diff --git a/Project/advanced_approach/README.assets/1532696294043.png b/Project/advanced_approach/README.assets/1532696294043.png
new file mode 100644
index 0000000..ebe6a5a
Binary files /dev/null and b/Project/advanced_approach/README.assets/1532696294043.png differ
diff --git a/Project/advanced_approach/README.assets/1532696345413.png b/Project/advanced_approach/README.assets/1532696345413.png
new file mode 100644
index 0000000..ad58f72
Binary files /dev/null and b/Project/advanced_approach/README.assets/1532696345413.png differ
diff --git a/Project/advanced_approach/README.assets/1532696365808.png b/Project/advanced_approach/README.assets/1532696365808.png
new file mode 100644
index 0000000..4b29f4f
Binary files /dev/null and b/Project/advanced_approach/README.assets/1532696365808.png differ
diff --git a/Project/advanced_approach/README.assets/1532696382476.png b/Project/advanced_approach/README.assets/1532696382476.png
new file mode 100644
index 0000000..7396b44
Binary files /dev/null and b/Project/advanced_approach/README.assets/1532696382476.png differ
diff --git a/Project/advanced_approach/README.md b/Project/advanced_approach/README.md
index b5a0a07..e56a296 100644
--- a/Project/advanced_approach/README.md
+++ b/Project/advanced_approach/README.md
@@ -36,27 +36,29 @@ In short: the user interface for methods provided in [twitter_learning.py](#twit
### load datasets
-
+![1532696294043](README.assets/1532696294043.png)
+
+
* setup the root folder for the json files containing our twitter samples
* then you can set the range of files that will be loaded. Also you can setup more preprocessing steps, like top-emoji usage filtering or only load samples containing a specific set of emojis (given as string)
### create/save/load classifier
-
+![1532696345413](README.assets/1532696345413.png)
* create a new keras pipeline including a vectorizer. you can setup the number of neurons and the activation function per layer (the last layer will be automatically adjusted to fit the label-dimension). if doc2vec is deactivated, tfidf is used. In the image above is the configuration we used for our final classifier.
* you can also save a trained classifier and load it later again by a .pipeline configuration file (either by selecting them in the file selector box or by give the full path to the `clf_file` box)
### train classifier
-
+![1532696365808](README.assets/1532696365808.png)
* if sample data and classifier are loaded, the classifier can be trained here
### Test single predictions
-
+![1532696382476](README.assets/1532696382476.png)
* in the playground tab you can predict single sentences and get the nearest emoji in sentiment space of a given emoji set. Also you can plot the predictions of samples with given labels (as emoji) of the validation set of the currently loaded twitter data