Add neural net class

pywatts/neural.py

@@ -0,0 +1,31 @@
+import tensorflow as tf
+
+
+class Net:
+    regressor = None
+
+
+    def net(self, feature_cols):
+        self.regressor = tf.estimator.DNNRegressor(feature_columns=feature_cols,
+                                                   hidden_units=[50, 50],
+                                                   model_dir='tf_pywatts_model')
+
+    def pywatts_input_fn(X, y=None, num_epochs=None, shuffle=True, batch_size=400):
+        return tf.estimator.inputs.pandas_input_fn(x=X,
+                                                   y=y,
+                                                   num_epochs=num_epochs,
+                                                   shuffle=shuffle,
+                                                   batch_size=batch_size)
+
+    def train(self, training_data, steps):
+        self.regressor.train(input_fn=self.pywatts_input_fn(training_data, num_epochs=None, shuffle=True), steps=steps)
+
+    def evaluate(self, eval_data):
+        self.regressor.evaluate(input_fn=self.pywatts_input_fn(eval_data, num_epochs=1, shuffle=False), steps=1)
+
+    def predict1h(self, df):
+        df = df.drop(['month', 'day', 'hour'])
+        predictions = self.regressor.predict(input_fn=self.pywatts_input_fn(df, num_epochs=1, shuffle=False))
+
+
+