def test_misfit_calculation_lowess():
profile = create_profile_for_calculating_misfit()
profile_manual = create_manual_profile_for_calculating_misfit()
misfit_df = misfit_calculation(profile,
profile_manual,
method="lowess",
bin_max=11,
bin_min=5,
bin_step=2,
theta_T_max=40,
theta_T_min=30,
theta_T_step=2,
phi_T=5,
plot=False,
save_file=False)
misfit_df['misfit_height'] = misfit_df['misfit_height'].fillna(0).astype(
int)
misfit_df['misfit_width'] = misfit_df['misfit_width'].fillna(0).astype(int)
misfit_df['misfit_slope'] = misfit_df['misfit_slope'].fillna(0).astype(int)
misfit_df['misfit_total'] = misfit_df['misfit_total'].fillna(0).astype(int)
assert misfit_df.loc[0].to_list() == [1, 'lowess', 1, 5, 4, 30, 5, 10]
def find_best(df, df_manual, methods, bin_max, bin_min,
bin_step, theta_t_max,
theta_t_min, theta_t_step, phi_t, criteria_weights,
plot):
"""
Finds the best set of parameters to use in the algorithm
:param: df <DataFrame> The DataFrame with profile data in
:param: df_manual <DataFrame> The DataFrame with the manual results in to compare to
:param: methods <list> A list of methods to be used
:param: bin_max <int> The maximum smoothing bin value to use
:param: bin_min <int> The minimum smoothing bin value to use
:param: bin_step <int> The smoothing bin step value to use
:param: theta_t_max <int> The maximum theta value to use
:param: theta_t_min <int> The minimum theta value to use
:param: theta_t_step <int> The theta step value to use
:param: phi_t <int> The phi value to use
:param: plot <boolean> Whether to make plots or not
:param: criteria_weights <list> The list of weights to apply to the results
"""
for method in methods:
print(f"Performing misfit analysis for method: {method}")
for n, profile in tqdm(enumerate(df_manual.profile)):
profile_number = df_manual['profile'][n]
profile = df[df['profile'] == profile_number].reset_index()
profile_manual = df_manual[df_manual['profile'] == profile_number]
df_tmp = misfit_calculation(profile, profile_manual,
method, bin_max, bin_min,
bin_step, theta_t_max,
theta_t_min, theta_t_step, phi_t,
plot = False, save_file = False)
try:
df_out = df_out.append(df_tmp)
except NameError:
df_out = df_tmp
df_out_average = df_out.groupby(['method', 'theta_t', 'bin']).mean().reset_index().drop('profile', axis = 1)
if plot:
for method in methods:
df_out_average_method = df_out_average[df_out_average['method'] == method]
for value in ['height', 'width', 'slope']:
df_out_average_plot = df_out_average_method[[f'misfit_{value}', 'bin', 'theta_t']].groupby(
['theta_t', 'bin']).mean().unstack(
level = 0)
df_out_average_plot.columns = df_out_average_plot.columns.droplevel()
ax = sns.heatmap(df_out_average_plot, linewidths = 0.1, annot = False, cbar = True,
cbar_kws = {'label': f'misfit_{value}'}, vmin = 0, cmap = "viridis")
ax.set_title(f'{method}')
plt.show()
df_out_average['misfit_total'] = \
df_out_average['misfit_height'] * criteria_weights[0] + \
df_out_average['misfit_width'] * criteria_weights[1] + \
df_out_average['misfit_slope'] * criteria_weights[2]
for criteria in ['total', 'height', 'width', 'slope']:
best_value = min(df_out_average[f'misfit_{criteria}'])
the_best = df_out_average[df_out_average[f'misfit_{criteria}'] == best_value]
print(f'\nThe best parameters using criteria {criteria} are:')
print(the_best.to_string())
try:
df_out_best = df_out_best.append(the_best)
except NameError:
df_out_best = the_best
save_to_csv(df_out_average, name = 'all')
save_to_csv(df_out_best, name = 'best')