reduce_radius_after=20,
# 2 Means that neighbour neurons will have high learning
# rates during the first iterations
std=2,
# Defines a rate at which parameter `std` will be reduced.
# Reduction is monotonic and reduces after each epoch.
# In 50 epochs std = 2 / 2 = 1 and after 100 epochs
# std = 2 / 3 and so on.
reduce_std_after=50,
# Step (or learning rate)
step=0.3,
# Defines a rate at which parameter `step` will reduced.
# Reduction is monotonic and reduces after each epoch.
# In 50 epochs step = 0.3 / 2 = 0.15 and after 100 epochs
# std = 0.3 / 3 = 0.1 and so on.
reduce_step_after=50,
)
sofm.train(data, epochs=20)
red, blue = ('#E24A33', '#348ABD')
plt.scatter(*data.T, color=blue)
plt.scatter(*sofm.weight, color=red)
weights = sofm.weight.reshape((2, GRID_HEIGHT, GRID_WIDTH))
plot_2d_grid(weights, color=red)
plt.show()
n_columns = len(configurations)
plt.figure(figsize=(12, 5))
for index, conf in enumerate(configurations, start=1):
sofm = algorithms.SOFM(
n_inputs=2,
features_grid=(GRID_HEIGHT, GRID_WIDTH),
verbose=True,
shuffle_data=True,
grid_type=conf['grid_type'],
learning_radius=8,
reduce_radius_after=5,
std=2,
reduce_std_after=5,
step=0.3,
reduce_step_after=5,
)
sofm.train(data, epochs=40)
plt.subplot(1, n_columns, index)
plt.title(conf['title'])
plt.scatter(*data.T, color=blue, alpha=0.05)
plt.scatter(*sofm.weight, color=red)
weights = sofm.weight.reshape((2, GRID_HEIGHT, GRID_WIDTH))
plot_2d_grid(weights, color=red, hexagon=conf['use_hexagon_grid'])
plt.show()