def plot_map_of_elites(self, perfs, log_dir_name):
"""
Plot a heatmap of elites
"""
plot_heatmap(perfs,
self.feature_dimensions[1].name,
self.feature_dimensions[0].name,
savefig_path=log_dir_name)
def plot_map_of_elites(self):
"""
Plot a heatmap of elites
"""
plot_heatmap(self.performances,
self.feature_dimensions[1].name,
self.feature_dimensions[0].name,
savefig_path=self.log_dir_path,
)
def plot_map_of_elites(self, iteration):
"""
Plot a heatmap of elites
"""
# Stringify the bins to be used as strings in the plot axes
x_ax = [str(d) for d in self.bins['bin_{}'.format(self.descriptors[5])]]
y_ax = [str(d) for d in self.bins['bin_{}'.format(self.descriptors[6])]]
#x_ax = [str(d) for d in self.bins['bin_{}'.format(self.descriptors[0])]]
#y_ax = [str(d) for d in self.bins['bin_{}'.format(self.descriptors[1])]]
plot_heatmap(self.performances,
x_ax,
y_ax,
v_min=0.0,
v_max=5.0,
savefig_path=self.log_dir_path,
iteration=iteration,
title=f"MAP-Elites for the city of Boston",
**self.plot_args)
if len(diff) != 1:
name_parts.append("differences" if run.diff ==
"diff" else "no differences")
ddiff = "a" if run.diff == "nodiff" else "b"
data = datas.setdefault(
(run.dataset, run.bounds, run.method, ddiff), {
"name": " ".join(name_parts),
"runs": set()
})
data["runs"].add(run)
if figure == "success":
plot_heatmap(datas,
fig,
map2success,
"Successes (out of 5)",
flat=args.flat,
grid=grid,
vmin=0,
vmax=5)
elif figure == "normdist":
plot_heatmap(datas,
fig,
map2normdist,
"Normdist",
flat=args.flat,
grid=grid)
elif figure == "row":
plot_heatmap(datas, fig, map2row, "Row", flat=args.flat, grid=grid)
elif figure == "blocks":
plot_heatmap(datas,
fig,
"#FF8031",
"#02D4F9",
"#4F4C4B",
]
sns.set_palette(flatui)
sns.palplot(sns.color_palette())
colors = plt.rcParams['axes.prop_cycle'].by_key()['color']
if not os.path.exists('figures'):
os.makedirs('figures')
plot_utils.plot_heatmap(
'saves/mnist_grid/grid_mnist_mlp_lr:inverse-time-decay_sgdm_T_5000_N_100/seed_3/result.pkl',
key='train_sum_loss',
xlabel='Decay',
ylabel='Initial LR',
cmap=plt.cm.Purples_r,
levels=10,
figsize=(8, 6),
)
plt.xticks([-2, -1, 0, 1, 2], fontsize=18)
plt.yticks([-3, -2, -1, 0, 1], fontsize=18)
es_K100 = plot_utils.load_log(
'saves/mnist_lr_decay/train_sum_loss/es-mnist-mlp-obj:train_sum_loss-tune:lr:itd-T:5000-K:100-N:100-sigma:0.1-olr:0.1-seed:3',
fname='frequent.csv')
es_K10 = plot_utils.load_log(
'saves/mnist_lr_decay/train_sum_loss/es-mnist-mlp-obj:train_sum_loss-tune:lr:itd-T:5000-K:10-N:100-sigma:0.1-olr:0.01-seed:3',
fname='frequent.csv')
es_K1 = plot_utils.load_log(
sns.palplot(sns.color_palette())
colors = plt.rcParams['axes.prop_cycle'].by_key()['color']
figure_dir = 'figures/mnist_heatmaps'
if not os.path.exists(figure_dir):
os.makedirs(figure_dir)
# Plot train loss heatmap
# ---------------------------------------------------------------------------
plot_utils.plot_heatmap(
'saves/mnist_grid_sgdm/grid_mnist_mlp_lr:inverse-time-decay_sgdm_T_5000_N_40/seed_3/result.pkl',
key='unroll_obj',
xlabel='Log LR Decay',
ylabel='Log Init LR',
cmap=plt.cm.Purples_r,
levels=30,
sigma=1.0,
use_smoothing=True,
show_contours=True,
contour_alpha=0.2,
figsize=(8, 6),
)
es_K10 = plot_utils.load_log(
'saves/mnist_lr_decay/train_sum_loss/es-mnist-mlp-obj:train_sum_loss-tune:lr:inverse-time-decay-T:5000-K:10-nc:1-npc:1000-sigma:0.1-olr:0.01-ob1:0.9-ob2:0.999-ic:-1-oc:-1-seed:3',
fname='frequent.csv')
pes_K10 = plot_utils.load_log(
'saves/mnist_lr_decay/train_sum_loss/pes-mnist-mlp-obj:train_sum_loss-tune:lr:inverse-time-decay-T:5000-K:10-nc:1-npc:1000-sigma:0.1-olr:0.01-ob1:0.9-ob2:0.999-ic:-1-oc:-1-seed:3',
fname='frequent.csv')
es_K100 = plot_utils.load_log(
def ftr():
print("Loading data...\n")
X = np.loadtxt('X.txt')
y = np.loadtxt('y.txt')
print("Randomly partitioning into training and test sets...\n")
n_train = 800
X_train, y_train, X_test, y_test = randomly_partition(X, y, n_train)
l_arr = [0.1, 0.35]
#l_arr = [0.2, 0.25, 0.30, 0.35, 0.40]
#l_arr = [0.1, 0.3, 0.5, 0.7, 0.9]
var_0_arr = [1, 0.5]
#var_0_arr = [0.45, 0.50, 0.55, 0.60, 0.65]
#var_0_arr = [0.1, 0.3, 0.5, 0.7, 0.9]
N = len(l_arr)
M = len(var_0_arr)
Z_matrix = np.zeros(shape=(N, M))
for i in range(0, N):
for j in range(0, M):
l = l_arr[i]
var_0 = var_0_arr[j]
print("( l , var_0 ) = ( {} , {} )".format(l, var_0))
X_tilde_train = get_x_tilde(
evaluate_basis_functions(l, X_train, X_train))
X_tilde_test = get_x_tilde(
evaluate_basis_functions(l, X_test, X_train))
w_map, log_Z, predict_laplace = find_map(X_tilde_train, y_train,
var_0)
Z_matrix[i][j] = log_Z
print("Log Z = {}\n".format(log_Z))
verbose = True
if (verbose):
predict_map = lambda X: predict(X_tilde=X, w=w_map)
probs_test_laplace = predict_laplace(X_tilde_test)
display_metrics(probs_test_laplace, y_test, "Test laplace")
probs_train_laplace = predict_laplace(X_tilde_train)
display_metrics(probs_train_laplace, y_train, "Train laplace")
probs_test_map = predict_map(X_tilde_test)
display_metrics(probs_test_map, y_test, "Test map")
probs_train_map = predict_map(X_tilde_train)
display_metrics(probs_train_map, y_train, "Train map")
expansion_function = lambda x: evaluate_basis_functions(
l, x, X_train)
plot_predictive_general(
X,
y,
predict_map,
expansion_function,
title="MAP predictions (s^2, l)=({}, {})".format(var_0, l))
plot_predictive_general(
X,
y,
predict_laplace,
expansion_function,
title="Laplace predictions (s^2, l)=({}, {})".format(
var_0, l))
plot_heatmap(Z_matrix, var_0_arr, l_arr)