[0, 0, 0],
[2, 1, 0],
])
partitions = np.vstack((day1partitions, day1partitions, day1partitions,
day4partitions, day5partitions, day5partitions))
#%%
alphas = np.power(10., np.arange(1, -50, -1))
posts = np.empty((len(alphas), 3)) * np.nan
for idx, alpha in enumerate(alphas):
animal = Animal(alpha, observationModelClass=Gauss)
for i in range(observations.shape[0]):
animal.experience(observations[i, :])
for i in range(3): # for each partition
posts[idx, i] = animal.partition_prob(partitions[:, i])
posts
#%%
figInd += 1
#%%
fig = plt.figure(figInd)
fig.clear()
bar_width = 0.2
idxs = np.array([0, 15, 30]) + 1 #+5
idxs = np.array([0, 20, 40]) + 1 #+5
for i in range(3):
plt.bar(
np.array([1, 2, 3]) + (i - 1) * bar_width, posts[idxs, i], bar_width)
#plt.plot(np.array([1,2,3])+(i-1)*bar_width/10, posts[idxs,i], 'x')
labels = []
np.pi).astype(int) # 2 clusters, animal1
partition2_1 = np.zeros((nExperiences, )) # 1 cluster, animal2
partition2_2 = np.mod(animal2.experiences[:, 0] - midpoint2,
2 * np.pi) > np.pi # 2 clusters, animal2
#%%
# partition_prob
figInd += 1
figInd = 2
fig = plt.figure(figInd)
fig.clear()
ax = plt.subplot(1, 2, 1)
bar_width = 0.2
inds = np.array([0, 1]) - bar_width / 2
rat1 = animal1.partition_prob(partition1_1) - animal1.partition_prob(
partition1_2)
rat2 = animal1.partition_prob(partition2_1) - animal1.partition_prob(
partition2_2)
plt.bar(np.array([-bar_width / 2, +bar_width / 2]),
-np.array([rat2, rat1]),
bar_width,
color='rb')
m = np.max(np.abs(plt.ylim()))
plt.ylim([-m, m])
adjustPlot(ax, fuzzyzero=True)
plt.xticks(np.array([-bar_width / 2, +bar_width / 2]), ['Random', 'Directed'])
plt.ylabel('Negative Partition Evidence Ratio')
plt.savefig('figures/' + filename + '-' + str(figInd) + '.pdf',
bbox_inches='tight')
def experiment(alpha=alpha_default,
sd0=sd0_default,
sd=sd_default,
N=N_default,
distance=distance_default):
np.random.seed(1)
_model = WorldModel(CRP(alpha), Gauss(sigma_0=sd0))
_animal = Animal(worldModel=_model)
experiences = np.nan * np.ones((N * nPresentations * 2))
partition1 = np.array([])
partition2 = np.array([])
partitionN = np.array([])
lps = np.nan * np.empty((N, 3))
posts = np.nan * np.empty((N * nPresentations * 2, 3))
similarity = np.nan * np.empty((N, 1))
for i in range(N):
c1 = stats.norm(loc=np.array([-distance / 2]),
scale=sd * np.ones((1, K)))
c2 = stats.norm(loc=np.array([+distance / 2]),
scale=sd * np.ones((1, K)))
for _i in range(nPresentations):
experience1 = c1.rvs()
experiences[i * nPresentations * 2 + _i * 2] = experience1
post1 = _animal.experience(experience1)
if i + _i == 0:
partition1 = np.array([[0]])
partition2 = np.array([[0]])
partitionN = np.array([[0]])
else:
partition1 = np.vstack((partition1, 0))
partition2 = np.vstack((partition2, 0))
partitionN = np.vstack(
(partitionN, 2 * i * nPresentations + _i))
if i > 0:
posts[i * nPresentations * 2 + _i, :] = post1.squeeze()
_animal.clusterAssignments = partition2
for _i in range(nPresentations):
experience2 = c2.rvs()
experiences[i * nPresentations * 2 + _i * 2 + 1] = experience2
post2 = _animal.experience(experience2)
partition1 = np.vstack((partition1, 0))
partition2 = np.vstack((partition2, 1))
partitionN = np.vstack(
(partitionN, (2 * i + 1) * nPresentations + _i))
if i > 0:
posts[i * nPresentations * 2 + nPresentations +
_i, :] = post2.squeeze()
_animal.clusterAssignments = partition2
#TODO: look over this for what to do about partitions2activity
ret = partitions2activity(_animal, experience1,
[partition1, partition2],
[zeta0, zeta1A, zeta1B])
activity1 = ret[0]
tmp_lps = ret[1]
ret = partitions2activity(_animal, experience2,
[partition1, partition2],
[zeta0, zeta1A, zeta1B])
activity2 = ret[0]
tmp_lps_2 = ret[1]
np.testing.assert_array_equal(tmp_lps, tmp_lps_2)
lps[i, 0:2] = tmp_lps
lps[i, 0] = _animal.partition_prob(partition1)
lps[i, 1] = _animal.partition_prob(partition2)
lps[i, 2] = _animal.partition_prob(partitionN)
similarity[i] = np.corrcoef(activity1, activity2)[0, 1]
return similarity, lps, _animal, partition1, partition2, c1, c2, experiences, posts