Update utilities for NAN codes:

dshemetov · dshemetov · commit 855730873cc6 · 2021-04-26T18:35:11.000-07:00
* update export utility to export, validate, and test the missing cols
* add deletion coding to the archiver, make it expect missing cols, and
  let it handle comparisons between missing and non-missing CSVs
diff --git a/_delphi_utils_python/delphi_utils/archive.py b/_delphi_utils_python/delphi_utils/archive.py
@@ -40,9 +40,11 @@
 from git import Repo
 from git.refs.head import Head
 import pandas as pd
+import numpy as np
 
 from .utils import read_params
 from .logger import get_structured_logger
+from .nancodes import Nans
 
 Files = List[str]
 FileDiffMap = Dict[str, Optional[str]]
@@ -73,8 +75,10 @@ def diff_export_csv(
         changed_df is the pd.DataFrame of common rows from after_csv with changed values.
         added_df is the pd.DataFrame of added rows from after_csv.
     """
-    export_csv_dtypes = {"geo_id": str, "val": float,
-                         "se": float, "sample_size": float}
+    export_csv_dtypes = {
+        "geo_id": str, "val": float, "se": float, "sample_size": float,
+        "missing_val": int, "missing_se": int, "missing_sample_size": int
+    }
 
     before_df = pd.read_csv(before_csv, dtype=export_csv_dtypes)
     before_df.set_index("geo_id", inplace=True)
@@ -89,12 +93,27 @@ def diff_export_csv(
     before_df_cmn = before_df.reindex(common_idx)
     after_df_cmn = after_df.reindex(common_idx)
 
-    # Exact comparisons, treating NA == NA as True
-    same_mask = before_df_cmn == after_df_cmn
-    same_mask |= pd.isna(before_df_cmn) & pd.isna(after_df_cmn)
+    # If CSVs have different columns (no missingness), mark all values as new
+    if ("missing_val" in before_df_cmn.columns) ^ ("missing_val" in after_df_cmn.columns):
+        same_mask = after_df_cmn.copy()
+        same_mask.loc[:] = False
+    else:
+        # Exact comparisons, treating NA == NA as True
+        same_mask = before_df_cmn == after_df_cmn
+        same_mask |= pd.isna(before_df_cmn) & pd.isna(after_df_cmn)
+
+    # Code deleted entries as nans with the deleted missing code
+    deleted_df = before_df.loc[deleted_idx, :].copy()
+    deleted_df[["val", "se", "sample_size"]] = np.nan
+    deleted_df[["missing_val", "missing_se", "missing_sample_size"]] = Nans.DELETED
+
+    # Code deleted entries as nans with the deleted missing code
+    deleted_df = before_df.loc[deleted_idx, :].copy()
+    deleted_df[["val", "se", "sample_size"]] = np.nan
+    deleted_df[["missing_val", "missing_se", "missing_sample_size"]] = Nans.DELETED
 
     return (
-        before_df.loc[deleted_idx, :],
+        deleted_df,
         after_df_cmn.loc[~(same_mask.all(axis=1)), :],
         after_df.loc[added_idx, :])
 
@@ -227,11 +246,11 @@ def diff_exports(self) -> Tuple[Files, FileDiffMap, Files]:
 
             deleted_df, changed_df, added_df = diff_export_csv(
                 before_file, after_file)
-            new_issues_df = pd.concat([changed_df, added_df], axis=0)
+            new_issues_df = pd.concat([deleted_df, changed_df, added_df], axis=0)
 
             if len(deleted_df) > 0:
                 print(
-                    f"Warning, diff has deleted indices in {after_file} that will be ignored")
+                    f"Diff has deleted indices in {after_file} that have been coded as nans.")
 
             # Write the diffs to diff_file, if applicable
             if len(new_issues_df) > 0:
diff --git a/_delphi_utils_python/delphi_utils/export.py b/_delphi_utils_python/delphi_utils/export.py
@@ -3,10 +3,32 @@
 from datetime import datetime
 from os.path import join
 from typing import Optional
+import logging
 
