplots(figsize=(6.5,5))
    
    # horizontal line
    ax1.plot([0,150], [5,5], color='0.7', linestyle='--', linewidth=2.5)
    
    # nb ob movements in each episode
    ax1.plot(s, 'k-')
    ax1.set_xlabel('Episodes')
    ax1.set_ylabel('Movements to terminal state', color='k')
    for tl in ax1.get_yticklabels():
        tl.set_color('k')

    # average q values
    ax2 = ax1.twinx()
    ax2.plot(qs, 'b-')
    ax2.set_ylabel('Average Q-values', color='b')
    for tl in ax2.get_yticklabels():
        tl.set_color('b')

    plt.xlim(0,150)
    plt.show()


def heatmap():
    fig, ax = plt.subplots(figsize=(6.5,5))
    heatmap = ax.pcolor(Q[-1], cmap=plt.cm.Blues, alpha=0.8)

    fig.colorbar(heatmap, ax=ax)

    fig = plt.gcf()
    #fig.set_size_inches(6.5,5)

    # turn off the frame
    ax.set_frame_on(False)

    # put the major ticks at the middle of each cell
    ax.set_yticks(np.arange(Q[-1].shape[0])+0.5, minor=False)
    ax.set_xticks(np.arange(Q[-1].shape[1])+0.5, minor=False)

    # want a more natural, table-like display
    ax.invert_yaxis()
    ax.xaxis.tick_top()

    labels = ['left', 'straight', 'right']

    ax.set_xticklabels(labels, minor=False)
    ax.set_yticklabels(range(10), minor=False)

    ax.set_xlabel('Actions')
    ax.set_ylabel('States')

    ax.grid(False)

    ax = plt.gca()

    for t in ax.xaxis.get_major_ticks(): 
        t.tick1On = False 
        t.tick2On = False 
    for t in ax.yaxis.get_major_ticks(): 
        t.tick1On = False 
        t.tick2On = False  

    plt.show()


def proxvalues():
    fig, ax = plt.subplots(figsize=(6.5,5))

    v = np.loadtxt("proximity_values.txt")
    x = np.arange(-18,18)*10

    colormap = plt.cm.coolwarm

    fig.gca().set_color_cycle([colormap(i) for i in np.linspace(0, 1.0, v.shape[1])])

    ax.plot(x,v)

    plt.xticks(np.arange(-8,9)*45)
    plt.xlim(-180,170)

    ax.set_xlabel('Boxposition [deg]')
    ax.set_ylabel('Proximity value')

    plt.show()
    
if __name__ == '__main__':
    learning_curve()