perspective-python 4.2.0__cp311-abi3-win_amd64.whl

perspective/tests/table/test_view.py

@@ -0,0 +1,2261 @@
+#  ┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
+#  ┃ ██████ ██████ ██████       █      █      █      █      █ █▄  ▀███ █       ┃
+#  ┃ ▄▄▄▄▄█ █▄▄▄▄▄ ▄▄▄▄▄█  ▀▀▀▀▀█▀▀▀▀▀ █ ▀▀▀▀▀█ ████████▌▐███ ███▄  ▀█ █ ▀▀▀▀▀ ┃
+#  ┃ █▀▀▀▀▀ █▀▀▀▀▀ █▀██▀▀ ▄▄▄▄▄ █ ▄▄▄▄▄█ ▄▄▄▄▄█ ████████▌▐███ █████▄   █ ▄▄▄▄▄ ┃
+#  ┃ █      ██████ █  ▀█▄       █ ██████      █      ███▌▐███ ███████▄ █       ┃
+#  ┣━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┫
+#  ┃ Copyright (c) 2017, the Perspective Authors.                              ┃
+#  ┃ ╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌ ┃
+#  ┃ This file is part of the Perspective library, distributed under the terms ┃
+#  ┃ of the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0). ┃
+#  ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
+
+import pandas as pd
+import numpy as np
+from perspective import PerspectiveError
+from datetime import date, datetime
+from pytest import approx, mark, raises
+
+import perspective as psp
+
+client = psp.Server().new_local_client()
+Table = client.table
+
+
+def date_timestamp(date):
+    return int(datetime.combine(date, datetime.min.time()).timestamp()) * 1000
+
+
+def compare_delta(received, expected):
+    """Compare an arrow-serialized row delta by constructing a Table."""
+    tbl = Table(received)
+    assert tbl.view().to_columns() == expected
+
+
+class TestView(object):
+    def test_view_zero(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view()
+        dimms = view.dimensions()
+        assert dimms["num_view_rows"] == 2
+        assert dimms["num_view_columns"] == 2
+        assert view.schema() == {"a": "integer", "b": "integer"}
+        assert view.to_records() == data
+
+    def test_view_one(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"])
+        dimms = view.dimensions()
+        assert dimms["num_view_rows"] == 3
+        assert dimms["num_view_columns"] == 2
+        assert view.schema() == {"a": "integer", "b": "integer"}
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "a": 4, "b": 6},
+            {"__ROW_PATH__": [1], "a": 1, "b": 2},
+            {"__ROW_PATH__": [3], "a": 3, "b": 4},
+        ]
+
+    def test_view_two(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"], split_by=["b"])
+        dimms = view.dimensions()
+        assert dimms["num_view_rows"] == 3
+        assert dimms["num_view_columns"] == 4
+        assert view.schema() == {"a": "integer", "b": "integer"}
+        assert view.to_records() == [
+            {"2|a": 1, "2|b": 2, "4|a": 3, "4|b": 4, "__ROW_PATH__": []},
+            {"2|a": 1, "2|b": 2, "4|a": None, "4|b": None, "__ROW_PATH__": [1]},
+            {"2|a": None, "2|b": None, "4|a": 3, "4|b": 4, "__ROW_PATH__": [3]},
+        ]
+
+    def test_view_two_column_only(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(split_by=["b"])
+        dimms = view.dimensions()
+        assert dimms["num_view_rows"] == 2
+        assert dimms["num_view_columns"] == 4
+        assert view.schema() == {"a": "integer", "b": "integer"}
+        assert view.to_records() == [
+            {"2|a": 1, "2|b": 2, "4|a": None, "4|b": None},
+            {"2|a": None, "2|b": None, "4|a": 3, "4|b": 4},
+        ]
+
+    # column path
+
+    def test_view_column_path_zero(self):
+        data = {"a": [1, 2, 3], "b": [1.5, 2.5, 3.5]}
+        tbl = Table(data)
+        view = tbl.view()
+        paths = view.column_paths()
+        assert paths == ["a", "b"]
+
+    def test_view_column_path_zero_schema(self):
+        data = {"a": "integer", "b": "float"}
+        tbl = Table(data)
+        view = tbl.view()
+        paths = view.column_paths()
+        assert paths == ["a", "b"]
+
+    def test_view_column_path_zero_hidden(self):
+        data = {"a": [1, 2, 3], "b": [1.5, 2.5, 3.5]}
+        tbl = Table(data)
+        view = tbl.view(columns=["b"])
+        paths = view.column_paths()
+        assert paths == ["b"]
+
+    def test_view_column_path_zero_respects_order(self):
+        data = {"a": [1, 2, 3], "b": [1.5, 2.5, 3.5]}
+        tbl = Table(data)
+        view = tbl.view(columns=["b", "a"])
+        paths = view.column_paths()
+        assert paths == ["b", "a"]
+
+    def test_view_column_path_one(self):
+        data = {"a": [1, 2, 3], "b": [1.5, 2.5, 3.5]}
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"])
+        paths = view.column_paths()
+        assert paths == ["a", "b"]
+
+    def test_view_column_path_one_numeric_names(self):
+        data = {"a": [1, 2, 3], "b": [1.5, 2.5, 3.5], "1234": [5, 6, 7]}
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"], columns=["b", "1234", "a"])
+        paths = view.column_paths()
+        assert paths == ["b", "1234", "a"]
+
+    def test_view_column_path_two(self):
+        data = {"a": [1, 2, 3], "b": [1.5, 2.5, 3.5]}
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"], split_by=["b"])
+        paths = view.column_paths()
+        assert paths == [
+            "1.5|a",
+            "1.5|b",
+            "2.5|a",
+            "2.5|b",
+            "3.5|a",
+            "3.5|b",
+        ]
+
+    def test_view_column_path_two_column_only(self):
+        data = {"a": [1, 2, 3], "b": [1.5, 2.5, 3.5]}
+        tbl = Table(data)
+        view = tbl.view(split_by=["b"])
+        paths = view.column_paths()
+        assert paths == ["1.5|a", "1.5|b", "2.5|a", "2.5|b", "3.5|a", "3.5|b"]
+
+    def test_view_column_path_hidden_sort(self):
+        data = {"a": [1, 2, 3], "b": [1.5, 2.5, 3.5], "c": [3, 2, 1]}
+        tbl = Table(data)
+        view = tbl.view(columns=["a", "b"], sort=[["c", "desc"]])
+        paths = view.column_paths()
+        assert paths == ["a", "b"]
+
+    def test_view_column_path_hidden_col_sort(self):
