def init():
U = np.stack([emb[cond==i,:].mean(0) for i in np.unique(cond)])
qq = len(np.unique(cond))
for ix in combinations(range(qq),2):
ax.plot(U[ix,0],U[ix,1],U[ix,2],color=(0.5,0.5,0.5))
# ax.plot(U[[1,3],0],U[[1,3],1],U[[1,3],2],color=(0.5,0.5,0.5))
# ax.plot(U[[3,2],0],U[[3,2],1],U[[3,2],2],color=(0.5,0.5,0.5))
# ax.plot(U[[2,0],0],U[[2,0],1],U[[2,0],2],color=(0.5,0.5,0.5))
# ax.plot(U[[0,3],0],U[[0,3],1],U[[0,3],2],color=(0.5,0.5,0.5))
# ax.plot(U[[1,2],0],U[[1,2],1],U[[1,2],2],color=(0.5,0.5,0.5))
ax.scatter(U[:,0],U[:,1],U[:,2],s=50, marker='s',c=np.unique(cond))
scat = ax.scatter(emb[:,0],emb[:,1], emb[:,2], c=colorby)
util.set_axes_equal(ax)
ax.set_xticklabels([])
ax.set_yticklabels([])
ax.set_zticklabels([])
# plt.xticks([])
# plt.yticks([])
# plt.zticks([])
# plt.legend(np.unique(cond), np.unique(cond))
cb = plt.colorbar(scat,
ticks=np.unique(colorby),
drawedges=True,
values=np.unique(colorby))
cb.set_ticklabels(np.unique(colorby)+1)
cb.set_alpha(1)
cb.draw_all()
return fig,
# # z[pos,:] = fake_labels[pos,:]@basis.T
z = emb(torch.tensor(fake_labels).float()).numpy() + np.random.randn(
ndat, dim) * noise
#%% Visualize
x0 = -1 # training context
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
w = weight_func(fake_labels[:, 1], x0, sigma=0.5)
# w = w_tot
ax.scatter(z[:, 0], z[:, 1], z[:, 2], c=w, s=1)
util.set_axes_equal(ax)
#%% Compute
# x0 = -1 # training context
# x_tst = np.linspace(-0.9,0.9,25)
bw = 0.05
n_ps = 300 # subsample size for parallelism
b_ps = 1 # number of 'batches' for parallelism
eps_ps = 0.5 # epsilon for parallelism
n_neigh = 10
dist_weight = 0.0
x = fake_labels[:, 1]
y = fake_labels[:, 0]
