Rm pheno-data for cases without sample(s).

gdc.py

@@ -416,14 +416,14 @@ def get_samples_clinical(projects=None):
     # be reduced into "naked" ``dict``). Therefore, it cannot be normalized
     # correctly with ``record_path `` "samples". Use the raw json instead.
     # Besides, there are cases (34 by 12/11/2017) which doesn't have any
-    # samples and thus doesn't have the key "samples". Such cases are fixed
-    # with the ``[{}]`` default for ``json_normalize``.
-    samples_json = []
-    for r in res:
-        r.setdefault('samples', [{}])
-        samples_json.append(r)
-    samples_df = pd.io.json.json_normalize(samples_json, 'samples', 'id',
-                                           record_prefix='samples.')
+    # samples and thus doesn't have the key "samples". Ignore them.
+#    for r in res:
+#        r.setdefault('samples', [{}])
+#        samples_json.append(r)
+    samples_df = pd.io.json.json_normalize(
+            [r for r in res if 'samples' in r], 'samples', 'id',
+            record_prefix='samples.'
+        )
     return pd.merge(cases_df, samples_df, how='inner', on='id')