+        data = {"a": [1, 2, 3], "b": [1.5, 2.5, 3.5], "c": [3, 2, 1]}
+        tbl = Table(data)
+        view = tbl.view(split_by=["a"], columns=["a", "b"], sort=[["c", "col desc"]])
+        paths = view.column_paths()
+        assert paths == ["1|a", "1|b", "2|a", "2|b", "3|a", "3|b"]
+
+    def test_view_column_path_pivot_by_bool(self):
+        data = {"a": [1, 2, 3], "b": [True, False, True], "c": [3, 2, 1]}
+        tbl = Table(data)
+        view = tbl.view(split_by=["b"], columns=["a", "b", "c"])
+        paths = view.column_paths()
+        assert paths == ["false|a", "false|b", "false|c", "true|a", "true|b", "true|c"]
+
+    # schema correctness
+
+    def test_string_view_schema(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view()
+        assert view.schema() == {"a": "integer", "b": "integer"}
+
+    def test_zero_view_schema(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view()
+        assert view.schema() == {"a": "integer", "b": "integer"}
+
+    def test_one_view_schema(self):
+        data = [{"a": "abc", "b": 2}, {"a": "abc", "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"], aggregates={"a": "distinct count"})
+        assert view.schema() == {"a": "integer", "b": "integer"}
+
+    def test_two_view_schema(self):
+        data = [{"a": "abc", "b": "def"}, {"a": "abc", "b": "def"}]
+        tbl = Table(data)
+        view = tbl.view(
+            group_by=["a"], split_by=["b"], aggregates={"a": "count", "b": "count"}
+        )
+        assert view.schema() == {"a": "integer", "b": "integer"}
+
+    # aggregates and column specification
+
+    def test_view_no_columns(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(columns=[])
+        assert view.dimensions()["num_view_columns"] == 0
+        assert view.to_records() == []
+
+    def test_view_no_columns_pivoted(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"], columns=[])
+        assert view.dimensions()["num_view_columns"] == 0
+        assert view.to_records() == [
+            {"__ROW_PATH__": []},
+            {"__ROW_PATH__": [1]},
+            {"__ROW_PATH__": [3]},
+        ]
+
+    def test_view_specific_column(self):
+        data = [{"a": 1, "b": 2, "c": 3, "d": 4}, {"a": 3, "b": 4, "c": 5, "d": 6}]
+        tbl = Table(data)
+        view = tbl.view(columns=["a", "c", "d"])
+        assert view.dimensions()["num_view_columns"] == 3
+        assert view.to_records() == [{"a": 1, "c": 3, "d": 4}, {"a": 3, "c": 5, "d": 6}]
+
+    def test_view_column_order(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(columns=["b", "a"])
+        assert view.to_records() == [{"b": 2, "a": 1}, {"b": 4, "a": 3}]
+
+    def test_view_dataframe_column_order(self):
+        table = Table(
+            pd.DataFrame(
+                {
+                    "0.1": [5, 6, 7, 8],
+                    "-0.05": [5, 6, 7, 8],
+                    "0.0": [1, 2, 3, 4],
+                    "-0.1": [1, 2, 3, 4],
+                    "str": ["a", "b", "c", "d"],
+                }
+            )
+        )
+        view = table.view(columns=["-0.1", "-0.05", "0.0", "0.1"], group_by=["str"])
+        assert view.column_paths() == ["-0.1", "-0.05", "0.0", "0.1"]
+
+    def test_view_aggregate_order_with_columns(self):
+        """If `columns` is provided, order is always guaranteed."""
+        data = [{"a": 1, "b": 2, "c": 3, "d": 4}, {"a": 3, "b": 4, "c": 5, "d": 6}]
+        tbl = Table(data)
+        view = tbl.view(
+            group_by=["a"],
+            columns=["a", "b", "c", "d"],
+            aggregates={"d": "avg", "c": "avg", "b": "last", "a": "last"},
+        )
+
+        order = ["a", "b", "c", "d"]
+        assert view.column_paths() == order
+
+    def test_view_df_aggregate_order_with_columns(self):
+        """If `columns` is provided, order is always guaranteed."""
+        data = pd.DataFrame(
+            {"a": [1, 2, 3], "b": [2, 3, 4], "c": [3, 4, 5], "d": [4, 5, 6]},
+            columns=["d", "a", "c", "b"],
+        )
+        tbl = Table(data)
+        view = tbl.view(
+            group_by=["a"],
+            aggregates={"d": "avg", "c": "avg", "b": "last", "a": "last"},
+        )
+
+        order = ["index", "d", "a", "c", "b"]
+        assert view.column_paths() == order
+
+    def test_view_aggregates_with_no_columns(self):
+        data = [{"a": 1, "b": 2, "c": 3, "d": 4}, {"a": 3, "b": 4, "c": 5, "d": 6}]
+        tbl = Table(data)
+        view = tbl.view(
+            group_by=["a"], aggregates={"c": "avg", "a": "last"}, columns=[]
+        )
+        assert view.column_paths() == []
+        assert view.to_records() == [
+            {"__ROW_PATH__": []},
+            {"__ROW_PATH__": [1]},
+            {"__ROW_PATH__": [3]},
+        ]
+
+    def test_view_aggregates_default_column_order(self):
+        """Order of columns are entirely determined by the `columns` kwarg. If
+        it is not provided, order of columns is default based on the order
+        of table.columns()."""
+        data = [{"a": 1, "b": 2, "c": 3, "d": 4}, {"a": 3, "b": 4, "c": 5, "d": 6}]
+        tbl = Table(data)
+        cols = tbl.columns()
+        view = tbl.view(group_by=["a"], aggregates={"c": "avg", "a": "last"})
+
+        assert view.column_paths() == cols
+
+        # check that default aggregates have been applied
+        result = view.to_columns()
+        assert result["b"] == [6, 2, 4]
+        assert result["d"] == [10, 4, 6]
+
+        # and that specified aggregates are applied
+        assert result["a"] == [3, 1, 3]
+        assert result["c"] == [4, 3, 5]
+
+    # row and split by paths
+    def test_view_group_by_datetime_row_paths_are_same_as_data(self, util):
+        """Tests row paths for datetimes in UTC. Timezone-related tests are
+        in the `test_table_datetime` file."""
