def exposure_model(self, model, print_model_results=True):
"""Estimation of g(A=1,W), which is Pr(A=1|W)
model:
-Independent variables to predict the exposure. Example) 'var1 + var2 + var3'
print_model_results:
-Whether to print the fitted model results. Default is True (prints results)
"""
self._exp_model = self._exposure + ' ~ ' + model
fitmodel = propensity_score(self.df,
self._exp_model,
mresult=print_model_results)
self.gA1 = fitmodel.predict(self.df)
self._fit_exposure_model = True
def exposure_model(self, model, print_results=True):
"""Used to specify the propensity score model. Model used to predict the exposure via a logistic regression
model
model:
-Independent variables to predict the exposure. Example) 'var1 + var2 + var3'
print_results:
-Whether to print the fitted model results. Default is True (prints results)
"""
self._exp_model = self._exposure + ' ~ ' + model
fitmodel = propensity_score(self.df,
self._exp_model,
print_results=print_results)
self.df['ps'] = fitmodel.predict(self.df)
self._fit_exposure_model = True
def exposure_model(self,
model,
custom_model=None,
bound=False,
print_results=True):
"""Estimation of Pr(A=1|L), which is termed as g(A=1|L) in the literature
Parameters
----------
model : str
Independent variables to predict the exposure. Example) 'var1 + var2 + var3'
custom_model : optional
Input for a custom model that is used in place of the logit model (default). The model must have the
"fit()" and "predict()" attributes. Both sklearn and supylearner are supported as custom models. In the
background, TMLE will fit the custom model and generate the predicted probablities
bound : float, list, optional
Value between 0,1 to truncate predicted probabilities. Helps to avoid near positivity violations.
Specifying this argument can improve finite sample performance for random positivity violations. However,
inference becomes limited to the restricted population. Default is False, meaning no truncation of
predicted probabilities occurs. Providing a single float assumes symmetric trunctation. A collection of
floats can be provided for asymmetric trunctation
print_results : bool, optional
Whether to print the fitted model results. Default is True (prints results)
"""
self._exp_model = self._exposure + ' ~ ' + model
# Step 3) Estimation of g-model (exposure model)
if custom_model is None:
fitmodel = propensity_score(self.df,
self._exp_model,
print_results=print_results)
self.g1W = fitmodel.predict(self.df)
# User-specified prediction model
else:
data = patsy.dmatrix(model + ' - 1', self.df)
self.g1W = _exposure_machine_learner(xdata=np.asarray(data),
ydata=np.asarray(
self.df[self._exposure]),
ml_model=custom_model,
print_results=print_results)
self.g0W = 1 - self.g1W
if bound: # Bounding predicted probabilities if requested
self.g1W = self._bounding(self.g1W, bounds=bound)
self.g0W = self._bounding(self.g0W, bounds=bound)
self._fit_exposure_model = True
def exposure_model(self, model, print_results=True):
"""Specify the propensity score / inverse probability weight model. Model used to predict the exposure via a
logistic regression model
Parameters
----------
model : str
Independent variables to predict the exposure. For example, 'var1 + var2 + var3'
print_results : bool, optional
Whether to print the fitted model results. Default is True (prints results)
"""
self._exp_model = self._exposure + ' ~ ' + model
fitmodel = propensity_score(self.df,
self._exp_model,
print_results=print_results)
self.df['ps'] = fitmodel.predict(self.df)
self._fit_exposure_model = True
def exposure_model(self, model, print_results=True):
r"""Specify the propensity score / inverse probability weight model. Model used to predict the exposure via a
logistic regression model. This model estimates
.. math::
\widehat{\Pr}(A=1|L) = logit^{-1}(\widehat{\beta_0} + \widehat{\beta} L)
Parameters
----------
model : str
Independent variables to predict the exposure. For example, 'var1 + var2 + var3'
print_results : bool, optional
Whether to print the fitted model results. Default is True (prints results)
"""
self._exp_model = self._exposure + ' ~ ' + model
fitmodel = propensity_score(self.df, self._exp_model, weights=self._weight_, print_results=print_results)
self.df['_ps_'] = fitmodel.predict(self.df)
self._fit_exposure_ = True
def missing_model(self, model, custom_model=None, print_results=True):
"""Estimation of Pr(M=1|A,L), which is the missing data mechanism for the outcome. The corresponding observation
probabilities are used to update the clever covariates for estimation of Qn.
