-
Notifications
You must be signed in to change notification settings - Fork 0
/
posterior_plot.py
239 lines (203 loc) · 9.53 KB
/
posterior_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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
import numpy as np
import matplotlib.pyplot as plt
import os
import boirlscenarios.constants as constants
from scipy import stats
plt_xlabels = {constants.GRIDWORLD2D: "Steepness, " + r"$\theta_0$",
constants.VIRTBORLANGE: "Time, " + r"$\theta_0$",
constants.MAZE: "x-position, " + r"$\theta_0$",
constants.FETCH: "Distance Threshold, " + r"$\theta_0$"}
plt_ylabels = {constants.GRIDWORLD2D: "Midpoint, " + r"$\theta_1$",
constants.VIRTBORLANGE: "LeftTurn, " + r"$\theta_1$",
constants.MAZE: "y-position, " + r"$\theta_1$",
constants.FETCH: "Penalty, " + r"$\theta_0$"}
plt_xlims = {constants.GRIDWORLD2D: [-2, 2],
constants.VIRTBORLANGE: [-2.5, 0],
constants.MAZE: [-1, 1],
constants.FETCH: [0., 0.25]}
plt_ylims = {constants.GRIDWORLD2D: [-10, 10],
constants.VIRTBORLANGE: [-2.5, 0],
constants.MAZE: [-1., 1],
constants.FETCH: [-1, 1]}
plt_xticks = {constants.GRIDWORLD2D: [-2, 0, 2],
constants.VIRTBORLANGE: [-2.5, 1.2, 0],
constants.MAZE: [-1, 0, 1],
constants.FETCH: [0, 0.12, 0.25]}
plt_yticks = {constants.GRIDWORLD2D: [-10, 0, 10],
constants.VIRTBORLANGE: [-2.5, 1.2, 0],
constants.MAZE: [-1, 0, 1],
constants.FETCH: [-1, 0., 1]}
plt_gt = {constants.GRIDWORLD2D: np.array([[1.25, 5]]),
constants.VIRTBORLANGE: np.array([[-2., -1]]),
constants.MAZE: np.array([[0.3, 0.5]]),
constants.FETCH: np.array([[0.05, -1]])}
baseDir = "PosteriorPlotsFromPaper"
resultsDir = os.path.join(baseDir, "Data")
def prep_data(loaddir, trial, is_ours=True, should_truncate=False):
if is_ours:
# Load X and Y positions
X1 = np.load(os.path.join(loaddir, "X1%d.npy") % trial)
X2 = np.load(os.path.join(loaddir, "X2%d.npy") % trial)
x1, x2 = np.meshgrid(X1, X2)
X = np.hstack((x1.reshape((len(X1) * len(X1), 1)), x2.reshape((len(X1) * len(X1), 1))))
# Load the m
m = np.load(os.path.join(loaddir, "m%d.npy") % trial).squeeze()
# Load the v
v = np.load(os.path.join(loaddir, "v%d.npy") % trial).squeeze()
# Load the ground truth
gt_lik = np.load(os.path.join(loaddir, "gt_lik.npy"), allow_pickle=True).tolist()
# Load learned x and y
x_choices = np.load(os.path.join(loaddir, "x_choices.npy")).squeeze()
y_choices = np.load(os.path.join(loaddir, "y_choices.npy"))
# Find those corresponding to the current trial under investigation
trialtracks = np.load(os.path.join(loaddir, "trial_tracks.npy"))
trinds = np.where(trialtracks == trial)[0]
bo_valid_x = x_choices[trinds]
bo_valid_y = y_choices[trinds]
#
if should_truncate:
trunc_ind = np.where(X[:, 0] < -0.6)[0]
newX = X[trunc_ind]
newm = m[trunc_ind]
newv = v[trunc_ind]
trunc_ind_bodata = np.where(bo_valid_x[:, 0] < -0.6)[0]
bo_valid_x = bo_valid_x[trunc_ind_bodata]
bo_valid_y = bo_valid_y[trunc_ind_bodata]
xlims = plt_xlims[env]
xlims[1] = -0.6
plt_xlims[env] = xlims
xticks = plt_xticks[env]
xticks[2] = -0.6
plt_xticks[env] = xticks
else:
newX = X
newm = m
newv = v
broadnearInds = np.where(bo_valid_y <= 1.1 * gt_lik)[0]
broadfarInds = np.where(bo_valid_y > 1.01 * gt_lik)[0]
nearInds = np.where(bo_valid_y <= 1.01 * gt_lik)[0]
farInds = np.where(bo_valid_y >= gt_lik)[0]
notgreatInds = np.intersect1d(broadnearInds, broadfarInds)
goodishInds = np.intersect1d(nearInds, farInds)
bestInds = np.where(bo_valid_y < gt_lik)[0]
bo_notgreat_xdata = bo_valid_x[notgreatInds]
bo_goodish_xdata = bo_valid_x[goodishInds]
bo_best_xdata = bo_valid_x[bestInds]
bo_notgreat_ydata = bo_valid_y[notgreatInds].squeeze()
bo_goodish_ydata = bo_valid_y[goodishInds].squeeze()
bo_best_ydata = bo_valid_y[bestInds].squeeze()
else:
xmin = plt_xlims[env][0]
xmax = plt_xlims[env][1]
ymin = plt_ylims[env][0]
ymax = plt_ylims[env][1]
X, Y = np.mgrid[xmin:xmax:100j, ymin:ymax:100j]
newX = np.vstack([X.ravel(), Y.ravel()])
weights = np.load(os.path.join(loaddir, "rewards%d.npy") % trial)
weights = weights.T
kernel = stats.gaussian_kde(weights)
newm = kernel(newX).T
newv = None
newX = newX.T
bo_goodish_xdata = None