+        data = {"a": [datetime(2019, 7, 11, 12, 30)], "b": [1]}
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"])
+        data = view.to_columns()
+
+        for rp in data["__ROW_PATH__"]:
+            if len(rp) > 0:
+                assert rp[0] == util.to_timestamp(datetime(2019, 7, 11, 12, 30))
+
+        assert tbl.view().to_columns() == {
+            "a": [util.to_timestamp(datetime(2019, 7, 11, 12, 30))],
+            "b": [1],
+        }
+
+    def test_view_split_by_datetime_names_utc(self):
+        """Tests column paths for datetimes in UTC. Timezone-related tests are
+        in the `test_table_datetime` file."""
+        data = {"a": [datetime(2019, 7, 11, 12, 30)], "b": [1]}
+        tbl = Table(data)
+        view = tbl.view(split_by=["a"])
+        cols = view.column_paths()
+        assert cols == ["2019-07-11 12:30:00.000|a", "2019-07-11 12:30:00.000|b"]
+
+    # TODO: time slightly off! thinks its NYE 1969
+    @mark.skip  # We do not support python datetimes.
+    def test_view_split_by_datetime_names_min(self):
+        """Tests column paths for datetimes in UTC. Timezone-related tests are
+        in the `test_table_datetime` file."""
+        import os
+
+        os.environ["TZ"] = "UTC"
+        data = {"a": [datetime.min], "b": [1]}
+        tbl = Table({"a": "datetime", "b": "integer"})
+        tbl.update(data)
+        view = tbl.view(split_by=["a"])
+        cols = view.column_paths()
+        assert cols == ["1970-01-01 00:00:00.000|a", "1970-01-01 00:00:00.000|b"]
+
+    @mark.skip  # We dont support python datetimes.
+    def test_view_split_by_datetime_names_max(self):
+        """Tests column paths for datetimes in UTC. Timezone-related tests are
+        in the `test_table_datetime` file."""
+        data = {"a": [datetime.max], "b": [1]}
+        tbl = Table(data)
+        view = tbl.view(split_by=["a"])
+        cols = view.column_paths()
+        assert cols == ["10000-01-01 00:00:00.000|a", "10000-01-01 00:00:00.000|b"]
+
+    # aggregate
+
+    def test_view_aggregate_int(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(aggregates={"a": "avg"}, group_by=["a"])
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "a": 2.0, "b": 6},
+            {"__ROW_PATH__": [1], "a": 1.0, "b": 2},
+            {"__ROW_PATH__": [3], "a": 3.0, "b": 4},
+        ]
+
+    def test_view_aggregate_str(self):
+        data = [{"a": "abc", "b": 2}, {"a": "def", "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(aggregates={"a": "count"}, group_by=["a"])
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "a": 2, "b": 6},
+            {"__ROW_PATH__": ["abc"], "a": 1, "b": 2},
+            {"__ROW_PATH__": ["def"], "a": 1, "b": 4},
+        ]
+
+    def test_view_aggregate_datetime(self, util):
+        data = [
+            {"a": datetime(2019, 10, 1, 11, 30)},
+            {"a": datetime(2019, 10, 1, 11, 30)},
+        ]
+        tbl = Table(data)
+        view = tbl.view(aggregates={"a": "distinct count"}, group_by=["a"])
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "a": 1},
+            {
+                "__ROW_PATH__": [util.to_timestamp(datetime(2019, 10, 1, 11, 30))],
+                "a": 1,
+            },
+        ]
+
+    def test_view_aggregate_datetime_leading_zeroes(self, util):
+        data = [
+            {"a": datetime(2019, 1, 1, 5, 5, 5)},
+            {"a": datetime(2019, 1, 1, 5, 5, 5)},
+        ]
+        tbl = Table(data)
+        view = tbl.view(aggregates={"a": "distinct count"}, group_by=["a"])
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "a": 1},
+            {
+                "__ROW_PATH__": [util.to_timestamp(datetime(2019, 1, 1, 5, 5, 5))],
+                "a": 1,
+            },
+        ]
+
+    def test_view_aggregate_mean(self):
+        data = [
+            {"a": "a", "x": 1, "y": 200},
+            {"a": "a", "x": 2, "y": 100},
+            {"a": "a", "x": 3, "y": None},
+        ]
+        tbl = Table(data)
+        view = tbl.view(aggregates={"y": "mean"}, group_by=["a"], columns=["y"])
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "y": 300 / 2},
+            {"__ROW_PATH__": ["a"], "y": 300 / 2},
+        ]
+
+    def test_view_aggregate_mean_from_schema(self):
+        data = [
+            {"a": "a", "x": 1, "y": 200},
+            {"a": "a", "x": 2, "y": 100},
+            {"a": "a", "x": 3, "y": None},
+        ]
+        tbl = Table({"a": "string", "x": "integer", "y": "float"})
+        view = tbl.view(aggregates={"y": "mean"}, group_by=["a"], columns=["y"])
+        tbl.update(data)
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "y": 300 / 2},
+            {"__ROW_PATH__": ["a"], "y": 300 / 2},
+        ]
+
+    def test_view_aggregate_weighted_mean(self):
+        data = [
+            {"a": "a", "x": 1, "y": 200},
+            {"a": "a", "x": 2, "y": 100},
+            {"a": "a", "x": 3, "y": None},
+        ]
+        tbl = Table(data)
+        view = tbl.view(
+            aggregates={"y": ("weighted mean", ["x"])},
+            group_by=["a"],
+            columns=["y"],
+        )
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "y": (1.0 * 200 + 2 * 100) / (1.0 + 2)},
+            {"__ROW_PATH__": ["a"], "y": (1.0 * 200 + 2 * 100) / (1.0 + 2)},
+        ]
+
+    def test_view_aggregate_weighted_mean_with_negative_weights(self):
+        data = [
+            {"a": "a", "x": 1, "y": 200},
+            {"a": "a", "x": -2, "y": 100},
+            {"a": "a", "x": 3, "y": None},
+        ]
+        tbl = Table(data)
+        view = tbl.view(
+            aggregates={"y": ("weighted mean", ["x"])},
+            group_by=["a"],
+            columns=["y"],
+        )
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "y": (1 * 200 + (-2) * 100) / (1 - 2)},
+            {"__ROW_PATH__": ["a"], "y": (1 * 200 + (-2) * 100) / (1 - 2)},
+        ]
+
+    def test_view_variance(self):
+        data = {"x": list(np.random.rand(10)), "y": ["a" for _ in range(10)]}
+
+        table = Table(data)
+        view = table.view(aggregates={"x": "var"}, group_by=["y"])
+
+        result = view.to_columns()
+        expected = np.var(data["x"])
+
+        assert result["x"] == approx([expected, expected])
+
+    def test_view_variance_multi(self):
+        data = {
