diff --git a/docs/deepmind1.png b/docs/deepmind1.png
new file mode 100644
index 0000000..543510d
Binary files /dev/null and b/docs/deepmind1.png differ
diff --git a/docs/replicate2.png b/docs/replicate2.png
new file mode 100644
index 0000000..d1d4299
Binary files /dev/null and b/docs/replicate2.png differ
diff --git a/readme.md b/readme.md
index 64e5a3d..58eaacc 100644
--- a/readme.md
+++ b/readme.md
@@ -71,6 +71,18 @@ Changes for stability:
 - use batchnorm and dropout on channel dimensions
 - check and skip nonfinite values because for extreme inputs we can still get nan's
 
+## Replicating tensorflow behaviour
+
+I put some work into replicating the behaviour shown in the [original deepmind tensorflow notebook](https://github.com/deepmind/neural-processes/blob/master/attentive_neural_process.ipynb).
+
+Compare deepmind:
+- ![](docs/deepmind1.png)
+
+And this repo (anp_1d_regression.ipynb)
+- ![](docs/replicate2.png)
+
+It's just a qualitative comparison but we see the same kind of overfitting with uncertainty being tight where lots of data points exist, and wide where they do not. However this repo seems to miss points occasionally.
+
 
 
 ## See also: