Adding the option of using different target columns

+ pandas update

util/williams.py

@@ -14,20 +14,22 @@
 def main(args):
     print
     print args.pred_file_a, args.pred_file_b
-    data_a = pd.DataFrame.from_csv(args.pred_file_a, index_col=None, sep='\t')
-    data_b = pd.DataFrame.from_csv(args.pred_file_b, index_col=None, sep='\t')
+    data_a = pd.read_csv(args.pred_file_a, index_col=None, sep='\t')
+    data_b = pd.read_csv(args.pred_file_b, index_col=None, sep='\t')
     assert(len(data_a) == len(data_b))
-    data_a = data_a.sort_values(by=['mr', 'system_output', 'human_rating']).reset_index()
-    data_b = data_b.sort_values(by=['mr', 'system_output', 'human_rating']).reset_index()
-    ref = np.array(data_a['human_rating'])
-    pred_a = np.array(data_a['system_rating'])
-    pred_b = np.array(data_b['system_rating'])
+    # ensure the order of instances is the same
+    data_a = data_a.sort_values(by=['mr', 'system_output', args.target + '_human_rating']).reset_index()
+    data_b = data_b.sort_values(by=['mr', 'system_output', args.target + '_human_rating']).reset_index()
+    # get the relevant columns
+    ref = np.array(data_a[args.target + '_human_rating'])
+    pred_a = np.array(data_a[args.target + '_system_rating'])
+    pred_b = np.array(data_b[args.target + '_system_rating'])
     for corr in [scipy.stats.pearsonr, scipy.stats.spearmanr]:
         c12, _ = corr(ref, pred_a)
         c13, _ = corr(ref, pred_b)
         c23, _ = corr(pred_a, pred_b)
         print corr.__name__, c12, c13, c23, len(data_a)
-        if c12 < c13:  # swap that 1st is always bigger
+        if c12 < c13:  # swap so that 1st is always bigger
             print 'SWAPPING A-B'
             c12, c13 = c13, c12
         os.system("R --no-save --args %f %f %f %d < %s/williams.R | grep '^P-value'" %
@@ -37,6 +39,8 @@ def main(args):
 if __name__ == '__main__':
     ap = ArgumentParser(description='Williams correlation test ' +
                         '(a wrapper for the R script by Y. Graham)')
+    ap.add_argument('-t', '--target', type=str, default='quality',
+                    help='Target column (default: quality)')
     ap.add_argument('pred_file_a', type=str, help='1st file with predictions to compare')
     ap.add_argument('pred_file_b', type=str, help='2nd file with predictions to compare')
     main(ap.parse_args())