+            "a": [
+                91.96,
+                258.576,
+                29.6,
+                243.16,
+                36.24,
+                25.248,
+                79.99,
+                206.1,
+                31.5,
+                55.6,
+            ],
+            "b": [1 if i % 2 == 0 else 0 for i in range(10)],
+        }
+        table = Table(data)
+        view = table.view(aggregates={"a": "var"}, group_by=["b"])
+
+        result = view.to_columns()
+        expected_total = np.var(data["a"])
+        expected_zero = np.var(
+            [data["a"][1], data["a"][3], data["a"][5], data["a"][7], data["a"][9]]
+        )
+        expected_one = np.var(
+            [data["a"][0], data["a"][2], data["a"][4], data["a"][6], data["a"][8]]
+        )
+
+        assert result["a"] == approx([expected_total, expected_zero, expected_one])
+
+    def test_view_variance_update_none(self):
+        data = {"a": [0.1, 0.5, None, 0.8], "b": [0, 1, 0, 1], "c": [1, 2, 3, 4]}
+        table = Table(data, index="c")
+        view = table.view(columns=["a"], group_by=["b"], aggregates={"a": "var"})
+        result = view.to_columns()
+        assert result["a"][0] == approx(np.var([0.1, 0.5, 0.8]))
+        assert result["a"][1] is None
+        assert result["a"][2] == approx(np.var([0.5, 0.8]))
+
+        table.update({"a": [0.3], "c": [3]})
+
+        result = view.to_columns()
+        assert result["a"] == approx(
+            [np.var([0.1, 0.5, 0.3, 0.8]), np.var([0.1, 0.3]), np.var([0.5, 0.8])]
+        )
+
+        table.update({"a": [None], "c": [1]})
+
+        result = view.to_columns()
+        assert result["a"][0] == approx(np.var([0.5, 0.3, 0.8]))
+        assert result["a"][1] is None
+        assert result["a"][2] == approx(np.var([0.5, 0.8]))
+
+    def test_view_variance_multi_update(self):
+        data = {
+            "a": [
+                91.96,
+                258.576,
+                29.6,
+                243.16,
+                36.24,
+                25.248,
+                79.99,
+                206.1,
+                31.5,
+                55.6,
+            ],
+            "b": [1 if i % 2 == 0 else 0 for i in range(10)],
+        }
+        table = Table(data)
+        view = table.view(aggregates={"a": "var"}, group_by=["b"])
+
+        result = view.to_columns()
+        expected_total = data["a"]
+        expected_zero = [
+            data["a"][1],
+            data["a"][3],
+            data["a"][5],
+            data["a"][7],
+            data["a"][9],
+        ]
+        expected_one = [
+            data["a"][0],
+            data["a"][2],
+            data["a"][4],
+            data["a"][6],
+            data["a"][8],
+        ]
+
+        assert result["a"] == approx(
+            [np.var(expected_total), np.var(expected_zero), np.var(expected_one)]
+        )
+
+        # 2 here should result in null var because the group size is 1
+        update_data = {"a": [15.12, 9.102, 0.99, 12.8], "b": [1, 0, 1, 2]}
+        table.update(update_data)
+
+        result = view.to_columns()
+        expected_total += update_data["a"]
+        expected_zero += [update_data["a"][1]]
+        expected_one += [update_data["a"][0], update_data["a"][2]]
+
+        assert result["__ROW_PATH__"] == [[], [0], [1], [2]]
+        assert result["a"][:-1] == approx(
+            [np.var(expected_total), np.var(expected_zero), np.var(expected_one)]
+        )
+        assert result["a"][-1] is None
+
+    def test_view_variance_multi_update_delta(self):
+        data = {
+            "a": [
+                91.96,
+                258.576,
+                29.6,
+                243.16,
+                36.24,
+                25.248,
+                79.99,
+                206.1,
+                31.5,
+                55.6,
+            ],
+            "b": [1 if i % 2 == 0 else 0 for i in range(10)],
+        }
+        table = Table(data)
+        view = table.view(aggregates={"a": "var"}, group_by=["b"])
+
+        result = view.to_columns()
+        expected_total = data["a"]
+        expected_zero = [
+            data["a"][1],
+            data["a"][3],
+            data["a"][5],
+            data["a"][7],
+            data["a"][9],
+        ]
+        expected_one = [
+            data["a"][0],
+            data["a"][2],
+            data["a"][4],
+            data["a"][6],
+            data["a"][8],
+        ]
+
+        assert result["a"] == approx(
+            [np.var(expected_total), np.var(expected_zero), np.var(expected_one)]
+        )
+
+        # 2 here should result in null var because the group size is 1
+        update_data = {"a": [15.12, 9.102, 0.99, 12.8], "b": [1, 0, 1, 2]}
+
+        def cb1(port_id, delta):
+            table2 = Table(delta)
+            view2 = table2.view()
+            result = view2.to_columns()
+
+            flat_view = table.view()
+            flat_data = flat_view.to_columns()
+            result = view.to_columns()
+
+            expected_total = flat_data["a"]
+            expected_zero = []
+            expected_one = []
+
+            for i, num in enumerate(expected_total):
+                if flat_data["b"][i] == 1:
+                    expected_one.append(num)
+                elif flat_data["b"][i] == 0:
+                    expected_zero.append(num)
+
+            assert result["a"][0] == approx(np.var(expected_total))
+            assert result["a"][1] == approx(np.var(expected_zero))
+            assert result["a"][2] == approx(np.var(expected_one))
+            assert result["a"][3] is None
+
+        view.on_update(cb1, mode="row")
+
+        table.update(update_data)
+
+    def test_view_variance_multi_update_indexed(self):
+        data = {
+            "a": [
+                91.96,
+                258.576,
+                29.6,
+                243.16,
+                36.24,
+                25.248,
+                79.99,
+                206.1,
+                31.5,
+                55.6,
+            ],
+            "b": [1 if i % 2 == 0 else 0 for i in range(10)],
+            "c": [i for i in range(10)],
+        }
+        table = Table(data, index="c")
+        view = table.view(aggregates={"a": "var"}, group_by=["b"])
+
+        result = view.to_columns()