 import numpy as np
 import pandas as pd
 
+from .nancodes import Nans
+
+def filter_contradicting_missing_codes(df, sensor, metric, date, logger=None):
+    """Find values with contradictory missingness codes, filter them, and log."""
+    columns = ["val", "se", "sample_size"]
+    # Get indicies where the XNOR is true (i.e. both are true or both are false).
+    masks = [
+        ~(df[column].isna() ^ df["missing_" + column].eq(Nans.NOT_MISSING))
+        for column in columns
+    ]
+    for mask in masks:
+        if not logger is None and df.loc[mask].size > 0:
+            logger.info(
+                "Filtering contradictory missing code in " +
+                "{0}_{1}_{2}.".format(sensor, metric, date.strftime(format="%Y-%m-%d"))
+            )
+            df = df.loc[~mask]
+        elif logger is None and df.loc[mask].size > 0:
+            df = df.loc[~mask]
+    return df
+
 def create_export_csv(
     df: pd.DataFrame,
     export_dir: str,
@@ -15,7 +37,8 @@ def create_export_csv(
     metric: Optional[str] = None,
     start_date: Optional[datetime] = None,
     end_date: Optional[datetime] = None,
-    remove_null_samples: Optional[bool] = False
+    remove_null_samples: Optional[bool] = False,
+    logger: Optional[logging.Logger] = None
 ):
     """Export data in the format expected by the Delphi API.
 
@@ -39,6 +62,8 @@ def create_export_csv(
         Latest date to export or None if no maximum date restrictions should be applied.
     remove_null_samples: Optional[bool]
         Whether to remove entries whose sample sizes are null.
+    logger: Optional[logging.Logger]
+        Pass a logger object here to log information about contradictory missing codes.
 
     Returns
     ---------
@@ -64,7 +89,20 @@ def create_export_csv(
         else:
             export_filename = f"{date.strftime('%Y%m%d')}_{geo_res}_{metric}_{sensor}.csv"
         export_file = join(export_dir, export_filename)
-        export_df = df[df["timestamp"] == date][["geo_id", "val", "se", "sample_size",]]
+        expected_columns = [
+            "geo_id",
+            "val",
+            "se",
+            "sample_size",
+            "missing_val",
+            "missing_se",
+            "missing_sample_size"
+        ]
+        export_df = df[df["timestamp"] == date].filter(items=expected_columns)
+        if "missing_val" in export_df.columns:
+            export_df = filter_contradicting_missing_codes(
+                export_df, sensor, metric, date, logger=logger
+            )
         if remove_null_samples:
             export_df = export_df[export_df["sample_size"].notnull()]
         export_df = export_df.round({"val": 7, "se": 7})
diff --git a/_delphi_utils_python/tests/test_archive.py b/_delphi_utils_python/tests/test_archive.py
@@ -13,30 +13,53 @@
 import pytest
 
 from delphi_utils.archive import ArchiveDiffer, GitArchiveDiffer, S3ArchiveDiffer,\
-    archiver_from_params
+    archiver_from_params, Nans
 
-CSV_DTYPES = {"geo_id": str, "val": float, "se": float, "sample_size": float}
+CSV_DTYPES = {
+    "geo_id": str, "val": float, "se": float, "sample_size": float,
+    "missing_val": int, "missing_se":int, "missing_sample_size": int
+    }
 
 CSVS_BEFORE = {
     # Common
     "csv0": pd.DataFrame({
         "geo_id": ["1", "2", "3"],
         "val": [1.000000001, 2.00000002, 3.00000003],
         "se": [0.1, 0.2, 0.3],
-        "sample_size": [10.0, 20.0, 30.0]}),
+        "sample_size": [10.0, 20.0, 30.0],
+        "missing_val": [Nans.NOT_MISSING] * 3,
+        "missing_se": [Nans.NOT_MISSING] * 3,
+        "missing_sample_size": [Nans.NOT_MISSING] * 3,
+        }),
 
     "csv1": pd.DataFrame({
         "geo_id": ["1", "2", "3"],
         "val": [1.0, 2.0, 3.0],
         "se": [np.nan, 0.20000002, 0.30000003],
-        "sample_size": [10.0, 20.0, 30.0]}),
+        "sample_size": [10.0, 20.0, 30.0],
+        "missing_val": [Nans.NOT_MISSING] * 3,
+        "missing_se": [Nans.NOT_MISSING] * 3,
+        "missing_sample_size": [Nans.NOT_MISSING] * 3,
+        }),
 
     # Deleted
     "csv2": pd.DataFrame({
         "geo_id": ["1"],
         "val": [1.0],
         "se": [0.1],
-        "sample_size": [10.0]}),
+        "sample_size": [10.0],
+        "missing_val": [Nans.NOT_MISSING],
+        "missing_se": [Nans.NOT_MISSING],
+        "missing_sample_size": [Nans.NOT_MISSING],
+        }),
+
+    # Common, but updated with missing columns
+    "csv4": pd.DataFrame({
+        "geo_id": ["1"],
+        "val": [1.0],
+        "se": [0.1],
+        "sample_size": [10.0]
+        }),
 }
 
