def test_varmax_without_intercept2(self):
ts_data = self.get_data_for_varmax()
f_name='varmax_without_intercept2.pmml'
model = VARMAX(ts_data, order=(1,1), trend=None)
result = model.fit()
StatsmodelsToPmml(result, f_name)
self.assertEqual(os.path.isfile(f_name),True)
def varma(self, df, host):
pd.plotting.register_matplotlib_converters()
df_diffed, no_diffs = Helper.diff_test(df)
print(df_diffed)
df_diffed.replace([np.inf, -np.inf], np.nan)
cols = df_diffed.columns
df_diffed = df_diffed.dropna()
nobs = int(len(df_diffed) / 10) + 2
train = df_diffed[:-nobs]
test = df_diffed[-nobs:]
model = VARMAX(train, order=(2,2), trend='c')
results = model.fit(maxiter=1000, disp=False)
print(results.summary())
df_forecast = results.forecast(nobs)
print(df_forecast)
df_fixed = Helper.reverse_diff(df_forecast, df, nobs, no_diffs)
print(df_fixed)
for col in df.columns:
print("-- RMSE --")
print(rmse(test[col], df_fixed[col + '_forecast']))
print("-- Mean --")
print(test[col].mean())
df[col].plot(legend=True)
df_fixed[col + '_forecast'].plot(legend=True)
plt.show()
def test_varmax_with_intercept(self):
ts_data = self.get_data_for_varmax()
f_name='varmax_with_intercept.pmml'
model = VARMAX(ts_data, order=(1,1))
result = model.fit()
StatsmodelsToPmml(result, f_name, conf_int=[80,95])
self.assertEqual(os.path.isfile(f_name),True)
def test_varmax_without_intercept(self):
ts_data = self.statsmodels_data_helper.get_data_for_varmax()
f_name = 'varmax_without_intercept.pmml'
model = VARMAX(ts_data, order=(1, 1), trend=None)
result = model.fit()
ArimaToPMML(result, f_name)
self.assertEqual(self.schema.is_valid(f_name), True)
def VARMAXfit(self, endog, exog, order=(2, 0), maxiter=1000):
model = VARMAX(endog.dropna(),
order=order,
trend='n',
exog=exog.dropna())
model_fit = model.fit(maxiter=maxiter, disp=False)
return model_fit
def VARMAXforecastPlot(self,
endog,
train_fraction=0.66,
steps=1,
order=(2, 0),
maxiter=1000,
verbose=False,
error_cov_type='diagonal',
title='',
titleFontSize=12,
labelFontSize=10,
figureSize=(20, 6),
lineWidth=1.0):
sns.set_theme()
cleaned = endog.dropna()
size = int(len(cleaned) * train_fraction)
timestamp = cleaned.index.to_timestamp().values.tolist()
train = cleaned[0:size].to_numpy().tolist()
test = cleaned[size:len(cleaned)].to_numpy().tolist()
history = [z for z in train]
predictions = train
interval = 10
for t in range(len(test)):
model = VARMAX(history, order=order, error_cov_type=error_cov_type)
model_fit = model.fit(maxiter=maxiter, disp=False)
output = model_fit.forecast(steps)
yhat = output[0].tolist()
predictions.append(yhat)
obs = test[t]
history.append(obs)
if (t % 10 == 0 and verbose):
print('percent done: ' + str(t / len(test)))
print('percent done: 1.00')
if (len(endog.dropna().to_numpy().tolist()) != len(predictions)):
print('List lengths do not match')
col = 0
for val in endog:
plt.subplots(figsize=figureSize)
plt.title(title, fontsize=titleFontSize)
plt.xlabel('Timestamp', fontsize=labelFontSize)
plt.ylabel(endog[val].name, fontsize=labelFontSize)
column = [row[col] for row in predictions]
plt.plot(timestamp,
column,
color='red',
linewidth=lineWidth,
linestyle='--')
col += 1
plt.plot(timestamp,
endog[val].dropna().to_numpy().tolist(),
linewidth=lineWidth)
plt.legend(labels=('prediction', endog[val].name))
plt.show()
return predictions
def fit_varmax(data, p, q, exog=None, summary=True):
model = VARMAX(data,
order=(p, q),
exog=exog,
initialization='approximate_diffuse')
model_fit = model.fit(maxiter=50, disp=False)
if summary:
print(model_fit.summary())
return model_fit
def VARMAXvma(self,
endog,
order=(2, 0),
error_cov_type='diagonal',
maxiter=1000,
disp=False):
model = VARMAX(endog.dropna(),
order=order,
error_cov_type=error_cov_type)
model_fit = model.fit(maxiter=maxiter, disp=disp)
return model_fit
# if predictions.shape[0] != Y.shape[0]:
# outcome = Y.tail(predictions.shape[0])
# else:
# outcome = Y
# accuracy_matrix = outcome*predictions
# accuracy_matrix.values
# accuracy.append(np.sum(np.sum((accuracy_matrix > 0)))/accuracy_matrix.size)
endog_Y = X.drop(columns=[
'XMRspread', 'XMRvolume', 'XMRbasevolume', 'XRPspread', 'XRPvolume',
'XRPbasevolume', 'LTCspread', 'LTCvolume', 'LTCbasevolume', 'DASHspread',
'DASHvolume', 'DASHbasevolume', 'ETHspread', 'ETHvolume', 'ETHbasevolume'
])
exog_X = X.drop(
columns=['ETHreturn', 'XRPreturn', 'LTCreturn', 'DASHreturn', 'XMRreturn'])
p = 1
model = VARMAX(endog=endog_Y, order=(p, 0), exog=exog_X)
results = model.fit(maxiter=0)
predictions = results.predict()
predictions = (predictions.shift(-1)).dropna()
if predictions.shape[0] != Y.shape[0]:
outcome = ((endog_Y.tail(predictions.shape[0])).shift(-1)).dropna()
else:
outcome = (endog_Y.shift(-1)).dropna()
accuracy_matrix = np.sign(outcome) * np.sign(predictions)
accuracy_matrix.values
accuracy = (np.sum(np.sum((accuracy_matrix > 0.0))) / accuracy_matrix.size)
# turn into numpy array
endog_Y_test = X_test.drop(columns=[
# Plot Diagnostics
model.plot_diagnostics(variable=4, figsize=(14,8), lags=24)
plt.gcf().suptitle('Beer - Diagnostics', fontsize=14)
plt.tight_layout()
plt.subplots_adjust(top=.93);
# Impulse Response Analysis
irf = var_results.irf(13)
irf.plot(orth=False)
# Forecast Error Variance Decomposition
var_results.fevd(13).plot()
# VARMAX
var_model = VARMAX(df_tran, order=(1,0), trend='c').fit(maxiter=1000)
var_model.summary()
model.plot_diagnostics(variable=3, figsize=(14,8), lags=24)
plt.gcf().suptitle('Beer - Diagnostics', fontsize=14)
plt.tight_layout()
plt.subplots_adjust(top=.93);
# Impulse-Response Function
changes = df_tran.diff().quantile(.05).tolist() # quantile=0.01 means negative change
# print (changes)
model.impulse_responses(steps=6,
impulse=changes).plot.bar(subplots=True,
figsize=(12, 6),
rot=0,
legend=False)