+        expected_total = data["a"]
+        expected_zero = [
+            data["a"][1],
+            data["a"][3],
+            data["a"][5],
+            data["a"][7],
+            data["a"][9],
+        ]
+        expected_one = [
+            data["a"][0],
+            data["a"][2],
+            data["a"][4],
+            data["a"][6],
+            data["a"][8],
+        ]
+
+        assert result["a"] == approx(
+            [np.var(expected_total), np.var(expected_zero), np.var(expected_one)]
+        )
+
+        # "b" = 2 here should result in null var because the group size is 1
+        update_data = {
+            "a": [15.12, 9.102, 0.99, 12.8],
+            "b": [1, 0, 1, 2],
+            "c": [1, 5, 2, 7],
+        }
+
+        table.update(update_data)
+
+        result = view.to_columns()
+
+        view2 = table.view()
+        flat_data = view2.to_columns()
+        expected_total = flat_data["a"]
+
+        expected_zero = []
+        expected_one = []
+
+        for i, val in enumerate(flat_data["a"]):
+            if flat_data["b"][i] == 1:
+                expected_one.append(val)
+            elif flat_data["b"][i] == 0:
+                expected_zero.append(val)
+
+        assert result["__ROW_PATH__"] == [[], [0], [1], [2]]
+        assert result["a"][:-1] == approx(
+            [np.var(expected_total), np.var(expected_zero), np.var(expected_one)]
+        )
+        assert result["a"][-1] is None
+
+    def test_view_variance_multi_update_indexed_delta(self):
+        data = {
+            "a": [
+                91.96,
+                258.576,
+                29.6,
+                243.16,
+                36.24,
+                25.248,
+                79.99,
+                206.1,
+                31.5,
+                55.6,
+            ],
+            "b": [1 if i % 2 == 0 else 0 for i in range(10)],
+            "c": [i for i in range(10)],
+        }
+        table = Table(data, index="c")
+        view = table.view(
+            aggregates={"a": "var", "b": "last", "c": "last"}, group_by=["b"]
+        )
+
+        result = view.to_columns()
+        expected_total = data["a"]
+        expected_zero = [
+            data["a"][1],
+            data["a"][3],
+            data["a"][5],
+            data["a"][7],
+            data["a"][9],
+        ]
+        expected_one = [
+            data["a"][0],
+            data["a"][2],
+            data["a"][4],
+            data["a"][6],
+            data["a"][8],
+        ]
+
+        assert result["a"] == approx(
+            [np.var(expected_total), np.var(expected_zero), np.var(expected_one)]
+        )
+
+        # 2 here should result in null var because the group size is 1
+        update_data = {
+            "a": [15.12, 9.102, 0.99, 12.8],
+            "b": [1, 0, 1, 2],
+            "c": [0, 4, 1, 6],
+        }
+
+        def cb1(port_id, delta):
+            table2 = Table(delta)
+            view2 = table2.view()
+            result = view2.to_columns()
+
+            flat_view = table.view()
+            flat_result = flat_view.to_columns()
+
+            new_a = flat_result["a"]
+            b = flat_result["b"]
+            expected_zero = []
+            expected_one = []
+
+            for i, num in enumerate(new_a):
+                if b[i] == 0:
+                    expected_zero.append(num)
+                elif b[i] == 1:
+                    expected_one.append(num)
+
+            assert result["a"][0] == approx(np.var(new_a))
+            assert result["a"][1] == approx(np.var(expected_zero))
+            assert result["a"][2] == approx(np.var(expected_one))
+            assert result["a"][3] is None
+            assert result["b"] == [2, 0, 1, 2]
+            assert result["c"] == [6, 9, 8, 6]
+
+        view.on_update(cb1, mode="row")
+
+        table.update(update_data)
+
+    def test_view_variance_less_than_two(self):
+        data = {"a": list(np.random.rand(10)), "b": [i for i in range(10)]}
+
+        table = Table(data)
+        view = table.view(aggregates={"a": "var"}, group_by=["b"])
+
+        result = view.to_columns()
+        assert result["a"][0] == approx(np.var(data["a"]))
+        assert result["a"][1:] == [None] * 10
+
+    def test_view_variance_normal_distribution(self):
+        data = {"a": list(np.random.standard_normal(100)), "b": [1] * 100}
+
+        table = Table(data)
+        view = table.view(aggregates={"a": "var"}, group_by=["b"])
+
+        result = view.to_columns()
+        assert result["a"] == approx([np.var(data["a"]), np.var(data["a"])])
+
+    def test_view_standard_deviation(self):
+        data = {"x": list(np.random.rand(10)), "y": ["a" for _ in range(10)]}
+
+        table = Table(data)
+        view = table.view(aggregates={"x": "stddev"}, group_by=["y"])
+
+        result = view.to_columns()
+        expected = np.std(data["x"])
+
+        assert result["x"] == approx([expected, expected])
+
+    def test_view_standard_deviation_multi(self):
+        data = {
+            "a": [
+                91.96,
+                258.576,
+                29.6,
+                243.16,
+                36.24,
+                25.248,
+                79.99,
+                206.1,
+                31.5,
+                55.6,
+            ],
+            "b": [1 if i % 2 == 0 else 0 for i in range(10)],
+        }
+        table = Table(data)
+        view = table.view(aggregates={"a": "stddev"}, group_by=["b"])
+
+        result = view.to_columns()
+        expected_total = np.std(data["a"])
+        expected_zero = np.std(
+            [data["a"][1], data["a"][3], data["a"][5], data["a"][7], data["a"][9]]
+        )
+        expected_one = np.std(
+            [data["a"][0], data["a"][2], data["a"][4], data["a"][6], data["a"][8]]
+        )
+
+        assert result["a"] == approx([expected_total, expected_zero, expected_one])
+
+    def test_view_standard_deviation_update_none(self):
+        data = {"a": [0.1, 0.5, None, 0.8], "b": [0, 1, 0, 1], "c": [1, 2, 3, 4]}
+        table = Table(data, index="c")
+        view = table.view(columns=["a"], group_by=["b"], aggregates={"a": "stddev"})
+        result = view.to_columns()
+        assert result["a"][0] == approx(np.std([0.1, 0.5, 0.8]))
+        assert result["a"][1] is None
