def _get_param_estimates_1d(x, z: np.array) -> lm.Parameters:
"""Returns lm.Parameters for x, z data"""
assert z.ndim == 1
z, x = CU.resample_data(z, x, max_num_pnts=500)
params = lm.Parameters()
s = pd.Series(
z) # Put into Pandas series so I can work with NaN's more easily
sx = pd.Series(x, index=s.index)
z = s[s.first_valid_index():s.last_valid_index() + 1] # type: pd.Series
x = sx[s.first_valid_index():s.last_valid_index() + 1]
if np.count_nonzero(~np.isnan(
z)) > 10: # Prevent trying to work on rows with not enough data
try:
smooth_gradient = np.gradient(
savgol_filter(x=z,
window_length=int(len(z) / 20) * 2 + 1,
polyorder=2,
mode='interp')) # window has to be odd
except np.linalg.linalg.LinAlgError: # Came across this error on 9/9/20 -- Weirdly works second time...
logger.warning('LinAlgError encountered, retrying')
smooth_gradient = np.gradient(
savgol_filter(x=z,
window_length=int(len(z) / 20) * 2 + 1,
polyorder=2,
mode='interp')) # window has to be odd
x0i = np.nanargmin(
smooth_gradient) # Index of steepest descent in data
mid = x.iloc[x0i] # X value of guessed middle index
amp = np.nanmax(z) - np.nanmin(
z) # If needed, I should look at max/min near middle only
lin = (z[z.last_valid_index()] - z[z.first_valid_index()] +
amp) / (x[z.last_valid_index()] - x[z.first_valid_index()])
theta = 5
const = z.mean()
G = 0
# add with tuples: (NAME VALUE VARY MIN MAX EXPR BRUTE_STEP)
params.add_many(('mid', mid, True, None, None, None, None),
('theta', theta, True, 0.01, None, None, None),
('amp', amp, True, 0, None, None, None),
('lin', lin, True, 0, None, None, None),
('const', const, True, None, None, None, None))
return params
