def test_Model():
""" """
M = sb.Model()
x = np.linspace(0.1, 0.9, 22)
target_mu = 0.5
target_sigma = 1
target_y = sb.cumgauss(x, target_mu, target_sigma)
F = M.fit(x, target_y, initial=[target_mu, target_sigma])
npt.assert_equal(F.predict(x), target_y)
def test_params_regression():
"""
Test for regressions in model parameter values from provided data
"""
model = sb.Model()
ortho_x, ortho_y, ortho_n = sb.transform_data(
op.join(data_path, 'ortho.csv'))
para_x, para_y, para_n = sb.transform_data(op.join(data_path, 'para.csv'))
ortho_fit = model.fit(ortho_x, ortho_y)
para_fit = model.fit(para_x, para_y)
npt.assert_almost_equal(ortho_fit.params[0], 0.46438638)
npt.assert_almost_equal(ortho_fit.params[1], 0.13845926)
npt.assert_almost_equal(para_fit.params[0], 0.57456788)
npt.assert_almost_equal(para_fit.params[1], 0.13684096)
def error_func(x, w):
return x * w
coefs = np.zeros([len(noise_levels), n_boots])
for ii, n_level in enumerate(noise_levels):
# Generate y for this noise level
y = w * x + n_level * np.random.randn(N)
for jj in range(n_boots):
# Pull subsets of data
ixs_boot = np.random.randint(0, N, N)
x_boot = x[ixs_boot]
y_boot = y[ixs_boot]
# Fit the model and return the coefs
model = sb.Model(error_func)
fit = model.fit(x_boot, y_boot, (.5,))
coefs[ii, jj] = fit.params[0]
###############################################################################
# Assessing coefficient stability
# -------------------------------
#
# Now we'll assess the stability of the fitted coefficient for varying levels
# of noise. Let's plot the raw values for each noise level, as well as the
# 95% confidence interval.
percentiles = np.percentile(coefs, [2.5, 97.5], axis=1).T
fig, ax = plt.subplots()
for n_level, i_coefs, percs in zip(noise_levels, coefs, percentiles):
ax.scatter(np.repeat(n_level, len(i_coefs)), i_coefs)
# First, we'll load some data into shablona. data_path = op.join(sb.__path__[0], 'data') ortho_x, ortho_y, ortho_n = sb.transform_data(op.join(data_path, 'ortho.csv')) para_x, para_y, para_n = sb.transform_data(op.join(data_path, 'para.csv')) ############################################################################### # Fitting a model # --------------- # # With shablona, models are created with the :ref:Model class. # This class has a `fit` method that returns the coefficients for the given # input data. # Instantiate our model and fit it on two datasets model = sb.Model() ortho_fit = model.fit(ortho_x, ortho_y) para_fit = model.fit(para_x, para_y) # These are the parameters that our model has discovered print(ortho_fit.params) print(para_fit.params) ############################################################################### # Visualizing results # ------------------- # # Now we will visualize the results of our model fit. We'll generate # a vector of input points, and use them to determine the model's output # for each input. Then we'll plot what these curves look like.