Change feedback generation, and change off topic indicator

feature_extractor.py

@@ -12,6 +12,7 @@ import os
 from itertools import chain
 import copy
 import operator
+import logging
 
 base_path = os.path.dirname(__file__)
 sys.path.append(base_path)
@@ -21,6 +22,8 @@ import util_functions
 if not base_path.endswith("/"):
     base_path=base_path+"/"
 
+log = logging.getLogger(__name__)
+
 
 class FeatureExtractor(object):
     def __init__(self):
@@ -46,7 +49,7 @@ class FeatureExtractor(object):
                 self._grammar_errors_per_character=1-(sum(self._get_grammar_errors
                     (e_set._pos,e_set._text,e_set._tokens)[0])/float(sum([len(t) for t in e_set._text])))
                 bag_feats=self.gen_bag_feats(e_set)
-                f_row_sum=numpy.sum(bag_feats[:,:])/bag_feats.shape[0]
+                f_row_sum=numpy.sum(bag_feats[:,:])
                 self._mean_f_prop=f_row_sum/float(sum([len(t) for t in e_set._text]))
                 ret = "ok"
             else:
@@ -224,7 +227,7 @@ class FeatureExtractor(object):
             if features is not None:
                 f_row_sum=numpy.sum(features[m,12:])
                 f_row_prop=f_row_sum/len(e_set._text[m])
-                if f_row_prop<(self._mean_f_prop):
+                if f_row_prop<(self._mean_f_prop/1.5) or len(e_set._text[m])<20:
                     individual_feedback['topicality']="Essay may be off topic."
             markup_string=" ".join(markup_tokens)
             individual_feedback['markup_text']=markup_string

grade.py

@@ -74,7 +74,8 @@ error_template = u"""
 def grade(grader_path,grader_config,submission,sandbox=None):
 
     if not grader_path.endswith(".p"):
-        model_path+=".p"
+        grader_path+=".p"
+
     log.debug("Grader path: {0}\n Submission: {1}".format(grader_path,submission))
     results = {'errors': [],'tests': [],'correct': False,'score': 0, 'feedback' : ""}
 
@@ -100,7 +101,7 @@ def grade(grader_path,grader_config,submission,sandbox=None):
     #Try to extract features from submission and assign score via the model
     try:
         grader_feats=grader_data['extractor'].gen_feats(grader_set)
-        feedback=grader_data['extractor'].gen_feedback(grader_set)[0]
+        feedback=grader_data['extractor'].gen_feedback(grader_set,grader_feats)[0]
         results['score']=int(grader_data['model'].predict(grader_feats)[0])
     except :
         results['errors'].append("Could not extract features and score essay.")
@@ -130,4 +131,5 @@ def grade(grader_path,grader_config,submission,sandbox=None):
     return results
 
 
+