def test_l1_fit_runs_correctly(self):
# how low can we shrink this tolerance and still pass?
tol = 0.03
index = np.arange(self.num)
result = l1_fit(index, self.y,
seasonality_matrix=self.seasonality_matrix)
model = result['model']
self.assertEquals(len(model), self.num)
for x_value, y_value, model_value in zip(index, self.y, model):
diff = model_value - y_value
print x_value, y_value, model_value, diff
self.assertLess(abs(diff), tol)
assert True
def fit(self):
self.solution = l1_fit(self.x,
self.y,
beta_d1=self.params['beta_d1'],
beta_d2=self.params['beta_d2'],
beta_seasonal=self.params['beta_seasonal'],
beta_step=self.params['beta_step'],
growth=self.params['growth'],
seasonality_matrix=self.seasonality_matrix)
y2 = self.solution['base'][-1]
y1 = self.solution['base'][-2]
x2 = self.x[-1]
x1 = self.x[-2]
self.slope = (y2 - y1) / float(x2 - x1)
self.offset = y2 - self.slope * x2
# fill in the seasonal parameters
# remember that the last one is 1-sum_of_rest
# as they sum to zero
seasonal_params = self.solution['seasonal_parameters']
sum_seasonal = seasonal_params.sum()
self.seasonal = np.array(list(seasonal_params) + [-sum_seasonal])
# define this function to gracefully handle
# unconstrained seasonal params
def date_to_seasonal_component_function(the_date):
the_index = self.seasonality_function(the_date)
if the_index not in self.compression_dict:
return 0.0
compressed_index = self.compression_dict[the_index]
return self.seasonal[compressed_index]
self.date_to_seasonal_component = date_to_seasonal_component_function
# set this function for interpolating
self.interpolate = interp1d(self.x, self.solution['model'])
self.extrapolate_without_seasonal = lambda new_x: self.offset + self.slope * new_x
def test_l1_fit_runs(self):
index = np.arange(self.num)
result = l1_fit(index, self.y,
seasonality_matrix=self.seasonality_matrix)
self.assertEquals(len(result['model']), self.num)
