def plot_model(df, model_type, n_components, title='', plot_CIs=True, repetitions=20, save_file_to='model.pdf',
maxiter=5000, maxfun=5000, method='nm', period=24):
rows, cols = hlp.get_factors(1)
fig = plt.figure(figsize=(8 * cols, 8 * rows))
gs = gridspec.GridSpec(rows, cols)
results, df_result, _ = dproc.fit_to_model(df, n_components, model_type, period, maxiter, maxfun, method, 0)
# plot
ax = fig.add_subplot(gs[0])
if plot_CIs:
CIs = subplot_confidence_intervals(df, n_components, model_type, ax, repetitions=repetitions, maxiter=maxiter,
maxfun=maxfun, period=period, method=method)
subplot_model(df['X'], df['Y'], df_result['X_test'], df_result['Y_test'], ax, color='blue', title=title,
fit_label='fitted curve')
ax_list = fig.axes
for ax in ax_list:
ax.legend(loc='upper left', fontsize='large')
fig.tight_layout()
plt.show()
# save
try:
hlp.make_results_dir()
fig.savefig(r'results\/' + save_file_to)
except:
print("Can not save plot.")
if plot_CIs:
return CIs
def plot_confidence_intervals(df, model_type, n_components, title='', repetitions=20, maxiter=5000, maxfun=5000,
period=24, method='nm', save_file_to='CIs.pdf'):
rows, cols = hlp.get_factors(1)
fig = plt.figure(figsize=(8 * cols, 8 * rows))
gs = gridspec.GridSpec(rows, cols)
ax = fig.add_subplot(gs[0])
Y = df['Y']
results, df_result, X_fit_test = dproc.fit_to_model(df, n_components, model_type, period, maxiter, maxfun, method,
0)
# CI
res2 = copy.deepcopy(results)
params = res2.params
CIs = dproc.calculate_confidence_intervals(df, n_components, model_type, repetitions, maxiter, maxfun, method,
period)
N2 = round(10 * (0.7 ** n_components) + 4)
P = np.zeros((len(params), N2))
i = 0
for index, CI in CIs.iterrows():
P[i, :] = np.linspace(CI[0], CI[1], N2)
i = i + 1
param_samples = hlp.lazy_cartesian_product(P)
size = param_samples.max_size
N = round(df.shape[0] - df.shape[0] / 3)
for i in range(0, N):
j = random.randint(0, size)
p = param_samples.entry_at(j)
res2.initialize(results.model, p)
if model_type == 'zero_nb' or model_type == "zero_poisson":
Y_test_CI = res2.predict(X_fit_test, exog_infl=X_fit_test)
else:
Y_test_CI = res2.predict(X_fit_test)
if i == 0:
ax.plot(df_result['X_test'], Y_test_CI, color='tomato', alpha=0.05, linewidth=0.1,
label='confidence intervals')
else:
ax.plot(df_result['X_test'], Y_test_CI, color='tomato', alpha=0.05, linewidth=0.1)
subplot_model(df['X'], Y, df_result['X_test'], df_result['Y_test'], ax, title=title, plot_model=False)
ax_list = fig.axes
for ax in ax_list:
ax.legend(loc='upper left', fontsize='large')
fig.tight_layout()
plt.show()
# save
try:
hlp.make_results_dir()
fig.savefig(r'results\/' + save_file_to)
except:
print("Can not save plot.")
return CIs
def plot_models(dfs,
model_type,
n_components,
title=[''],
rows=1,
cols=1,
period=24,
maxiter=5000,
maxfun=5000,
method='nm',
disp=0,
plot_CIs=True,
repetitions=50,
save_file_to='win.pdf'):
fig = plt.figure(figsize=(8 * cols, 8 * rows))
gs = gridspec.GridSpec(rows, cols)
i = 0
for df in dfs:
_, df_result, _ = dproc.fit_to_model(df, n_components[i],
model_type[i], period, maxiter,
maxfun, method, disp)
# plot
ax = fig.add_subplot(gs[i])
if plot_CIs:
subplot_confidence_intervals(df,
n_components[i],
model_type[i],
ax,
repetitions=repetitions,
period=period,
maxiter=maxiter,
maxfun=maxfun,
method=method)
subplot_model(df['X'],
df['Y'],
df_result['X_test'],
df_result['Y_test'],
ax,
color='blue',
title=title[i],
fit_label='prilagojena krivulja',
raw_label='izvorni podatki')
i = i + 1
ax_list = fig.axes
for ax in ax_list:
ax.legend(loc='upper left')
fig.tight_layout()
plt.show()
# save
try:
hlp.make_results_dir()
fig.savefig(r'results\/' + save_file_to)
except:
print("Can not save plot.")
def subplot_confidence_intervals(df,
n_components,
model_type,
ax,
repetitions=20,
maxiter=5000,
maxfun=5000,
period=24,
method='nm'):
results, df_result, X_fit_test = dproc.fit_to_model(
df, n_components, model_type, period, maxiter, maxfun, method, 0)
# CI
res2 = copy.deepcopy(results)
params = res2.params
CIs = dproc.calculate_confidence_intervals(df, n_components, model_type,
repetitions, maxiter, maxfun,
method, period)
N2 = round(10 * (0.7**n_components) + 4)
P = np.zeros((len(params), N2))
i = 0
for index, CI in CIs.iterrows():
P[i, :] = np.linspace(CI[0], CI[1], N2)
i = i + 1
param_samples = hlp.lazy_cartesian_product(P)
size = param_samples.max_size
N = round(df.shape[0] - df.shape[0] / 3)
for i in range(0, N):
j = random.randint(0, size)
p = param_samples.entry_at(j)
res2.initialize(results.model, p)
if model_type == 'zero_nb' or model_type == "zero_poisson":
Y_test_CI = res2.predict(X_fit_test, exog_infl=X_fit_test)
else:
Y_test_CI = res2.predict(X_fit_test)
if i == 0:
ax.plot(df_result['X_test'],
Y_test_CI,
color='tomato',
alpha=0.3,
linewidth=0.15)
else:
ax.plot(df_result['X_test'],
Y_test_CI,
color='tomato',
alpha=0.3,
linewidth=0.15)
return CIs