-
Notifications
You must be signed in to change notification settings - Fork 1
/
plot.py
178 lines (164 loc) · 6.34 KB
/
plot.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
'''
Plot runs.
'''
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d.axes3d import Axes3D
from learn import *
import numpy as np
import pickle
import simulator
import scipy.stats as t
def average_return(returns):
''' Compute the average return for a set of episodes. '''
total = 0
values = np.zeros(returns.shape)
for i in range(returns.size):
total += returns[i]
values[i] = total / (i + 1)
return values
def sliding_window(returns):
count = 0.0
prob = np.zeros(returns.shape)
window = 1000.0
for i in range(returns.size):
if i > window:
if returns[i-window] == 0:
count -= 1.0
denom = window
else:
denom = (i + 1)
if returns[i] == 0:
count += 1.0
prob[i] = count / denom
return prob
def plot_return(agent, returns, data = None):
''' Plot return over time. '''
plt.plot(returns, agent.colour, label = agent.legend)
n = returns.size
interval = 2000
if data != None:
for i in range(n/interval):
plt.errorbar(1+i*interval, returns[i*interval],
yerr = t.sem(data[:, i*interval]), fmt= agent.colour)
plt.axis([0, returns.size, -10, 20])
plt.xlabel('Episodes')
plt.title('Average Return')
plt.ylabel('Average Return')
def plot_run(agent, run):
''' Plot a single run. '''
item_name = './runs/'+agent.name+'/'+str(run)
returns = average_return(np.load(item_name + '.npy'))
plot_return(agent, returns)
plt.savefig(item_name, bbox_inches='tight')
def plot_average_goals(agent, returns, data = None):
''' Plot the average goals over time. '''
prob = average_return(returns)
n = returns.size
interval = 2000
if data != None:
for i in range(data.shape[0]):
data[i, :] = average_return(data[i, :])
for i in range(n/interval):
plt.errorbar(1+i*interval, prob[i*interval],
yerr = t.sem(data[:, i*interval]), fmt= agent.colour)
plt.plot(prob, agent.colour, label = agent.legend)
plt.title('Goal Scoring Probability')
plt.axis([0, prob.size, 0.0, 0.475])
plt.xlabel('Episodes')
plt.ylabel('Probability')
def goal_count(returns):
''' Boolean array for returns based on goals. '''
return 1. * np.array([(val == 50.) for val in returns])
def plot_goals(agent, run):
''' Plot a single run. '''
item_name = './runs/'+agent.name+'/'+str(run)
returns = goal_count(np.load(item_name + '.npy'))
plot_average_goals(agent, returns)
plt.savefig(item_name+'_gb', bbox_inches='tight')
def plot_return_agents(agents, max_runs, runs = 20):
''' Plot all the average returns for all agents. '''
plt.clf()
for agent in agents:
returns = np.zeros((max_runs,))
data = np.zeros((runs, max_runs))
for run in range(1, runs + 1):
ret = np.load('./runs/' + agent.name + '/' + str(run) + '.npy')
ret = average_return(ret[:max_runs])
returns += ret / runs
data[run-1, :] = ret
plot_return(agent, returns, data)
plt.legend(loc = 'upper left')
plt.savefig('./runs/return', bbox_inches='tight')
def plot_goals_agents(agents, max_runs, runs = 20):
''' Plot all the goals for all agents. '''
plt.clf()
for agent in agents:
returns = np.zeros((max_runs,))
data = np.zeros((runs, max_runs))
for run in range(1, runs + 1):
ret = np.load('./runs/'+agent.name + '/' + str(run) + '.npy')
print ret.shape
ret = goal_count(ret[:max_runs])
returns += ret / runs
data[run-1, :] = ret
plot_average_goals(agent, returns, data)
#plt.plot(0.2*np.ones(max_runs), '--k', label = 'SARSA($\\theta_0$)')
plt.legend(loc = 'upper left')
plt.savefig('./runs/goals', bbox_inches='tight')
def plot_action_weights(weights, name, run):
''' Plot the action weights as a function of x,y. '''
plt.clf()
fig = plt.figure()
plot = fig.add_subplot(111, projection = '3d')
xxrange = np.arange(0, 30, 0.5)
yyrange = np.arange(-15, 15, 0.5)
xgrid, ygrid = np.meshgrid(xxrange, yyrange)
colours = [[1,0,0],[0,1,0],[0,0,1]]
for weight, colour in zip(weights, colours):
function = lambda x,y: weight.dot(fourier_basis(np.array([x, y, 0, 0, 0])))
zarray = np.array([function(x, y) for x,y in zip(np.ravel(xgrid), np.ravel(ygrid))])
zgrid = zarray.reshape(xgrid.shape)
plot.plot_surface(xgrid, ygrid, zgrid, color = colour)
plot.set_xlabel('x')
plot.set_ylabel('y')
plot.set_zlabel('Action-Value')
plt.savefig('./runs/'+name+'/'+str(run)+'_weight', bbox_inches='tight')
def plot_episode(agent, run):
''' Plot an example run. '''
with file('./runs/'+agent.name+'/'+str(run)+'.obj', 'r') as file_handle:
agent = pickle.load(file_handle)
sim = simulator.Simulator()
import interface
agent.run_episode(sim)
interface.Interface().draw_episode(sim, 'after')
def plot_policy(agent, run):
''' Plot policies. '''
with file('./runs/'+agent.name+'/'+str(run)+'.obj', 'r') as file_handle:
agent = pickle.load(file_handle)
plot_action_weights(agent.action_weights, agent.name, run)
def print_policy(agent, run):
with file('./runs/'+agent.name+'/'+str(run)+'.obj', 'r') as file_handle:
agent = pickle.load(file_handle)
print agent.action_weights
print agent.parameter_weights
print agent.alpha
def plot_vector_field(agent):
''' Plot policies. '''
agent = agent()
X, Y = np.meshgrid(np.arange(0, PITCH_LENGTH/2, 0.5), np.arange(-PITCH_WIDTH/2, PITCH_WIDTH/2, 0.5))
U, V = np.zeros(X.shape), np.zeros(Y.shape)
for i in range(X.shape[0]):
for j in range(X.shape[1]):
ball = vector(X[i, j], Y[i,j])
keeper = vector(PITCH_LENGTH/2-5.5, 0)
state = np.zeros((14,))
state[10:12] = ball
state[5:7] = keeper
feat = ball_features(state)
kickto = agent.parameter_weights[0].T.dot(to_matrix(feat))
U[i, j] = kickto[0, 0] - ball[0]
V[i, j] = kickto[1, 0] - ball[1]
Q = plt.quiver(X[::3, ::3], Y[::3, ::3],10*U[::3, ::3],10*V[::3, ::3], pivot='tail', color='r')
qk = plt.quiverkey(Q, 0.5, 0.03, 1, 'T')
plt.plot(X[::3,::3], Y[::3, ::3], 'k.')
plt.savefig('./runs/vf')