def test_lasso_signal_primal(setup_param_lasso, expected_beta):
expected_beta_lasso_signal = expected_beta['values_signal']['lasso']
calculated_beta_primal = fused_lasso_primal(**setup_param_lasso['data_'
'primal'])
assert_allclose(expected_beta_lasso_signal,
calculated_beta_primal,
atol=2e-2)
def test_lasso_general_primal(setup_param_lasso, expected_beta):
setup_param_lasso['data_primal']['X'] = np.array([[1, 0.5], [0.5, 1]])
setup_param_lasso['data_primal']['s1'] = 2.09999
expected_beta_lasso_general = expected_beta['values_general']['lasso']
calculated_beta_lagrange = fused_lasso_primal(
**setup_param_lasso['data_primal'])
assert_allclose(expected_beta_lasso_general,
calculated_beta_lagrange,
atol=2e-2)
def fit(self, X, y):
"""Fit unkown parameters *beta_hat* to *X* and *y*.
Example:
--------
>>> reg = FusedLassoEstimator( s1=1, s2=10).fit(X, y)
Args:
| X (np.ndarray): 2d array of independent variables.
| y (np.ndarray): 1d array of dependent variables.
Returns:
| b.value (np.ndarray)
"""
self.beta = fused_lasso_primal(y, X, self.s1, self.s2)
return self.beta
def test_fused_lasso_primal_wrong_dimension_penalty(setup_param_lasso):
setup_param_lasso['data_primal']['s1'] = [1, 2]
with pytest.raises(TypeError):
fused_lasso_primal(**setup_param_lasso['data_primal'])
def test_fused_lasso_primal_negative_penalty(setup_param_lasso):
setup_param_lasso['data_primal']['s1'] = -1
with pytest.raises(ValueError):
fused_lasso_primal(**setup_param_lasso['data_primal'])
"""Calculate fused lasso estimates for CGH data and plot them."""
import matplotlib.pyplot as plt
import numpy as np
from bld.project_paths import project_paths_join as ppj
from src.model_code.fused_lasso_primal import fused_lasso_primal
if __name__ == "__main__":
# Set penalty constants, load CGH data and estimate fused lasso.
CGH_DATA = np.loadtxt(ppj("IN_DATA", "cgh.txt"))
BETA_HAT = fused_lasso_primal(CGH_DATA, np.identity(len(CGH_DATA)), 160,
15)
# Plot CGH data.
plt.xlabel('Genome Order')
plt.ylabel('Copy Number')
plt.plot(CGH_DATA, "bo")
plt.axhline(color='r')
plt.savefig(ppj("OUT_FIGURES", "cgh_plot_raw.png"))
# Plot CGH data with fused lasso estimates.
plt.xlabel('Genome Order')
plt.ylabel('Copy Number')
plt.plot(CGH_DATA, "bo")
plt.axhline(color='r')
plt.plot(BETA_HAT, color='orange')
plt.savefig(ppj("OUT_FIGURES", "cgh_plot_beta.png"))
plt.clf()
# Load data from json file.
N_FEATURES = SIM_DICT["p"]
N_OBS = SIM_DICT["n"]
N_SIMULATIONS = SIM_DICT['num_simulations']
# Building containers to store simulation results.
BETA_HAT = np.empty((N_FEATURES, N_SIMULATIONS))
Y_HAT = np.empty((N_OBS, N_SIMULATIONS))
RESIDUALS = np.empty((N_OBS, N_SIMULATIONS))
# Calculate beta estimates for cross-validated penalty terms.
for i in range(N_SIMULATIONS):
start_time_estimation = time()
BETA_HAT[:, i] = fused_lasso_primal(y[:, i], X, S1_OPT, S2_OPT)
#fle(penalty_cv[0], penalty_cv[1]).fit(X, y[:, i])
Y_HAT[:, i] = np.matmul(X, BETA_HAT[:, i])
RESIDUALS[:, i] = y[:, i] - Y_HAT[:, i]
end_time_estimation = time()
# Save timings for fused lasso setting.
if REG_NAME == 'fused':
with open(ppj("OUT_ANALYSIS", "time_list_{}.pickle".
format(SIM_NAME)), "rb") as in_file:
TIMINGS = pickle.load(in_file)
TIME_ESTIMATION = np.round((end_time_estimation - start_time_estimation), 2)
TIMINGS.append(TIME_ESTIMATION)