def plot_test_vs_validation_set(output_name, C, sigma, M, xlim=None, ylim=None, xticks=None, yticks=None):
"""
Plot a single figure which compares the expected validation and generalization errors
for various numbers of training samples (n), using either m=1 or m=n validation samples.
"""
import matplotlib.pyplot as plt
plt.ioff()
plt.style.use('./latex-paper.mplstyle')
plt.figure()
ax = plt.axes()
ax.yaxis.grid(True)
d = pickler.load(f'categorical_K2_C{C}_sigma{sigma:.2f}_M{M}'.replace('.','_'))
x_values = [x.n_train for x in d.xs]
plt.plot(x_values, d.results[:,0], 'C0-', label=r'$\mathrm{e}_{\mathrm{val}}(m=n)$')
plt.plot(x_values, d.results[:,1], 'C1-', label=r'$\mathrm{e}_{\mathrm{gen}}(m=n)$')
d = pickler.load(f'categorical_LOO_C{C}_sigma{sigma:.2f}_M{M}'.replace('.','_'))
x_values = [x.n_train for x in d.xs]
plt.plot(x_values, d.results[:,0], 'C0--', label=r'$\mathrm{e}_{\mathrm{val}}(m=1)$')
plt.plot(x_values, d.results[:,1], 'C1--', label=r'$\mathrm{e}_{\mathrm{gen}}(m=1)$')
plt.xlabel('n')
plt.ylabel('MSE')
if xlim is not None:
plt.xlim(xlim)
if ylim is not None:
plt.ylim(ylim)
if xticks is not None:
plt.xticks(xticks)
if yticks is not None:
plt.yticks(yticks)
plt.legend(loc='best')
simulations_framework.save_figure(output_name + f'_reps{d.n_repetitions}')
def plot_res(p, df, sigma, with_intercept):
(xs, null_model_mse, correct_mses, incorrect_mses,
n_reps) = read_results_with_confintervals(p, df, sigma, with_intercept)
plt.style.use('./latex-paper.mplstyle')
fig, ax = plt.subplots()
X = [x.n_train + x.n_validation for x in xs]
ax.fill_between(X, correct_mses[:, 0], correct_mses[:, 1], alpha=0.2)
ax.fill_between(X, incorrect_mses[:, 0], incorrect_mses[:, 1], alpha=0.2)
ax.plot(X, np.mean(correct_mses, axis=1), '-', label='Correct')
ax.plot(X, np.mean(incorrect_mses, axis=1), '-', label='Incorrect')
ax.plot([X[0], X[-1]], [null_model_mse] * 2,
'k:',
linewidth=1.0,
label='Null model')
ax.legend()
ax.set_xticks(X)
ax.set_xlim([X[0], X[-1]])
ax.set_xlabel('N')
ax.set_ylabel('MSE')
#latexsigma = r'\sigma'
#ax.set_title(f'$p={p} \ df={df} \ {latexsigma}={sigma} \ intr={int(with_intercept)} \ reps={n_reps}$')
ax.grid(axis='y')
save_figure(
f'normalized_lasso_cv_pipeline_p{p}_df{df}_sigma{sigma}_10FOLDCV_reps{n_reps}',
'normalized_lasso_pipeline')
def plot_test_vs_validation_set(subdir,
filename_prefix,
M,
normalize,
xlim=None,
ylim=None,
xticks=None,
yticks=None):
"""
Plot a single figure which compares the expected validation and generalization errors
for various numbers of training samples (n), using either m=1 or m=n validation samples.
