def _apply_pi_theorem(self, df):
if self.apply_pi_theorem and self.units:
ureg = pint.UnitRegistry(auto_reduce_dimensions=True,
autoconvert_offset_to_baseunit=True)
parsed_units = _parse_units(self.units, ureg, self.verbose)
# use only original features
parsed_units = {
c: parsed_units[c]
for c in self.feateng_cols_
if not parsed_units[c].dimensionless
}
if self.verbose:
print("[AutoFeat] Applying the Pi Theorem")
pi_theorem_results = ureg.pi_theorem(parsed_units)
for i, r in enumerate(pi_theorem_results, 1):
if self.verbose:
print("[AutoFeat] Pi Theorem %i: " % i,
pint.formatter(r.items()))
# compute the final result by multiplying and taking the power of
cols = sorted(r)
# only use data points where non of the affected columns are NaNs
not_na_idx = df[cols].notna().all(axis=1)
ptr = df[cols[0]].to_numpy()[not_na_idx]**r[cols[0]]
for c in cols[1:]:
ptr *= df[c].to_numpy()[not_na_idx]**r[c]
df.loc[not_na_idx,
"PT%i_%s" %
(i, pint.formatter(r.items()).replace(" ", ""))] = ptr
return df
def to_dimensionless(quant, basis, dimensional=False):
dim = lambda q: pint.formatter(q.dimensionality.items())
quantities = { '_': dim(quant) }
for system, quantity in basis:
quantities['{0}.{1}'.format(type(system).__name__, quantity)] = dim(system._asdict()[quantity])
powers = pint.pi_theorem(quantities)[0]
power = powers.pop('_')
expression = lambda sgn: pint.formatter([ (item[0], sgn*item[1]/power) for item in powers.items() ])
if args.verbose:
print('\t\t{}'.format('=' if dimensional else '*'), expression(-1))
return eval(expression(-1 if dimensional else 1))*(ureg('') if dimensional else quant)
def _apply_pi_theorem(self, df):
if self.units:
ureg = pint.UnitRegistry(auto_reduce_dimensions=True,
autoconvert_offset_to_baseunit=True)
parsed_units = _parse_units(self.units, ureg)
# use only original features
parsed_units = {
c: parsed_units[c]
for c in self.feateng_cols if not parsed_units[c].dimensionless
}
print("[AutoFeatRegression] Applying the Pi Theorem")
pi_theorem_results = ureg.pi_theorem(parsed_units)
for i, r in enumerate(pi_theorem_results, 1):
print("Pi Theorem %i: " % i, pint.formatter(r.items()))
# compute the final result by multiplying and taking the power of
cols = sorted(r)
ptr = df[cols[0]].values**r[cols[0]]
for c in cols[1:]:
ptr *= df[c].values**r[c]
df["PT%i: %s" %
(i, pint.formatter(r.items()).replace(" ", ""))] = ptr
return df