Merge pull request #49 from titipata/revise-backend

Revise backend

backend/README.md

@@ -2,7 +2,7 @@
 
 ## Fetch events
 
-`fetch_events.py` contains functions to fetch Penn events. We can run GROBID and fetch to update data in `data/events.json` file 
+`fetch_events.py` contains functions to fetch Penn events. We can run GROBID and fetch to update data in `data/events.json` file
 as follows
 
 ```sh
@@ -20,12 +20,12 @@ python cron_fetch_events.py
 ```
 
 
-## Running Flask API locally
+## Running Hug API locally
 
-Start Flask API by running
+Start Hug API by running:
 
 ```sh
-python api.py
+hug -f hug_api.py -p 8888
 ```
 
 

backend/hug_api.py

@@ -0,0 +1,57 @@
+import hug
+import json
+import numpy as np
+from datetime import datetime
+import dateutil
+from scipy.spatial.distance import cosine
+
+# enable CORS
+api = hug.API(__name__)
+api.http.add_middleware(hug.middleware.CORSMiddleware(api))
+
+
+path_data, path_vector = 'data/events.json', 'data/events_vector.json'
+event_vectors = json.load(open(path_vector, 'r'))
+events = json.load(open(path_data, 'r'))
+event_vectors_map = {e['event_index']: e['event_vector']
+                     for e in event_vectors}
+
+
+def get_future_event(date):
+    """
+    Function return True if the event happens after now
+    """
+    try:
+        if dateutil.parser.parse(date) > datetime.now():
+            return True
+        else:
+            return False
+    except:
+        return False
+
+
+@hug.post("/recommendations")
+def recommendations(body):
+    """
+    Suggest event from a given comma separated indices.
+    Body is sent as JSON from the frontend with data as a value of a key 'payload':
+
+    {
+        "payload": data
+    }
+
+    The body is then passed as an argument to this function, as a dictionary.
+    """
+    event_indices = body['payload']
+    pref_indices = [int(event_idx) for event_idx in event_indices]
+    pref_vector = np.mean([np.array(event_vectors_map[idx])
+                           for idx in pref_indices], axis=0)
+    # get indices of event happens after current time
+    future_event_indices = [e['event_index'] for e in
+                            filter(lambda r: get_future_event(r['date_dt']), events)]
+
+    # rank indices by cosine distance and get indices
+    rank_indices = np.argsort([cosine(pref_vector, event_vectors_map[idx])
+                               for idx in future_event_indices])[0:15]
+    indices_recommendation = [future_event_indices[i] for i in rank_indices]
+    return json.dumps(indices_recommendation)