def test_bilinear_regression_estimator():
from numpy.testing import assert_array_almost_equal, assert_almost_equal
a = np.arange(1, 5)
b = np.arange(6, 10)
abmat = a[:, np.newaxis] * b
X = np.random.randn(100, abmat.size)
y = X.dot(abmat.ravel())
from sklearn.linear_model import LinearRegression
lr = LinearRegression(fit_intercept=False)
br = BilinearRegression(fit_intercept=False)
lr.fit(X, y)
br.fit(X.reshape(X.shape[0], 1, 1, -1), y)
br_coef = br.coef_b_.ravel() * br.coef_a_.ravel()[0]
assert_array_almost_equal(lr.coef_, br_coef)
br.fit(X.reshape(X.shape[0], 1, -1, 1), y)
br_coef = br.coef_a_.ravel() * br.coef_b_.ravel()[0]
assert_array_almost_equal(lr.coef_, br_coef)
offset = 2
lr.fit_intercept = True
br.fit_intercept = True
lr.fit(X, y + offset)
X.shape = X.shape[0], 1, len(a), len(b)
br.fit(X, y + offset)
assert_almost_equal(br.intercept_, lr.intercept_)
def _get_estimator(self):
if self.estimator is None:
estimator = LinearRegression()
else:
estimator = clone(self.estimator)
estimator.fit_intercept = False
return estimator
def correlationRegression():
scatddx = "POA_Irradiance"
scatddy = "Specific_ACPower"
reg = Reg()
reg.fit_intercept = False
reg.fit(np.vstack(df[scatddx]), df[scatddy])
pred = reg.predict(np.vstack(df[scatddx]))
data = go.Scatter(x=df[scatddx],
y=df[scatddy],
mode='markers',
opacity=0.8,
marker={'size': 4},
showlegend=False)
layout = go.Layout(
xaxis=dict(title=scatddx,
titlefont=dict(size=10),
tickfont=dict(size=10),
color='#e6e6e6',
gridwidth=0.5,
gridcolor="#333",
zerolinecolor="#333"),
yaxis=dict(title=scatddy,
titlefont=dict(size=10),
tickfont=dict(size=10),
color='#e6e6e6',
gridwidth=0.5,
gridcolor="#333",
zerolinecolor="#333"),
plot_bgcolor='#282828',
paper_bgcolor='#222222',
margin=dict(l=5, r=5, t=50, b=50),
)
figure = go.Figure(data=data, layout=layout)
figure.add_trace(
go.Scatter(x=df[scatddx],
y=pred,
mode='lines',
opacity=0.8,
line={'width': 2},
showlegend=False))
return dict(msgCorrReg=T('Correlation and Regression'),
figCorrReg=figure.to_json())
def linear_model(x, attr, xvars, fit_intercept=None, name=None,
cut=None, residuals=True, quiet=True,
model='LinearRegression'):
"""Make a linear model for attr based on xvars as free parameters.
Currently only model='LinearRegression' implmented.
Uses scikit-learn.
residuals: Name of attribute for residuals (default: attr+"_residuals")
"""
if model is not 'LinearRegression':
raise Exception("Currently only model='LinearRegression' implmented.")