The initial estimate of Q is still based on complete observations only
Parameters
----------
model : str
Independent variables to predict the exposure. Example) 'var1 + var2 + var3'. The treatment must be
included for the missing data model
custom_model : optional
Input for a custom model that is used in place of the logit model (default). The model must have the
"fit()" and "predict()" attributes. Both sklearn and supylearner are supported as custom models. In the
background, TMLE will fit the custom model and generate the predicted probablities
print_results : bool, optional
Whether to print the fitted model results. Default is True (prints results)
"""
# Error if no missing outcome data
if not self._miss_flag:
raise ValueError(
"No missing outcome data is present in the data set")
# Warning if exposure is not included in the missingness of outcome model
if self._exposure not in model:
warnings.warn(
"For the specified missing outcome model, the exposure variable should be included in the "
"model", UserWarning)
self._miss_model = self._missing_indicator + ' ~ ' + model
# Step 3b) Prediction for M if missing outcome data exists
if custom_model is None: # Logistic Regression model for predictions
fitmodel = propensity_score(self.df,
self._miss_model,
print_results=print_results)
dfx = self.df.copy()
dfx[self._exposure] = 1
self.m1W = fitmodel.predict(dfx)
dfx = self.df.copy()
dfx[self._exposure] = 0
self.m0W = fitmodel.predict(dfx)
# User-specified model
else:
data = patsy.dmatrix(model + ' - 1', self.df)
dfx = self.df.copy()
dfx[self._exposure] = 1
adata = patsy.dmatrix(model + ' - 1', dfx)
dfx = self.df.copy()
dfx[self._exposure] = 0
ndata = patsy.dmatrix(model + ' - 1', dfx)
self.m1W, self.m0W = _missing_machine_learner(
xdata=np.array(data),
mdata=self.df[self._missing_indicator],
all_a=adata,
none_a=ndata,
ml_model=custom_model,
print_results=print_results)
self._fit_missing_model = True
def exposure_model(self,
model,
custom_model=None,
bound=False,
print_results=True):
"""Estimation of Pr(A=1|L), which is termed as g(A=1|L) in the literature
Parameters
----------
model : str
Independent variables to predict the exposure. Example) 'var1 + var2 + var3'
custom_model : optional
Input for a custom model that is used in place of the logit model (default). The model must have the
"fit()" and "predict()" attributes. Both sklearn and supylearner are supported as custom models. In the
background, TMLE will fit the custom model and generate the predicted probablities
bound : float, list, optional
Value between 0,1 to truncate predicted probabilities. Helps to avoid near positivity violations.
Specifying this argument can improve finite sample performance for random positivity violations. However,
inference becomes limited to the restricted population. Default is False, meaning no truncation of
predicted probabilities occurs. Providing a single float assumes symmetric trunctation. A collection of
floats can be provided for asymmetric trunctation
print_results : bool, optional
Whether to print the fitted model results. Default is True (prints results)
"""
self._exp_model = self._exposure + ' ~ ' + model
# Step 3) Estimation of g-model (exposure model)
if custom_model is None:
fitmodel = propensity_score(self.df,
self._exp_model,
print_results=print_results)
self.g1W = fitmodel.predict(self.df)
# User-specified prediction model
else:
data = patsy.dmatrix(model + ' - 1', self.df)
try:
fm = custom_model.fit(X=data, y=self.df[self._outcome])
except TypeError:
raise TypeError(
"Currently custom_model must have the 'fit' function with arguments 'X', 'y'. This "
"covers both sklearn and supylearner. If there is a predictive model you would "
"like to use, please open an issue at https://github.com/pzivich/zepid and I "
"can work on adding support")
if print_results and hasattr(fm, 'summarize'):
fm.summarize()
if hasattr(fm, 'predict_proba'):
self.g1W = fm.predict_proba(data)[:, 1]
elif hasattr(fm, 'predict'):
self.g1W = fm.predict(data)
else:
raise ValueError(
"Currently custom_model must have 'predict' or 'predict_proba' attribute"
)
self.g0W = 1 - self.g1W
if bound: # Bounding predicted probabilities if requested
self.g1W = self._bounding(self.g1W, bounds=bound)
if bound: # Bounding predicted probabilities if requested
self.g0W = self._bounding(self.g0W, bounds=bound)
self._fit_exposure_model = True
def exposure_model(self,
model,
custom_model=None,
bound=False,
print_results=True):
"""Estimation of g(A=1,W), which is Pr(A=1|W)
model:
-Independent variables to predict the exposure. Example) 'var1 + var2 + var3'
custom_model:
-Input for a custom model. The model must already be estimated and have the "predict()" attribute to work.
This allows the user to use any outside model they want and bring it into TMLE. For example, you can use
any sklearn model, ensemble model (SuPyLearner), or just different statsmodels regression models than
logistic regression. Please see online for an example
NOTE: if a custom model is used, patsy in the background does the data filtering from the equation above.
The equation order of variables MUST match that of the custom_model when it was fit. If not, this can lead
to unexpected estimates
bound:
-Value between 0,1 to truncate predicted probabilities. Helps to avoid near positivity violations. Default
is False, meaning no truncation of predicted probabilities occurs. Providing a single float assumes
symmetric trunctation. A collection of floats can be provided for asymmetric trunctation
print_results:
-Whether to print the fitted model results. Default is True (prints results)
"""
self._exp_model = self._exposure + ' ~ ' + model
# Base logistic regression model to generated predicted probabilities
if custom_model is None:
fitmodel = propensity_score(self.df,
self._exp_model,
print_results=print_results)
self.gW = fitmodel.predict(self.df)
if bound is not False: # Bounding predicted probabilities if requested
self.gW = self._bounding(self.gW, bounds=bound)
# User-specified prediction model
else:
try: # This two-stage 'try' filters whether the data needs an intercept, then has the predict() attr
data = patsy.dmatrix(model, self.df)
try:
self.gW = custom_model.predict(data)
if bound is not False: # Bounding predicted probabilities if requested
self.gW = self._bounding(self.gW, bounds=bound)
except AttributeError:
raise AttributeError(
"custom_model does not have the 'predict()' attribute")
except ValueError:
data = patsy.dmatrix(model + ' - 1', self.df)
try:
if hasattr(custom_model, 'predict_proba'):
self.gW = custom_model.predict_proba(data)[:, 1]
else:
self.gW = custom_model.predict(data)
if bound is not False: # Bounding predicted probabilities if requested
self.gW = self._bounding(self.gW, bounds=bound)
except AttributeError:
raise AttributeError(
"custom_model does not have the 'predict()' or 'predict_proba()' attribute"
)
# Setting flag for fit() check to make sure exposure model was specified
self._fit_exposure_model = True