From 5921d3627c2651d8b55c421c2f0a0af8ae5f942f Mon Sep 17 00:00:00 2001
From: Rawan Mahdi <111605471+rawanmahdi@users.noreply.github.com>
Date: Sat, 15 Jul 2023 13:30:51 -0400
Subject: [PATCH] Added keras support (#345)

---
 python/README.md                    |  2 +-
 python/examples/keras_classifier.py | 52 +++++++++++++++++++++++++++++
 python/shaprpy/explain.py           | 16 +++++++--
 3 files changed, 67 insertions(+), 3 deletions(-)
 create mode 100644 python/examples/keras_classifier.py

diff --git a/python/README.md b/python/README.md
index 740fa3a85..6b5b1238a 100644
--- a/python/README.md
+++ b/python/README.md
@@ -56,7 +56,7 @@ df_shapley, pred_explain, internal, timing = explain(
 print(df_shapley)
 ```
 
-`shaprpy` knows how to explain predictions from models from `sklearn` and `xgboost`. 
+`shaprpy` knows how to explain predictions from models from `sklearn`, `keras` and `xgboost`. 
 For other models, one can provide a custom `predict_model` function (and optionally a custom `get_model_specs`) to `shaprpy.explain`.
 
 See `/examples` for runnable examples, including an example of a custom PyTorch model.
diff --git a/python/examples/keras_classifier.py b/python/examples/keras_classifier.py
new file mode 100644
index 000000000..73617443e
--- /dev/null
+++ b/python/examples/keras_classifier.py
@@ -0,0 +1,52 @@
+from keras import Sequential
+from keras import layers
+from keras import utils
+from shaprpy import explain
+from shaprpy.datasets import load_binary_iris
+
+
+dfx_train, dfx_test, dfy_train, dfy_test = load_binary_iris()
+
+utils.set_random_seed(1)
+
+## Build model
+model = Sequential([
+    layers.Dense(units=8, activation='relu'), 
+    layers.Dense(units=16, activation='relu'), 
+    layers.Dense(units=8, activation='relu'), 
+    layers.Dense(units=1, activation='sigmoid')
+])
+model.compile(optimizer="adam", 
+              loss ="binary_crossentropy", 
+              metrics=["accuracy"])
+
+## Fit Model
+model.fit(dfx_train, dfy_train, 
+          epochs=10, 
+          validation_data=(dfx_test, dfy_test))
+## Shapr
+df_shapley, pred_explain, internal, timing = explain(
+    model = model,
+    x_train = dfx_train,
+    x_explain = dfx_test,
+    approach = 'empirical',
+    prediction_zero = dfy_train.mean().item(),
+)
+print(df_shapley)
+
+""" 
+       none  sepal length (cm)  sepal width (cm)  petal length (cm)  \
+1  0.494737           0.042263          0.037911           0.059232   
+2  0.494737           0.034217          0.029183           0.045027   
+3  0.494737           0.045776          0.031752           0.058278   
+4  0.494737           0.014977          0.032691           0.014280   
+5  0.494737           0.022742          0.025851           0.027427   
+
+   petal width (cm)  
+1          0.058412  
+2          0.053639  
+3          0.070650  
+4          0.018697  
+5          0.026814  
+
+ """
\ No newline at end of file
diff --git a/python/shaprpy/explain.py b/python/shaprpy/explain.py
index 1d77cb7d1..0b214a3f5 100644
--- a/python/shaprpy/explain.py
+++ b/python/shaprpy/explain.py
@@ -45,7 +45,7 @@ def explain(
     Parameters
     ----------
     model: The model whose predictions we want to explain. 
-      `shaprpy` natively supports `sklearn` and `xgboost` models.
+      `shaprpy` natively supports `sklearn`, `xgboost` and `keras` models.
       Unsupported models can still be explained by passing `predict_model` and (optionally) `get_model_specs`.
     x_explain: Contains the the features, whose predictions ought to be explained.
     x_train: Contains the data used to estimate the (conditional) distributions for the features
@@ -79,7 +79,8 @@ def explain(
       and a pandas.DataFrame to compute predictions for. The function must give the prediction as a numpy.Array.
       `None` (the default) uses functions specified internally.
       Can also be used to override the default function for natively supported model classes.
-    get_model_specs: An optional function for checking model/data consistency when `model` is not natively supported.
+    get_model_specs: An optional function for checking model/data consistency when `model` is not natively supported. 
+      This method has yet to be implemented for keras models.
       The function takes `model` as argument and provides a `dict with 3 elements:
       - labels: list[str] with the names of each feature.
       - classes: list[str] with the classes of each features.
@@ -305,6 +306,17 @@ def prebuilt_predict_model(model):
       return lambda m, x: m.predict(xgb.DMatrix(x))
   except:
     pass
+  
+  # Look for keras
+  try:
+    from keras.models import Model
+    if isinstance(model, Model):
+      def predict_fn(m,x):
+        pred = m.predict(x)
+        return pred.reshape(pred.shape[0],)
+      return predict_fn
+  except:
+    pass
 
   return None