def should_plot(name):
return plot_setting(section="fit_quantity", name=name)
def should_plot(name):
return plot_setting(section="ray_tracing", name=name)
def should_plot(name):
return plot_setting(section="hyper", name=name)
def visualize_stochastic_histogram(
self,
stochastic_log_likelihoods: np.ndarray,
max_log_evidence: float,
histogram_bins: int = 10,
):
"""
Certain `Inversion`'s have stochasticity in their log likelihood estimate.
For example, the `VoronoiBrightnessImage` pixelization, which changes the likelihood depending on how different
KMeans seeds change the pixel-grid.
A log likelihood cap can be applied to model-fits performed using these `Inversion`'s to improve error and
posterior estimates. This log likelihood cap is estimated from a list of stochastic log likelihoods, where
these log likelihoods are computed using the same model but with different KMeans seeds.
This function plots a histogram representing the distribution of these stochastic log likelihoods with a 1D
Gaussian fit to the likelihoods overlaid. This figure can be used to determine how subject the fit to this
dataset is to the stochastic likelihood effect.
Parameters
----------
stochastic_log_likelihoods
The stochastic log likelihood which are used to plot the histogram and Gaussian.
max_log_evidence
The maximum log likelihood value of the non-linear search, which will likely be much larger than any of the
stochastic log likelihoods as it will be boosted high relative to most samples.
histogram_bins
The number of bins in the histogram used to visualize the distribution of stochastic log likelihoods.
Returns
-------
float
A log likelihood cap which is applied in a stochastic model-fit to give improved error and posterior
estimates.
"""
if stochastic_log_likelihoods is None:
return
if plot_setting("other", "stochastic_histogram"):
file_path = path.join(self.visualize_path, "other")
try:
os.makedirs(file_path)
except FileExistsError or IsADirectoryError:
pass
filename = path.join(file_path, "stochastic_histogram.png")
if path.exists(filename):
try:
os.rmdir(filename)
except Exception:
pass
(mu, sigma) = norm.fit(stochastic_log_likelihoods)
n, bins, patches = plt.hist(
x=stochastic_log_likelihoods, bins=histogram_bins, density=1
)
y = norm.pdf(bins, mu, sigma)
plt.plot(bins, y, "--")
plt.xlabel("log evidence")
plt.title("Stochastic Log Evidence Histogram")
plt.axvline(max_log_evidence, color="r")
plt.savefig(filename, bbox_inches="tight")
plt.close()
def should_plot(name):
return plot_setting(section=["positions"], name=name)