WIP: User preference prediction openday

Dockerfile

@@ -1,14 +1,16 @@
 FROM ubuntu:16.04
 EXPOSE 8000
 # TODO: Do we really need python3-dev?
-RUN apt-get update && apt-get install -y python3-pip python3 dirmngr
+RUN apt-get update && apt-get install -y python3-pip python3 dirmngr graphviz
 
 RUN apt-key adv --keyserver pgp.skewed.de --recv-key 612DEFB798507F25
 RUN echo "deb http://downloads.skewed.de/apt/xenial xenial universe" | tee -a /etc/apt/sources.list
 RUN echo "deb-src http://downloads.skewed.de/apt/xenial xenial universe" | tee -a /etc/apt/sources.list
 RUN apt-get update && apt-get install -y libboost-all-dev
 RUN apt-get update -qq && apt-get install -y python3-graph-tool
 
+RUN pip3 install jupyter
+
 COPY . /32de-python/
 
 WORKDIR /32de-python

Dockerfile-Notebook

@@ -0,0 +1,16 @@
+FROM server:latest
+
+RUN apt-get update && apt-get install -y graphviz
+RUN pip3 install jupyter
+
+# Add Tini. Tini operates as a process subreaper for jupyter. This prevents
+# kernel crashes.
+ENV TINI_VERSION v0.6.0
+ADD https://github.com/krallin/tini/releases/download/${TINI_VERSION}/tini /usr/bin/tini
+RUN chmod +x /usr/bin/tini
+ENTRYPOINT ["/usr/bin/tini", "--"]
+
+COPY . /32de-python/
+
+EXPOSE 8888
+CMD ["jupyter", "notebook", "--config", "deployment/jupyter_notebook_config.py"]

deployment/build-notebook.sh

@@ -0,0 +1,2 @@
+#!/bin/bash
+docker build -t notebook -f Dockerfile-Notebook ${1:-.}

deployment/jupyter_notebook_config.py

@@ -0,0 +1,7 @@
+# Set ip to '*' to bind on all interfaces (ips) for the public server
+c.NotebookApp.ip = '*'
+c.NotebookApp.password = u'sha1:ba8ffcde0b68:95fa25d7225a3915db1db76799f1695a0483afb4'
+c.NotebookApp.open_browser = False
+
+c.NotebookApp.port = 8888
+c.NotebookApp.allow_root = True

deployment/run-notebook.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+docker stop notebook-container
+docker rm notebook-container
+docker run --name notebook-container \
+           --publish=${1:-8888}:8888 \
+           -d \
+           notebook

deployment/run-server.sh

@@ -4,4 +4,5 @@ docker rm server-container
 docker run --name server-container \
            --publish=${1:-8000}:8000 \
            -d \
+           -e METAEXP_DEV=${2:-'false'}\
            server

deployment/stop-notebook.sh

@@ -0,0 +1,2 @@
+#!/bin/bash
+docker stop notebook-container

domain_scoring/domain_scoring.py

@@ -1,5 +1,7 @@
 from typing import List, Tuple
 import numpy
+from sklearn.ensemble import RandomForestRegressor
+
 from util.datastructures import MetaPathRatingGraph
 from util.datastructures import MetaPath
 from util.lists import all_pairs
@@ -45,8 +47,8 @@ def fit(self, metapath_graph: MetaPathRatingGraph, test_size: float = None) -> N
 
         self.classifier = self.classifier.fit(self._preprocess(x_train), y_train)
 
-        if test_size is not None:
-            print('Test accuracy is {}'.format(self.classifier.score(X=self._preprocess(x_test), y=y_test)))
+        if test_size:
+            self._test_score(x_test, y_test)
 
     def predict(self, metapath_unrated: List[MetaPath]) -> List[Tuple[MetaPath, int]]:
         """
@@ -78,8 +80,8 @@ def _transform_to_domain_values(self,
         """
         Transforms the classified ordering of all meta-paths pairs to the domain values.
 
-        :param inferred_ratings: user-defined and inferred rating for all meta-paths
-        :return: Total order of all meta-paths with values in [0,1]
+        :param inferred_ratings: user-defined and inferred rating for all meta-paths.
+        :return: Total order of all meta-paths with values in [0,1].
         """
 
         return self.domain_value_transformer.transform(metapaths_pairs, classification)
@@ -94,10 +96,10 @@ def _fit_vectorizer(self, metapath_graph: MetaPathRatingGraph) -> None:
 
     def _extract_data_labels(self, metapath_graph: MetaPathRatingGraph) -> (List[Tuple[MetaPath]], List[int]):
         """
-        Computes all pairwise tuples (a, b) of the meta-paths with their feature vector. If a is ranked higher than b
+        Computes all pairwise tuples (a, b) of the meta-paths. If a is ranked higher than b
         a > b then the label is 1, 0 otherwise.
 
-        :param metapath_graph: The meta-path graph representing the ordering of all meta-path
+        :param metapath_graph: The meta-path graph representing the ordering of all meta-path.
         :return: (x, y) The feature vector and class labels.
         """
 
@@ -112,3 +114,44 @@ def _extract_data_labels(self, metapath_graph: MetaPathRatingGraph) -> (List[Tup
             metapath_labels.append(LARGER)  # >
 
         return metapath_pairs, metapath_labels
+
+    def _test_score(self, x_test, y_test):
+        print('Test accuracy is {}'.format(self.classifier.score(X=self._preprocess(x_test), y=y_test)))
+
+class DomainScoringRegressor(DomainScoring):
+
+    def __init__(self):
+        """
+        Extracts the domain value of meta-paths via regression.
+        """
+        super()
+        self.classifier = RandomForestRegressor()
+
+    def _extract_data_labels(self, metapath_graph: MetaPathRatingGraph) -> (List[Tuple[MetaPath]], List[int]):
+        """
+        Computes all pairwise distances (a, b) of the meta-paths.
+
+        :param metapath_graph: The meta-path graph representing the ordering of all meta-path.
+        :return: (x, y) The meta-paths pairs and their respective distance.
+        """
+
+        metapath_pairs = []
+        metapath_labels = []
+
+        for superior, inferior, distance in metapath_graph.stream_meta_path_distances():
+            metapath_pairs.append((inferior, superior))
+            metapath_labels.append(distance)  # <
+
+            metapath_pairs.append((superior, inferior))
+            metapath_labels.append(-distance)  # >
+
+        return metapath_pairs, metapath_labels
+
+    def _test_score(self, x_test, y_test):
+        """
+        Converts regression result into a binary classification and uses mean accuracy.
+        """
+        test_predict = self.classifier.predict(self._preprocess(x_test))
+        score = numpy.mean(numpy.logical_and(numpy.array(y_test) > 0, numpy.array(test_predict) > 0))
+        print('Test accuracy is {}'.format(score))
+        print('R^2 is {}'.format(self.classifier.score(X=self._preprocess(x_test), y=y_test)))