"""
import matplotlib.pyplot as plt
plt.ioff()
plt.style.use('./latex-paper.mplstyle')
plt.figure()
ax = plt.axes()
ax.yaxis.grid(True)
d = pickler.load(f'{filename_prefix}_K2_M{M}_normalize{normalize}')
x_values = [x.n_train for x in d.xs]
plt.plot(x_values,
d.results[:, 0],
'C0-',
linewidth=1.5,
label=r'$\mathrm{e}_{\mathrm{val}}(m=n)$')
plt.plot(x_values,
d.results[:, 1],
'C1-',
linewidth=1.5,
label=r'$\mathrm{e}_{\mathrm{gen}}(m=n)$')
d = pickler.load(f'{filename_prefix}_LOO_M{M}_normalize{normalize}')
x_values = [x.n_train for x in d.xs]
plt.plot(x_values,
d.results[:, 0],
'C0--',
linewidth=1.5,
label=r'$\mathrm{e}_{\mathrm{val}}(m=1)$')
plt.plot(x_values,
d.results[:, 1],
'C1--',
linewidth=1.5,
label=r'$\mathrm{e}_{\mathrm{gen}}(m=1)$')
plt.xlabel('$n$')
plt.ylabel('MSE')
plt.xlim(xlim if xlim is not None else [min(x_values), max(x_values)])
if ylim is not None:
plt.ylim(ylim)
if xticks is not None:
plt.xticks(xticks)
if yticks is not None:
plt.yticks(yticks)
#ax.set_yscale('log')
plt.legend(loc='best')
output_name = f'{filename_prefix}_M{M}_normalize{normalize}_reps{d.n_repetitions}'
save_figure(output_name, subdir=subdir)
def plot_test_vs_validation_set(filename_prefix, D, df, K_strong_columns, strong_column_multiplier, K, noise_multiplier, xlim=None, ylim=None, xticks=None, yticks=None):
"""
Plot a single figure which compares the expected validation and generalization errors
for various numbers of training samples (n), using either m=1 or m=n validation samples.
"""
import matplotlib.pyplot as plt
plt.ioff()
plt.style.use('./latex-paper.mplstyle')
plt.figure()
ax = plt.axes()
ax.yaxis.grid(True)
d = pickler.load(f'variable_selected_linear_regression_K2_D{D}_df{df}_Kstrong{K_strong_columns}_multiplier{strong_column_multiplier}_K{K}_noisemul{noise_multiplier:.2f}')
x_values = [x.n_train for x in d.xs]
plt.plot(x_values, d.results[:,0], 'C0-', linewidth=1.5, label=r'$\mathrm{e}_{\mathrm{val}}(m=n)$')
plt.plot(x_values, d.results[:,1], 'C1-', linewidth=1.5, label=r'$\mathrm{e}_{\mathrm{gen}}(m=n)$')
d = pickler.load(f'variable_selected_linear_regression_LOO_D{D}_df{df}_Kstrong{K_strong_columns}_multiplier{strong_column_multiplier}_K{K}_noisemul{noise_multiplier:.2f}')
x_values = [x.n_train for x in d.xs]
plt.plot(x_values, d.results[:,0], 'C0--', linewidth=1.5, label=r'$\mathrm{e}_{\mathrm{val}}(m=1)$')
plt.plot(x_values, d.results[:,1], 'C1--', linewidth=1.5, label=r'$\mathrm{e}_{\mathrm{gen}}(m=1)$')
d = pickler.load(f'variable_selected_linear_regression_null_model_D{D}_df{df}_Kstrong{K_strong_columns}_multiplier{strong_column_multiplier}_noisemul{noise_multiplier:.2f}')
null_mse = d.results[0][0]
plt.plot([min(x_values), max(x_values)], [null_mse, null_mse], 'k:', linewidth=1.0, label='Null model')
plt.xlabel('$n$')
plt.ylabel('MSE')
plt.xlim(xlim if xlim is not None else [min(x_values), max(x_values)])
if ylim is not None:
plt.ylim(ylim)
if xticks is not None:
plt.xticks(xticks)
if yticks is not None:
plt.yticks(yticks)
plt.legend(loc='best')
output_name = f'{filename_prefix}_D{D}_df{df}_Kstrong{K_strong_columns}_multiplier{strong_column_multiplier}_K{K}_noisemul{noise_multiplier:.2f}_reps{d.n_repetitions}'
save_figure(output_name, f'noise_{noise_multiplier:.2f}')