Try converting all columns to numerical type

The dtype inference in augur.io.read_metadata does not support numerical
columns with empty values (because it calls pandas.read_csv with
na_filter=False¹). This gets around that limitation by converting
columns before querying.

I also considered infer_objects and convert_dtypes, but those are not
useful here since they only support soft (not hard) conversions².

¹ a1bfce4
² https://stackoverflow.com/a/60278450

augur/filter/include_exclude_rules.py

@@ -178,6 +178,10 @@ def filter_by_query(metadata, query) -> FilterFunctionReturn:
     set()
 
     """
+    # Try converting all columns to numeric.
+    for column in metadata.columns:
+        metadata[column] = pd.to_numeric(metadata[column], errors='ignore')
+
     return set(metadata.query(query).index.values)
 
 

tests/functional/filter/cram/filter-query-numerical.t

@@ -12,15 +12,13 @@ Create metadata file for testing.
   > SEQ_4	
   > ~~
 
-Ideally, the 'coverage' column should be query-able by numerical comparisons.
-This does not currently work since the empty string is causing that column to be
-parsed as a non-numerical type.
+The 'coverage' column should be query-able by numerical comparisons.
 
   $ ${AUGUR} filter \
   >  --metadata metadata.tsv \
   >  --query "coverage >= 0.95" \
   >  --output-strains filtered_strains.txt > /dev/null
-  ERROR: Internal Pandas error when applying query:
-  	'>=' not supported between instances of 'str' and 'float'
-  Ensure the syntax is valid per <https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#indexing-query>.
-  [2]
+
+  $ sort filtered_strains.txt
+  SEQ_2
+  SEQ_3