 CSVS_AFTER = {
@@ -45,23 +68,45 @@
         "geo_id": ["1", "2", "3"],
         "val": [1.0, 2.0, 3.0],
         "se": [0.10000001, 0.20000002, 0.30000003],
-        "sample_size": [10.0, 20.0, 30.0]}),
+        "sample_size": [10.0, 20.0, 30.0],
+        "missing_val": [Nans.NOT_MISSING] * 3,
+        "missing_se": [Nans.NOT_MISSING] * 3,
+        "missing_sample_size": [Nans.NOT_MISSING] * 3,
+        }),
 
     "csv1": pd.DataFrame({
         "geo_id": ["1", "2", "4"],
         "val": [1.0, 2.1, 4.0],
         "se": [np.nan, 0.21, np.nan],
-        "sample_size": [10.0, 21.0, 40.0]}),
+        "sample_size": [10.0, 21.0, 40.0],
+        "missing_val": [Nans.NOT_MISSING] * 3,
+        "missing_se": [Nans.NOT_MISSING] * 3,
+        "missing_sample_size": [Nans.NOT_MISSING] * 3,
+        }),
 
     # Added
     "csv3": pd.DataFrame({
         "geo_id": ["2"],
         "val": [2.0000002],
         "se": [0.2],
-        "sample_size": [20.0]}),
+        "sample_size": [20.0],
+        "missing_val": [Nans.NOT_MISSING],
+        "missing_se": [Nans.NOT_MISSING],
+        "missing_sample_size": [Nans.NOT_MISSING],
+        }),
+
+    # Common, but updated with missing columns
+    "csv4": pd.DataFrame({
+        "geo_id": ["1"],
+        "val": [1.0],
+        "se": [0.1],
+        "sample_size": [10.0],
+        "missing_val": [Nans.NOT_MISSING],
+        "missing_se": [Nans.NOT_MISSING],
+        "missing_sample_size": [Nans.NOT_MISSING],
+        }),
 }
 
-
 class TestArchiveDiffer:
 
     def test_stubs(self):
@@ -80,10 +125,14 @@ def test_diff_and_filter_exports(self, tmp_path):
         mkdir(export_dir)
 
         csv1_diff = pd.DataFrame({
-            "geo_id": ["2", "4"],
-            "val": [2.1, 4.0],
-            "se": [0.21, np.nan],
-            "sample_size": [21.0, 40.0]})
+            "geo_id": ["3", "2", "4"],
+            "val": [np.nan, 2.1, 4.0],
+            "se": [np.nan, 0.21, np.nan],
+            "sample_size": [np.nan, 21.0, 40.0],
+            "missing_val": [Nans.DELETED] + [Nans.NOT_MISSING] * 2,
+            "missing_se": [Nans.DELETED] + [Nans.NOT_MISSING] * 2,
+            "missing_sample_size": [Nans.DELETED] + [Nans.NOT_MISSING] * 2,
+            })
 
         arch_diff = ArchiveDiffer(cache_dir, export_dir)
 
@@ -106,15 +155,15 @@ def test_diff_and_filter_exports(self, tmp_path):
         # Check return values
         assert set(deleted_files) == {join(cache_dir, "csv2.csv")}
         assert set(common_diffs.keys()) == {
-            join(export_dir, f) for f in ["csv0.csv", "csv1.csv"]}
+            join(export_dir, f) for f in ["csv0.csv", "csv1.csv", "csv4.csv"]}
         assert set(new_files) == {join(export_dir, "csv3.csv")}
         assert common_diffs[join(export_dir, "csv0.csv")] is None
         assert common_diffs[join(export_dir, "csv1.csv")] == join(
             export_dir, "csv1.csv.diff")
 
         # Check filesystem for actual files
         assert set(listdir(export_dir)) == {
-            "csv0.csv", "csv1.csv", "csv1.csv.diff", "csv3.csv"}
+            "csv0.csv", "csv1.csv", "csv1.csv.diff", "csv3.csv", "csv4.csv", "csv4.csv.diff"}
         assert_frame_equal(
             pd.read_csv(join(export_dir, "csv1.csv.diff"), dtype=CSV_DTYPES),
             csv1_diff)
@@ -132,7 +181,7 @@ def test_diff_and_filter_exports(self, tmp_path):
         arch_diff.filter_exports(common_diffs)
 
