Merge pull request #3 from MITx/vik/deployment_work

Vik/deployment work

feature_extractor.py

@@ -219,6 +219,7 @@ class FeatureExtractor(object):
         #Iterate through essays and create a feedback dict for each
         all_feedback=[]
         for m in xrange(0,len(e_set._text)):
+            #Be very careful about changing these messages!
             individual_feedback={'grammar' : "Grammar: Ok.", 'spelling' : "Spelling: Ok.",
                                  'topicality' : "Topicality: Ok.", 'markup_text' : "",
                                  'prompt_overlap' : "Prompt Overlap: Ok.",
@@ -253,7 +254,7 @@ class FeatureExtractor(object):
                 if f_row_prop<(self._mean_f_prop/1.5) or len(e_set._text[m])<20:
                     individual_feedback['topicality']="Topicality: Essay may be off topic."
 
-                if(features[m,9]>.5):
+                if(features[m,9]>.6):
                     individual_feedback['prompt_overlap']="Prompt Overlap: Too much overlap with prompt."
                     log.debug(features[m,9])
 

grade.py

@@ -64,7 +64,7 @@ def grade(grader_data,grader_config,submission):
 
     if not has_error:
 
-        if(len(feedback['prompt_overlap'])>4):
+        if(len(feedback['prompt_overlap'])>20):
             results['score']=0
             results['correct']=False
 

model_creator.py

@@ -127,9 +127,14 @@ def extract_features_and_generate_model(essays,additional_array=None):
 
     cv_error_results=get_cv_error(clf2,train_feats,essays._score)
 
-    set_score = numpy.asarray(essays._score, dtype=numpy.int)
-    clf.fit(train_feats, set_score)
-
+    try:
+        set_score = numpy.asarray(essays._score, dtype=numpy.int)
+        clf.fit(train_feats, set_score)
+    except ValueError:
+        log.exception("Not enough classes (0,1,etc) in sample.")
+        set_score[0]=1
+        set_score[1]=0
+        clf.fit(train_feats, set_score)
 
     return f, clf, cv_error_results