+        assert result["a"][2] == approx(np.std([0.5, 0.8]))
+
+        table.update({"a": [0.3], "c": [3]})
+
+        result = view.to_columns()
+        assert result["a"] == approx(
+            [np.std([0.1, 0.5, 0.3, 0.8]), np.std([0.1, 0.3]), np.std([0.5, 0.8])]
+        )
+
+        table.update({"a": [None], "c": [1]})
+
+        result = view.to_columns()
+        assert result["a"][0] == approx(np.std([0.5, 0.3, 0.8]))
+        assert result["a"][1] is None
+        assert result["a"][2] == approx(np.std([0.5, 0.8]))
+
+    def test_view_standard_deviation_multi_update(self):
+        data = {
+            "a": [
+                91.96,
+                258.576,
+                29.6,
+                243.16,
+                36.24,
+                25.248,
+                79.99,
+                206.1,
+                31.5,
+                55.6,
+            ],
+            "b": [1 if i % 2 == 0 else 0 for i in range(10)],
+        }
+        table = Table(data)
+        view = table.view(aggregates={"a": "stddev"}, group_by=["b"])
+
+        result = view.to_columns()
+        expected_total = data["a"]
+        expected_zero = [
+            data["a"][1],
+            data["a"][3],
+            data["a"][5],
+            data["a"][7],
+            data["a"][9],
+        ]
+        expected_one = [
+            data["a"][0],
+            data["a"][2],
+            data["a"][4],
+            data["a"][6],
+            data["a"][8],
+        ]
+
+        assert result["a"] == approx(
+            [np.std(expected_total), np.std(expected_zero), np.std(expected_one)]
+        )
+
+        # 2 here should result in null stddev because the group size is 1
+        update_data = {"a": [15.12, 9.102, 0.99, 12.8], "b": [1, 0, 1, 2]}
+        table.update(update_data)
+
+        result = view.to_columns()
+        expected_total += update_data["a"]
+        expected_zero += [update_data["a"][1]]
+        expected_one += [update_data["a"][0], update_data["a"][2]]
+
+        assert result["__ROW_PATH__"] == [[], [0], [1], [2]]
+        assert result["a"][:-1] == approx(
+            [np.std(expected_total), np.std(expected_zero), np.std(expected_one)]
+        )
+        assert result["a"][-1] is None
+
+    def test_view_standard_deviation_multi_update_delta(self):
+        data = {
+            "a": [
+                91.96,
+                258.576,
+                29.6,
+                243.16,
+                36.24,
+                25.248,
+                79.99,
+                206.1,
+                31.5,
+                55.6,
+            ],
+            "b": [1 if i % 2 == 0 else 0 for i in range(10)],
+        }
+        table = Table(data)
+        view = table.view(aggregates={"a": "stddev"}, group_by=["b"])
+
+        result = view.to_columns()
+        expected_total = data["a"]
+        expected_zero = [
+            data["a"][1],
+            data["a"][3],
+            data["a"][5],
+            data["a"][7],
+            data["a"][9],
+        ]
+        expected_one = [
+            data["a"][0],
+            data["a"][2],
+            data["a"][4],
+            data["a"][6],
+            data["a"][8],
+        ]
+
+        assert result["a"] == approx(
+            [np.std(expected_total), np.std(expected_zero), np.std(expected_one)]
+        )
+
+        # 2 here should result in null stddev because the group size is 1
+        update_data = {"a": [15.12, 9.102, 0.99, 12.8], "b": [1, 0, 1, 2]}
+
+        def cb1(port_id, delta):
+            table2 = Table(delta)
+            view2 = table2.view()
+            result = view2.to_columns()
+
+            flat_view = table.view()
+            flat_data = flat_view.to_columns()
+            result = view.to_columns()
+
+            expected_total = flat_data["a"]
+            expected_zero = []
+            expected_one = []
+
+            for i, num in enumerate(expected_total):
+                if flat_data["b"][i] == 1:
+                    expected_one.append(num)
+                elif flat_data["b"][i] == 0:
+                    expected_zero.append(num)
+
+            assert result["a"][0] == approx(np.std(expected_total))
+            assert result["a"][1] == approx(np.std(expected_zero))
+            assert result["a"][2] == approx(np.std(expected_one))
+            assert result["a"][3] is None
+
+        view.on_update(cb1, mode="row")
+
+        table.update(update_data)
+
+    def test_view_standard_deviation_multi_update_indexed(self):
+        data = {
+            "a": [
+                91.96,
+                258.576,
+                29.6,
+                243.16,
+                36.24,
+                25.248,
+                79.99,
+                206.1,
+                31.5,
+                55.6,
+            ],
+            "b": [1 if i % 2 == 0 else 0 for i in range(10)],
+            "c": [i for i in range(10)],
+        }
+        table = Table(data, index="c")
+        view = table.view(aggregates={"a": "stddev"}, group_by=["b"])
+
+        result = view.to_columns()
+        expected_total = data["a"]
+        expected_zero = [
+            data["a"][1],
+            data["a"][3],
+            data["a"][5],
+            data["a"][7],
+            data["a"][9],
+        ]
+        expected_one = [
+            data["a"][0],
+            data["a"][2],
+            data["a"][4],
+            data["a"][6],
+            data["a"][8],
+        ]
+
+        assert result["a"] == approx(
+            [np.std(expected_total), np.std(expected_zero), np.std(expected_one)]
+        )
+
+        # "b" = 2 here should result in null stddev because the group size is 1
+        update_data = {
+            "a": [15.12, 9.102, 0.99, 12.8],
+            "b": [1, 0, 1, 2],
+            "c": [1, 5, 2, 7],
+        }
+
+        table.update(update_data)
+
+        result = view.to_columns()
+
+        view2 = table.view()
+        flat_data = view2.to_columns()
+        expected_total = flat_data["a"]
+
+        expected_zero = []
+        expected_one = []
+
+        for i, val in enumerate(flat_data["a"]):
+            if flat_data["b"][i] == 1:
+                expected_one.append(val)
+            elif flat_data["b"][i] == 0:
+                expected_zero.append(val)
+
+        assert result["__ROW_PATH__"] == [[], [0], [1], [2]]
