Feature/swap out xgboost (#16)

* removed xgboost from TMLE

* updated comments and docs

* updated more docs

.travis.yml

@@ -13,7 +13,7 @@ env:
 
 before_install:
 # Here we download miniconda and install the dependencies
-- pip install black coverage future joblib numpy numpydoc pandas patsy progressbar2 pygam pytest python-dateutil python-utils pytz scikit-learn scipy six statsmodels xgboost
+- pip install black coverage future joblib numpy numpydoc pandas patsy progressbar2 pygam pytest python-dateutil python-utils pytz scikit-learn scipy six statsmodels
 
 install:
 - python setup.py install

causal_curve/tmle.py

@@ -9,8 +9,8 @@
 from pandas.api.types import is_float_dtype, is_numeric_dtype
 from scipy.interpolate import interp1d
 from scipy.stats import norm
+from sklearn.ensemble import GradientBoostingClassifier, GradientBoostingRegressor
 from statsmodels.genmod.generalized_linear_model import GLM
-from xgboost import XGBClassifier, XGBRegressor
 
 from causal_curve.core import Core
 from causal_curve.utils import rand_seed_wrapper
@@ -19,7 +19,7 @@
 class TMLE(Core):
     """
     Constructs a causal dose response curve through a series of TMLE comparisons across a grid
-    of the treatment values. XGBoost is used for prediction in Q model and G model.
+    of the treatment values. Gradient boosting is used for prediction in Q model and G model.
     Assumes continuous treatment and outcome variable.
 
     WARNING:
@@ -47,18 +47,14 @@ class TMLE(Core):
         treatment values between the bin edges will be used to generate the CDRC.
 
     n_estimators: int, optional (default = 100)
-        Optional argument to set the number of learners to use when XGBoost
+        Optional argument to set the number of learners to use when sklearn
         creates TMLE's Q and G models.
 
     learning_rate: float, optional (default = 0.1)
-        Optional argument to set the XGBoost's learning rate for TMLE's Q and G models.
+        Optional argument to set the sklearn's learning rate for TMLE's Q and G models.
 
     max_depth: int, optional (default = 5)
-        Optional argument to set XGBoost's maximum depth when creating TMLE's Q and G models.
-
-    gamma: float, optional (default = 1.0)
-        Optional argument to set XGBoost's gamma parameter (regularization) when
-        creating TMLE's Q and G models.
+        Optional argument to set sklearn's maximum depth when creating TMLE's Q and G models.
 
     random_seed: int, optional (default = None)
         Sets the random seed.
@@ -115,7 +111,6 @@ def __init__(
         n_estimators=100,
         learning_rate=0.1,
         max_depth=5,
-        gamma=1.0,
         random_seed=None,
         verbose=False,
     ):
@@ -124,7 +119,6 @@ def __init__(
         self.n_estimators = n_estimators
         self.learning_rate = learning_rate
         self.max_depth = max_depth
-        self.gamma = gamma
         self.random_seed = random_seed
         self.verbose = verbose
 
@@ -190,16 +184,6 @@ def _validate_init_params(self):
         if self.max_depth <= 0:
             raise TypeError("max_depth parameter must be greater than 0")
 
-        # Checks for gamma
-        if not isinstance(self.gamma, float):
-            raise TypeError(
-                f"gamma parameter must be a float, "
-                f"but found type {type(self.gamma)}"
-            )
-
-        if self.gamma <= 0:
-            raise TypeError("gamma parameter must be greater than 0")
-
         # Checks for random_seed
         if not isinstance(self.random_seed, (int, type(None))):
             raise TypeError(
@@ -263,11 +247,10 @@ def _initial_bucket_mean_prediction(self):
             self.t_data < self.treatment_grid_bins[1]
         ]
 
-        init_model = XGBRegressor(
+        init_model = GradientBoostingRegressor(
             n_estimators=self.n_estimators,
             max_depth=self.max_depth,
             learning_rate=self.learning_rate,
-            gamma=self.gamma,
             random_state=self.random_seed,
         ).fit(X, y)
 
@@ -495,11 +478,10 @@ def _q_model(self, temp_y, temp_x, temp_t):
         X = pd.concat([temp_t, temp_x], axis=1).to_numpy()
         y = temp_y.to_numpy()
 
-        Q_model = XGBRegressor(
+        Q_model = GradientBoostingRegressor(
             n_estimators=self.n_estimators,
             max_depth=self.max_depth,
             learning_rate=self.learning_rate,
-            gamma=self.gamma,
             random_state=self.random_seed,
         ).fit(X, y)
 
@@ -525,11 +507,10 @@ def _g_model(self, temp_x, temp_t):
         X = temp_x.to_numpy()
         t = temp_t.to_numpy()
 
-        G_model = XGBClassifier(
+        G_model = GradientBoostingClassifier(
             n_estimators=self.n_estimators,
             max_depth=self.max_depth,
             learning_rate=self.learning_rate,
-            gamma=self.gamma,
             random_state=self.random_seed,
         ).fit(X, t)
 

docs/TMLE_example.rst

@@ -13,7 +13,7 @@ tool to estimate the marginal causal curve of some continuous treatment on a con
 accounting for some mild confounding effects.
 
 Compared with the package's GPS method, this TMLE method is double robust against model
-misspecification, incorporates more powerful machine learning techniques internally,
+misspecification, incorporates more powerful machine learning techniques internally (gradient boosting),
 produces significantly smaller confidence intervals, however it is not computationally efficient
 and will take longer to run.
 
@@ -39,7 +39,7 @@ References
 ----------
 
 van der Laan MJ and Rubin D. Targeted maximum likelihood learning. In: ​U.C. Berkeley Division of
-Biostatistics Working Paper Series, 2006. 
+Biostatistics Working Paper Series, 2006.
 
 van der Laan MJ and Gruber S. Collaborative double robust penalized targeted
 maximum likelihood estimation. In: The International Journal of Biostatistics 6(1), 2010.

docs/changelog.rst

@@ -4,6 +4,12 @@
 Change Log
 ==========
 
+Version 0.3.4
+-------------
+- Removed XGBoost as dependency.
+- Now using sklearn's gradient boosting implementation. 
+
+
 Version 0.3.3
 -------------
 - Misc edits to paper and bibliography

docs/conf.py

@@ -22,7 +22,7 @@
 author = 'Roni Kobrosly'
 
 # The full version, including alpha/beta/rc tags
-release = '0.3.3'
+release = '0.3.4'
 
 # -- General configuration ---------------------------------------------------
 

docs/install.rst

@@ -27,7 +27,6 @@ causal-curve requires:
 - scipy
 - six
 - statsmodels
-- xgboost
 
 
 

docs/requirements.txt

@@ -16,4 +16,3 @@ scikit-learn
 scipy
 six
 statsmodels
-xgboost

requirements.txt

@@ -16,4 +16,3 @@ scikit-learn
 scipy
 six
 statsmodels
-xgboost

setup.py

@@ -5,7 +5,7 @@
 
 setuptools.setup(
     name="causal-curve",
-    version="0.3.3",
+    version="0.3.4",
     author="Roni Kobrosly",
     author_email="[email protected]",
     description="A python library with tools to perform causal inference using \
@@ -38,7 +38,6 @@
         'scikit-learn',
         'scipy',
         'six',
-        'statsmodels',
-        'xgboost'
+        'statsmodels'
     ]
 )