from sklearn.linear_model import LinearRegression
import pandas as pd
import numpy as np
import xarray as xr
lm = LinearRegression()
if not name:
name = '{:}_model'.format(attr)
if not quiet:
print '\nUsing scikit-learn LinearRegression to build model for {:} from variables:\n {:}'.format(attr, str(xvars))
allattrs = xvars + [attr]
if cut:
allattrs += [cut]
xx = x.reset_coords()[allattrs].where(np.isfinite(x.reset_coords()[attr]), drop=True)
df_xvars0 = xx[xvars].to_dataframe()
if cut:
df_xvars = xx[xvars].where(xx[cut] == 1, drop=True).to_dataframe()
xdata = xx[attr].where(xx[cut] == 1, drop=True).data
if not quiet:
print '\nUsing following cut in buildiing model'
print xx[cut]
else:
df_xvars = df_xvars0
xdata = xx[attr].data
if fit_intercept is not None:
lm.fit_intercept = fit_intercept
lm.fit(df_xvars, xdata)
#ft = pd.DataFrame(zip(df_xvars.columns,lm.coef_), columns=['params','estimatedCoefficients'])
x[name] = xr.DataArray(lm.predict(df_xvars0), coords=[('time', df_xvars0.index)])
#x[name] = (['time'], lm.predict(df_xvars0))
x[name].attrs.update(**lm.get_params())
x[name].attrs['unit'] = x[attr].attrs.get('unit','')
x[name].attrs['doc'] = 'LinearRegression scikit-learn model for {:} training data'.format(attr)
x[name].attrs['model'] = model
x[name].attrs['variables'] = xvars
x[name].attrs['coef_'] = lm.coef_
x[name].attrs['intercept_'] = lm.intercept_
x[name].attrs['score'] = lm.score(df_xvars, xdata)
if not quiet:
print '\n****Model Results****'
print x.reset_coords()[name]
if residuals:
if not isinstance(residuals, str):
residuals = '{:}_residuals'.format(attr)
x[residuals] = (['time'], x[attr]-x[name])
x[residuals].attrs['doc'] = 'Residuals for {:} based on LinearRegression model {:}'.format(attr, name)
if not quiet:
print '\n****Model Residuals****'
print x.reset_coords()[residuals]
return x
def linear_model(x,
attr,
xvars,
fit_intercept=None,
name=None,
cut=None,
residuals=True,
quiet=True,
model='LinearRegression'):
"""Make a linear model for attr based on xvars as free parameters.
Currently only model='LinearRegression' implmented.
Uses scikit-learn.
residuals: Name of attribute for residuals (default: attr+"_residuals")
"""
if model is not 'LinearRegression':
raise Exception("Currently only model='LinearRegression' implmented.")
from sklearn.linear_model import LinearRegression
import pandas as pd
import numpy as np
import xarray as xr
lm = LinearRegression()
if not name:
name = '{:}_model'.format(attr)
if not quiet:
print '\nUsing scikit-learn LinearRegression to build model for {:} from variables:\n {:}'.format(
attr, str(xvars))
allattrs = xvars + [attr]
if cut:
allattrs += [cut]
xx = x.reset_coords()[allattrs].where(np.isfinite(x.reset_coords()[attr]),
drop=True)
df_xvars0 = xx[xvars].to_dataframe()
if cut:
df_xvars = xx[xvars].where(xx[cut] == 1, drop=True).to_dataframe()
xdata = xx[attr].where(xx[cut] == 1, drop=True).data
if not quiet:
print '\nUsing following cut in buildiing model'
print xx[cut]
else:
df_xvars = df_xvars0
xdata = xx[attr].data
if fit_intercept is not None:
lm.fit_intercept = fit_intercept
lm.fit(df_xvars, xdata)
#ft = pd.DataFrame(zip(df_xvars.columns,lm.coef_), columns=['params','estimatedCoefficients'])
x[name] = xr.DataArray(lm.predict(df_xvars0),
coords=[('time', df_xvars0.index)])
#x[name] = (['time'], lm.predict(df_xvars0))
x[name].attrs.update(**lm.get_params())
x[name].attrs['unit'] = x[attr].attrs.get('unit', '')
x[name].attrs[
'doc'] = 'LinearRegression scikit-learn model for {:} training data'.format(
attr)
x[name].attrs['model'] = model
x[name].attrs['variables'] = xvars
x[name].attrs['coef_'] = lm.coef_
x[name].attrs['intercept_'] = lm.intercept_
x[name].attrs['score'] = lm.score(df_xvars, xdata)
if not quiet:
print '\n****Model Results****'
print x.reset_coords()[name]
if residuals:
if not isinstance(residuals, str):
residuals = '{:}_residuals'.format(attr)
x[residuals] = (['time'], x[attr] - x[name])
x[residuals].attrs[
'doc'] = 'Residuals for {:} based on LinearRegression model {:}'.format(
attr, name)
if not quiet:
print '\n****Model Residuals****'
print x.reset_coords()[residuals]
return x