+        assert result["a"][:-1] == approx(
+            [np.std(expected_total), np.std(expected_zero), np.std(expected_one)]
+        )
+        assert result["a"][-1] is None
+
+    def test_view_standard_deviation_multi_update_indexed_delta(self):
+        data = {
+            "a": [
+                91.96,
+                258.576,
+                29.6,
+                243.16,
+                36.24,
+                25.248,
+                79.99,
+                206.1,
+                31.5,
+                55.6,
+            ],
+            "b": [1 if i % 2 == 0 else 0 for i in range(10)],
+            "c": [i for i in range(10)],
+        }
+        table = Table(data, index="c")
+        view = table.view(
+            aggregates={"a": "stddev", "b": "last", "c": "last"}, group_by=["b"]
+        )
+
+        result = view.to_columns()
+        expected_total = data["a"]
+        expected_zero = [
+            data["a"][1],
+            data["a"][3],
+            data["a"][5],
+            data["a"][7],
+            data["a"][9],
+        ]
+        expected_one = [
+            data["a"][0],
+            data["a"][2],
+            data["a"][4],
+            data["a"][6],
+            data["a"][8],
+        ]
+
+        assert result["a"] == approx(
+            [np.std(expected_total), np.std(expected_zero), np.std(expected_one)]
+        )
+
+        # 2 here should result in null stddev because the group size is 1
+        update_data = {
+            "a": [15.12, 9.102, 0.99, 12.8],
+            "b": [1, 0, 1, 2],
+            "c": [0, 4, 1, 6],
+        }
+
+        def cb1(port_id, delta):
+            table2 = Table(delta)
+            view2 = table2.view()
+            result = view2.to_columns()
+
+            flat_view = table.view()
+            flat_result = flat_view.to_columns()
+
+            new_a = flat_result["a"]
+            b = flat_result["b"]
+            expected_zero = []
+            expected_one = []
+
+            for i, num in enumerate(new_a):
+                if b[i] == 0:
+                    expected_zero.append(num)
+                elif b[i] == 1:
+                    expected_one.append(num)
+
+            assert result["a"][0] == approx(np.std(new_a))
+            assert result["a"][1] == approx(np.std(expected_zero))
+            assert result["a"][2] == approx(np.std(expected_one))
+            assert result["a"][3] is None
+            assert result["b"] == [2, 0, 1, 2]
+            assert result["c"] == [6, 9, 8, 6]
+
+        view.on_update(cb1, mode="row")
+
+        table.update(update_data)
+
+    def test_view_standard_deviation_less_than_two(self):
+        data = {"a": list(np.random.rand(10)), "b": [i for i in range(10)]}
+
+        table = Table(data)
+        view = table.view(aggregates={"a": "stddev"}, group_by=["b"])
+
+        result = view.to_columns()
+        assert result["a"][0] == approx(np.std(data["a"]))
+        assert result["a"][1:] == [None] * 10
+
+    def test_view_standard_deviation_normal_distribution(self):
+        data = {"a": list(np.random.standard_normal(100)), "b": [1] * 100}
+
+        table = Table(data)
+        view = table.view(aggregates={"a": "stddev"}, group_by=["b"])
+
+        result = view.to_columns()
+        assert result["a"] == approx([np.std(data["a"]), np.std(data["a"])])
+
+    # sort
+
+    def test_view_sort_int(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(sort=[["a", "desc"]])
+        assert view.to_records() == [{"a": 3, "b": 4}, {"a": 1, "b": 2}]
+
+    def test_view_sort_float(self):
+        data = [{"a": 1.1, "b": 2}, {"a": 1.2, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(sort=[["a", "desc"]])
+        assert view.to_records() == [{"a": 1.2, "b": 4}, {"a": 1.1, "b": 2}]
+
+    def test_view_sort_string(self):
+        data = [{"a": "abc", "b": 2}, {"a": "def", "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(sort=[["a", "desc"]])
+        assert view.to_records() == [{"a": "def", "b": 4}, {"a": "abc", "b": 2}]
+
+    def test_view_sort_date(self, util):
+        data = [{"a": date(2019, 7, 11), "b": 2}, {"a": date(2019, 7, 12), "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(sort=[["a", "desc"]])
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 12)), "b": 4},
+            {"a": util.to_timestamp(datetime(2019, 7, 11)), "b": 2},
+        ]
+
+    def test_view_sort_datetime(self, util):
+        data = [
+            {"a": datetime(2019, 7, 11, 8, 15), "b": 2},
+            {"a": datetime(2019, 7, 11, 8, 16), "b": 4},
+        ]
+        tbl = Table(data)
+        view = tbl.view(sort=[["a", "desc"]])
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 11, 8, 16)), "b": 4},
+            {"a": util.to_timestamp(datetime(2019, 7, 11, 8, 15)), "b": 2},
+        ]
+
+    def test_view_sort_hidden(self):
+        data = [{"a": 1.1, "b": 2}, {"a": 1.2, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(sort=[["a", "desc"]], columns=["b"])
+        assert view.to_records() == [{"b": 4}, {"b": 2}]
+
+    def test_view_sort_avg_nan(self):
+        data = {
+            "w": [3.5, 4.5, None, None, None, None, 1.5, 2.5],
+            "x": [1, 2, 3, 4, 4, 3, 2, 1],
+            "y": ["a", "b", "c", "d", "e", "f", "g", "h"],
+        }
+        tbl = Table(data)
+        view = tbl.view(
+            columns=["x", "w"],
+            group_by=["y"],
+            sort=[["w", "asc"]],
+            aggregates={"w": "avg", "x": "unique"},
+        )
+        assert view.to_columns() == {
+            "__ROW_PATH__": [
+                [],
+                ["c"],
+                ["d"],
+                ["e"],
+                ["f"],
+                ["g"],
+                ["h"],
+                ["a"],
+                ["b"],
+            ],
+            "w": [3, None, None, None, None, 1.5, 2.5, 3.5, 4.5],
+            "x": [None, 3, 4, 4, 3, 2, 1, 1, 2],
+        }
+
+    def test_view_sort_sum_nan(self):
+        data = {
+            "w": [3.5, 4.5, None, None, None, None, 1.5, 2.5],
+            "x": [1, 2, 3, 4, 4, 3, 2, 1],
+            "y": ["a", "b", "c", "d", "e", "f", "g", "h"],
+        }
+        tbl = Table(data)
+        view = tbl.view(
+            columns=["x", "w"],
+            group_by=["y"],
