def _fit_(self, fit_kwargs={}, **kwargs):
"""Fit the genotype-phenotype map for epistasis.
"""
# Fit with a linear epistasis model first, and use those values as initial guesses.
super(NonlinearEpistasisModel, self).fit()
# Expose the linear stuff to user and fill in important stuff.
self.linear.epistasis.values = self.epistasis.values
linear_phenotypes = np.dot(self.X, self.epistasis.values)
if self.linear.log_transform:
linear_phenotypes = 10**linear_phenotypes
self.linear.phenotypes = linear_phenotypes
# Construct an array of guesses, using the scale specified by user.
guess = np.ones(self.parameters.n)
# Add model's extra parameter guesses to input array
for kw in kwargs:
index = self.parameters._mapping[kw]
guess[index] = kwargs[kw]
# Curve fit the data using a nonlinear least squares fit
popt, pcov = curve_fit(self.function,
self.statistics.linear(),
self.phenotypes,
p0=guess,
**fit_kwargs)
for i in range(0, self.parameters.n):
self.parameters._set_param(i, popt[i])
# Score the fit
y_pred = self.statistics.predict()
self._score = pearson(self.phenotypes, y_pred)**2
def _fit_float_linear(self, guess_coeffs=None, fit_kwargs={}, **kwargs):
"""Fit using nonlinear least squares regression.
Parameters
----------
guess : array-like
array of guesses
fit_kwargs : dict
specific keyword arguments for scipy's curve_fit function. This is
not for parameters.
**kwargs :
guess values for `parameters` in the nonlinear function passed in as
keyword arguments
"""
# Construct an array of guesses, using the scale specified by user.
guess = np.ones(self.epistasis.n + self.parameters.n)
# Add model's extra parameter guesses to input array
for kw in kwargs:
index = self.parameters._mapping[kw]
guess[self.epistasis.n + index] = kwargs[kw]
# Create an initial guess array using fit values
if guess_coeffs is None:
# Fit with a linear epistasis model first, and use those values as initial guesses.
super(NonlinearEpistasisModel, self).fit()
# Add linear guesses to guess array
guess[:self.epistasis.n] = self.epistasis.values
else:
# Add user guesses for interactions
guess[:self.epistasis.n] = guess_coeffs
self._wrapped_function = self._nonlinear_function_wrapper(
self.function)
# Curve fit the data using a nonlinear least squares fit
popt, pcov = curve_fit(self._wrapped_function,
self.X,
self.phenotypes,
p0=guess,
**fit_kwargs)
# Set the Interaction values
self.epistasis.values = popt[0:len(self.epistasis.labels)]
# Set the other parameters from nonlinear function to fit results
for i in range(0, self.parameters.n):
self.parameters._set_param(i, popt[len(self.epistasis.labels) + i])
# Expose the linear stuff to user and fill in important stuff.
self.linear.epistasis.values = self.epistasis.values
linear_phenotypes = np.dot(self.X, self.epistasis.values)
if self.linear.log_transform:
linear_phenotypes = 10**linear_phenotypes
self.linear.phenotypes = linear_phenotypes
# Score the fit
y_pred = self._wrapped_function(self.X, *popt)
self._score = pearson(self.phenotypes, y_pred)**2
def score(self, X=None, y=None):
"""Score the model (using pearson coefficient).
Parameters
----------
X : array
array of genotypes.
y : array
array of phentoypes.
"""
ypred = self.predict(X=X)
return pearson(y, ypred)**2
def score(self, X=None, y=None, **kwargs):
"""Calculate the pearson coefficient between the models predictions and
a given y array.
"""
return pearson(self.gpm.phenotypes, self.predict(X=X))**2
def score(self, X=None, y=None):
x = self.Additive.predict(X=X)
ypred = self.minimizer.predict(x)
return pearson(y, ypred)**2
def score(self, X=None, y=None):
y_pred = X.dot(self.coef_)
return pearson(y, y_pred)**2
def score(self, X=None, y=None):
""" Calculates the squared-pearson coefficient between the model's
predictions and data.
"""
y_pred = self.predict(X)
return pearson(y, y_pred)**2