def _get_model(path):
with open(path, "rb") as fp:
contents = pickle.load(fp)
Js, Ks, gJs, gKs, converged, meta = contents
# Get best model.
J_best_mml, K_best_mml = grid_search.best(Js, Ks, meta["message_length"])
model = meta["best_models"]["mml"]
return model
def _get_model(path):
with open(path, "rb") as fp:
contents = pickle.load(fp)
Js, Ks, gJs, gKs, converged, meta, X_H, label_names = contents
#pickle.dump((Js, Ks, gJs, gKs, converged, meta, X_H, label_names), fp)
print(f"label_names from {path}: {label_names}")
# Get best model.
J_best_mml, K_best_mml = grid_search.best(Js, Ks, meta["message_length"])
model = meta["best_models"]["mml"]
return model
A_target = A_astrophysical
#A_target = A_original
#A_target[:-2, 0] = 0
#A_target[-2:, 0] = 1
stored_models = []
for i, comparison_model_path in enumerate(comparison_model_paths):
with open(comparison_model_path, "rb") as fp:
contents = pickle.load(fp)
Js, Ks, gJs, gKs, converged, meta, X_H, label_names = contents
# Get best model.
J_best_mml, K_best_mml = grid_search.best(Js, Ks, meta["message_length"])
comparison_model = meta["best_models"]["mml"]
# Perform rotation.
A_est = comparison_model.theta_[comparison_model.parameter_names.index("A")]
if A_est.shape == A_target.shape:
for rotation in range(max_rotations):
A_est = comparison_model.theta_[comparison_model.parameter_names.index("A")]
R, p_opt, cov, *_ = utils.find_rotation_matrix(A_target, A_est,
full_output=True)
R_opt = utils.exact_rotation_matrix(A_target, A_est,
p0=np.random.uniform(-np.pi, np.pi, comparison_model.n_latent_factors**2))
if config["subtract_mean"]:
X = X - np.nanmean(X, axis=0)
gJs, gKs, converged, meta = grid_search.grid_search(Js,
Ks,
X,
N_inits=N_inits,
mcfa_kwds=mcfa_kwds)
# Plot the grid space.
ll = meta["ll"]
bic = meta["bic"]
mml = meta["message_length"]
J_best_ll, K_best_ll = grid_search.best(Js, Ks, -ll)
J_best_bic, K_best_bic = grid_search.best(Js, Ks, bic)
J_best_mml, K_best_mml = grid_search.best(Js, Ks, mml)
print(
f"Best log likelihood for N = {size} at J = {J_best_ll} and K = {K_best_ll}"
)
print(
f"Best BIC value found for N = {size} at J = {J_best_bic} and K = {K_best_bic}"
)
print(
f"Best MML value found for N = {size} at J = {J_best_mml} and K = {K_best_mml}"
)
try:
# Plot some contours.
max_n_latent_factors = gs_options["max_n_latent_factors"]
max_n_components = gs_options["max_n_components"]
N_inits = gs_options["n_inits"]
Js = 1 + np.arange(max_n_latent_factors)
Ks = 1 + np.arange(max_n_components)
Jg, Kg, converged, meta = grid_search.grid_search(Js, Ks, X, N_inits=N_inits,
mcfa_kwds=mcfa_kwds)
ll = meta["ll"]
bic = meta["bic"]
pseudo_bic = meta["pseudo_bic"]
message_length = meta["message_length"]
J_best_ll, K_best_ll = grid_search.best(Js, Ks, -ll)
J_best_bic, K_best_bic = grid_search.best(Js, Ks, bic)
J_best_mml, K_best_mml = grid_search.best(Js, Ks, message_length)
print(f"Best log likelihood at J = {J_best_ll} and K = {K_best_ll}")
print(f"Best BIC value found at J = {J_best_bic} and K = {K_best_bic}")
print(f"Best MML value found at J = {J_best_mml} and K = {K_best_mml}")
# Plot some contours.
plot_filled_contours_kwds = dict(converged=converged,
marker_function=np.nanargmin,
cmap="Spectral_r")
fig_ll = mpl_utils.plot_filled_contours(Jg, Kg, -ll,
colorbar_label=r"$-\log\mathcal{L}$",
**plot_filled_contours_kwds)
savefig(fig_ll, "gridsearch-ll")