bo_best_xdata = None
bo_notgreat_xdata = None
return newX, newm, newv, bo_notgreat_xdata, bo_goodish_xdata, bo_best_xdata,
def plot_me(pos, env, algo, val, notgreat, goodish, best, is_ours, trial=0, ismean=True):
markersize = 1000
savedir = os.path.join(baseDir, "Plots", env)
if not is_ours:
savedir = os.path.join(savedir, "BIRL")
os.makedirs(savedir, exist_ok=True)
fontlabs = {'family': 'serif',
'color': 'black',
'weight': 'normal',
'size': 50,
'verticalalignment': 'center',
'horizontalalignment': 'center'
}
plt.close("all")
plt.figure(1, figsize=(16.75, 13.5), dpi=80)
if is_ours:
cmap = "viridis"
else:
cmap = "viridis_r"
p1 = plt.tricontourf(pos[:, 0], pos[:, 1], val.squeeze(), 30, vmin=np.amin(val.squeeze()),
vmax=np.amax(val.squeeze()), cmap=cmap)
plt.xlabel(plt_xlabels[env], fontdict=fontlabs, labelpad=30)
plt.ylabel(plt_ylabels[env], fontdict=fontlabs, labelpad=15)
plt.xticks(plt_xticks[env], fontsize=30)
plt.yticks(plt_yticks[env], fontsize=30)
cbar = plt.colorbar(p1, ticks=[np.amax(val), np.amin(val), (np.amax(val) + np.amin(val)) / 2.], format='%.1f')
cbar.ax.tick_params(labelsize=30)
if env == constants.GRIDWORLD2D:
if is_ours:
clr = "white"
else:
clr = "black"
plt.annotate(r"$\theta_2$=0", (-1.8, -9), fontsize=40, c=clr)
elif env == constants.VIRTBORLANGE:
plt.annotate(r"$\theta_2$=-1", (-2.4, -2.4), fontsize=40, c="black")
plt.axis((plt_xlims[env][0], plt_xlims[env][1], plt_ylims[env][0], plt_ylims[env][1]))
if notgreat is not None:
plt.scatter(notgreat[:, 0], notgreat[:, 1], edgecolors="r", marker='o', facecolors="none", linewidths=3,
s=markersize, zorder=2)
if goodish is not None:
plt.scatter(goodish[:, 0], goodish[:, 1], c='r', s=markersize, zorder=2)
if best is not None:
plt.scatter(best[:, 0], best[:, 1], marker="X", c="red", s=markersize, zorder=2)
if not (env == constants.FETCH):
plt.scatter(plt_gt[env][:, 0], plt_gt[env][:, 1], marker="*", c="lime", s=1500, zorder=1)
fname = "Posterior_BIRL_%d.png" % trial if not is_ours else (
"Posterior_mean_%s_%d.png" % (algo, trial) if ismean else "Posterior_std_%s_%d.png" % (algo, trial))
plt.savefig(os.path.join(savedir, fname), bbox_inches="tight")
plt.close("all")
def execute_plot(env, algo, load_dir, trial, should_truncate=False, is_ours=True):
pos, val, val_dev, notgreat, goodish, best = prep_data(loaddir=load_dir, trial=trial, is_ours=is_ours,
should_truncate=should_truncate)
plot_me(pos=pos, env=env, algo=algo, val=val, notgreat=notgreat, goodish=goodish, best=best, is_ours=is_ours,
trial=trial)
if val_dev is not None:
plot_me(pos=pos, env=env, algo=algo, val=val_dev, notgreat=notgreat, goodish=goodish, best=best, is_ours=is_ours,
trial=trial, ismean=False)
## Gridworld3d
env = constants.GRIDWORLD2D
algo = "rhorbf"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=3)
algo = "rbf"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=3)
algo = "matern"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=3)
## Vborlange
env = constants.VIRTBORLANGE
algo = "rhorbf"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=0, should_truncate=True)
algo = "rbf"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=0, should_truncate=True)
algo = "matern"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=0, should_truncate=True)
## MAZE
env = constants.MAZE
algo = "rhorbf"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=1)
algo = "rbf"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=1)
algo = "matern"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=1)
## FETCH
env = constants.FETCH
algo = "rhorbf"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=3)
algo = "rbf"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=3)
algo = "matern"
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, env, "%s")%algo, trial=3)
algo="birl"
# BIRL
env = constants.GRIDWORLD2D
for i in range(10):
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, "BayesianIRL", env), trial=i, is_ours=False)
# BIRL
env = constants.VIRTBORLANGE
for i in range(10):
execute_plot(env=env, algo=algo, load_dir=os.path.join(resultsDir, "BayesianIRL", env), trial=i, is_ours=False)