Update README.md

README.md

@@ -45,3 +45,16 @@ Code output is:
  ```
 
 As the probability of 1 (=It's a paraphrase) is 0.84 and the probability of 0 (=It is not a paraphrase) is 0.15, we can conclude, for our previous example, that "could be included" is a paraphrase of "may be addressed".
+
+
+
+# Evaluation results
+
+We have used as test dataset 16500 pairs of phrases human tagged. 
+
+Metrics obtained are:
+
+```
+metrics={'test_loss': 0.5660144090652466, 'test_accuracy': 0.8170742794799527, 'test_precision': 0.7043977055449331, 'test_recall': 0.5978578383641675, 'test_f1': 0.6467696629213483, 'test_matthews_correlation': 0.5276716223607356, 'test_runtime': 19.3345, 'test_samples_per_second': 568.88, 'test_steps_per_second': 17.792}
+
+```