         # Check exports directory just has incremental changes
-        assert set(listdir(export_dir)) == {"csv1.csv", "csv3.csv"}
+        assert set(listdir(export_dir)) == {"csv1.csv", "csv3.csv", "csv4.csv"}
         assert_frame_equal(
             pd.read_csv(join(export_dir, "csv1.csv"), dtype=CSV_DTYPES),
             csv1_diff)
@@ -259,12 +308,16 @@ def test_run(self, tmp_path, s3_client):
             assert_frame_equal(pd.read_csv(body, dtype=CSV_DTYPES), df)
 
         # Check exports directory just has incremental changes
-        assert set(listdir(export_dir)) == {"csv1.csv", "csv3.csv"}
+        assert set(listdir(export_dir)) == {"csv1.csv", "csv3.csv", "csv4.csv"}
         csv1_diff = pd.DataFrame({
-            "geo_id": ["2", "4"],
-            "val": [2.1, 4.0],
-            "se": [0.21, np.nan],
-            "sample_size": [21.0, 40.0]})
+            "geo_id": ["3", "2", "4"],
+            "val": [np.nan, 2.1, 4.0],
+            "se": [np.nan, 0.21, np.nan],
+            "sample_size": [np.nan, 21.0, 40.0],
+            "missing_val": [Nans.DELETED] + [Nans.NOT_MISSING] * 2,
+            "missing_se": [Nans.DELETED] + [Nans.NOT_MISSING] * 2,
+            "missing_sample_size": [Nans.DELETED] + [Nans.NOT_MISSING] * 2,
+            })
         assert_frame_equal(
             pd.read_csv(join(export_dir, "csv1.csv"), dtype=CSV_DTYPES),
             csv1_diff)
@@ -346,7 +399,11 @@ def test_diff_exports(self, tmp_path):
             "geo_id": ["1", "2", "3"],
             "val": [1.0, 2.0, 3.0],
             "se": [0.1, 0.2, 0.3],
-            "sample_size": [10.0, 20.0, 30.0]})
+            "sample_size": [10.0, 20.0, 30.0],
+            "missing_val": [Nans.NOT_MISSING] * 3,
+            "missing_se": [Nans.NOT_MISSING] * 3,
+            "missing_sample_size": [Nans.NOT_MISSING] * 3,
+            })
 
         # Write exact same CSV into cache and export, so no diffs expected
         csv1.to_csv(join(cache_dir, "csv1.csv"), index=False)
@@ -383,7 +440,11 @@ def test_archive_exports(self, tmp_path):
             "geo_id": ["1", "2", "3"],
             "val": [1.0, 2.0, 3.0],
             "se": [0.1, 0.2, 0.3],
-            "sample_size": [10.0, 20.0, 30.0]})
+            "sample_size": [10.0, 20.0, 30.0],
+            "missing_val": [Nans.NOT_MISSING] * 3,
+            "missing_se": [Nans.NOT_MISSING] * 3,
+            "missing_sample_size": [Nans.NOT_MISSING] * 3,
+            })
 
         # csv1.csv is now a dirty edit in the repo, and to be exported too
         csv1.to_csv(join(cache_dir, "csv1.csv"), index=False)
@@ -460,12 +521,16 @@ def test_run(self, tmp_path):
         original_branch.checkout()
 
         # Check exports directory just has incremental changes
-        assert set(listdir(export_dir)) == {"csv1.csv", "csv3.csv"}
+        assert set(listdir(export_dir)) == {"csv1.csv", "csv3.csv", "csv4.csv"}
         csv1_diff = pd.DataFrame({
-            "geo_id": ["2", "4"],
-            "val": [2.1, 4.0],
-            "se": [0.21, np.nan],
-            "sample_size": [21.0, 40.0]})
+            "geo_id": ["3", "2", "4"],
+            "val": [np.nan, 2.1, 4.0],
+            "se": [np.nan, 0.21, np.nan],
+            "sample_size": [np.nan, 21.0, 40.0],
+            "missing_val": [Nans.DELETED] + [Nans.NOT_MISSING] * 2,
+            "missing_se": [Nans.DELETED] + [Nans.NOT_MISSING] * 2,
+            "missing_sample_size": [Nans.DELETED] + [Nans.NOT_MISSING] * 2,
+            })
         assert_frame_equal(
             pd.read_csv(join(export_dir, "csv1.csv"), dtype=CSV_DTYPES),
             csv1_diff)
diff --git a/_delphi_utils_python/tests/test_export.py b/_delphi_utils_python/tests/test_export.py
