Merge pull request #19 from preinaj/save-predictions-in-evaluator

Save predictions in evaluator

ADLStream/evaluation/base_evaluator.py

@@ -96,6 +96,7 @@ def __init__(
 
         self._create_results_file()
 
+        self.visualizer = None
         if self.show_plot or self.plot_file is not None:
             self.visualizer = EvaluationVisualizer(self.dataset_name, self.ylabel)
 
@@ -105,7 +106,8 @@ def _create_results_file(self):
                 f.write("timestamp,instances,metric\n")
 
     def start(self):
-        self.visualizer.start()
+        if self.visualizer is not None:
+            self.visualizer.start()
         if self.predictions_file:
             self.predictions_file = open(self.predictions_file, "a")
         if self.results_file:

ADLStream/models/transformer.py

@@ -180,11 +180,12 @@ def call(self, x, enc_output, training, look_ahead_mask):
         Returns:
             tf.Tensor: Tensor with shape (batch_size, target_seq_len, d_model)
         """
+        if len(x.shape) == 4:
+            x = tf.squeeze(x, axis=[1])
 
         attn1 = self.mha1(x, x, x, look_ahead_mask)
         attn1 = self.dropout1(attn1, training=training)
         out1 = self.layernorm1(attn1 + x)
-
         attn2 = self.mha2(enc_output, enc_output, out1, None)
         attn2 = self.dropout2(attn2, training=training)
         out2 = self.layernorm2(attn2 + out1)
@@ -473,6 +474,7 @@ def predict_step(self, data):
         if self.attribute != None:
             tar_inp = tf.gather(x, [self.attribute], axis=-1)
             tar_inp = tf.gather(tar_inp, [tar_inp.shape[1] - 1], axis=1)
+            tar_inp = tf.squeeze(tar_inp, axis=[1])
 
         for i in range(self.target_shape[0]):
             output = self((x, combined_mask, tar_inp), False)
@@ -567,7 +569,6 @@ def Transformer(
         optimizer = tf.keras.optimizers.Adam(
             learning_rate, beta_1=0.9, beta_2=0.98, epsilon=1e-9
         )
-
     att_inp = input_shape[-1]
     att_out = output_shape[-1]