+            sort=[["w", "asc"]],
+            aggregates={"w": "sum", "x": "unique"},
+        )
+        assert view.to_columns() == {
+            "__ROW_PATH__": [
+                [],
+                ["c"],
+                ["d"],
+                ["e"],
+                ["f"],
+                ["g"],
+                ["h"],
+                ["a"],
+                ["b"],
+            ],
+            "w": [12, 0, 0, 0, 0, 1.5, 2.5, 3.5, 4.5],
+            "x": [None, 3, 4, 4, 3, 2, 1, 1, 2],
+        }
+
+    def test_view_sort_unique_nan(self):
+        data = {
+            "w": [3.5, 4.5, None, None, None, None, 1.5, 2.5],
+            "x": [1, 2, 3, 4, 4, 3, 2, 1],
+            "y": ["a", "b", "c", "d", "e", "f", "g", "h"],
+        }
+        tbl = Table(data)
+        view = tbl.view(
+            columns=["x", "w"],
+            group_by=["y"],
+            sort=[["w", "asc"]],
+            aggregates={"w": "unique", "x": "unique"},
+        )
+        assert view.to_columns() == {
+            "__ROW_PATH__": [
+                [],
+                ["c"],
+                ["d"],
+                ["e"],
+                ["f"],
+                ["g"],
+                ["h"],
+                ["a"],
+                ["b"],
+            ],
+            "w": [None, None, None, None, None, 1.5, 2.5, 3.5, 4.5],
+            "x": [None, 3, 4, 4, 3, 2, 1, 1, 2],
+        }
+
+    # filter
+
+    def test_view_filter_int_eq(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "==", 1]])
+        assert view.to_records() == [{"a": 1, "b": 2}]
+
+    def test_view_filter_int_neq(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "!=", 1]])
+        assert view.to_records() == [{"a": 3, "b": 4}]
+
+    def test_view_filter_int_gt(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", ">", 1]])
+        assert view.to_records() == [{"a": 3, "b": 4}]
+
+    def test_view_filter_int_lt(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "<", 3]])
+        assert view.to_records() == [{"a": 1, "b": 2}]
+
+    def test_view_filter_float_eq(self):
+        data = [{"a": 1.1, "b": 2}, {"a": 1.2, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "==", 1.2]])
+        assert view.to_records() == [{"a": 1.2, "b": 4}]
+
+    def test_view_filter_float_neq(self):
+        data = [{"a": 1.1, "b": 2}, {"a": 1.2, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "!=", 1.2]])
+        assert view.to_records() == [{"a": 1.1, "b": 2}]
+
+    def test_view_filter_string_eq(self):
+        data = [{"a": "abc", "b": 2}, {"a": "def", "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "==", "def"]])
+        assert view.to_records() == [{"a": "def", "b": 4}]
+
+    def test_view_filter_string_neq(self):
+        data = [{"a": "abc", "b": 2}, {"a": "def", "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "!=", "def"]])
+        assert view.to_records() == [{"a": "abc", "b": 2}]
+
+    def test_view_filter_string_gt(self):
+        data = [{"a": "abc", "b": 2}, {"a": "def", "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", ">", "abc"]])
+        assert view.to_records() == [{"a": "def", "b": 4}]
+
+    def test_view_filter_string_lt(self):
+        data = [{"a": "abc", "b": 2}, {"a": "def", "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "<", "def"]])
+        assert view.to_records() == [{"a": "abc", "b": 2}]
+
+    # @mark.skip # We do not support using `datetime.date` in operators.
+    def test_view_filter_date_eq(self, util):
+        data = [{"a": date(2019, 7, 11), "b": 2}, {"a": date(2019, 7, 12), "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "==", str(date(2019, 7, 12))]])
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 12)), "b": 4}
+        ]
+
+    # @mark.skip # We do not support using `datetime.date` in operators.
+    def test_view_filter_date_neq(self, util):
+        data = [{"a": date(2019, 7, 11), "b": 2}, {"a": date(2019, 7, 12), "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "!=", str(date(2019, 7, 12))]])
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 11)), "b": 2}
+        ]
+
+    def test_view_filter_date_str_eq(self, util):
+        data = [{"a": date(2019, 7, 11), "b": 2}, {"a": date(2019, 7, 12), "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "==", "2019/7/12"]])
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 12)), "b": 4}
+        ]
+
+    def test_view_filter_date_str_neq(self, util):
+        data = [{"a": date(2019, 7, 11), "b": 2}, {"a": date(2019, 7, 12), "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "!=", "2019/7/12"]])
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 11)), "b": 2}
+        ]
+
+    @mark.skip  # We do not support using `datetime.datetime` in operators
+    def test_view_filter_datetime_eq(self, util):
+        data = [
+            {"a": datetime(2019, 7, 11, 8, 15), "b": 2},
+            {"a": datetime(2019, 7, 11, 8, 16), "b": 4},
+        ]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "==", datetime(2019, 7, 11, 8, 15)]])
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 11, () * 1000)), "b": 2}
+        ]
+
+    @mark.skip  # We do not support using `datetime.datetime` in operators
+    def test_view_filter_datetime_neq(self, util):
+        data = [
+            {"a": datetime(2019, 7, 11, 8, 15), "b": 2},
+            {"a": datetime(2019, 7, 11, 8, 16), "b": 4},
+        ]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "!=", datetime(2019, 7, 11, 8, 15)]])
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 11, () * 1000)), "b": 4}
+        ]
+
+    @mark.skip  # We do not support numpy anymore
+    def test_view_filter_datetime_np_eq(self, util):
+        data = [
+            {"a": datetime(2019, 7, 11, 8, 15), "b": 2},
+            {"a": datetime(2019, 7, 11, 8, 16), "b": 4},
+        ]
+        tbl = Table(data)
+        view = tbl.view(
+            filter=[["a", "==", np.datetime64(datetime(2019, 7, 11, 8, 15))]]
+        )
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 11, () * 1000)), "b": 2}
+        ]
+
+    @mark.skip  # We do not support numpy anymore.
+    def test_view_filter_datetime_np_neq(self, util):
+        data = [
+            {"a": datetime(2019, 7, 11, 8, 15), "b": 2},
+            {"a": datetime(2019, 7, 11, 8, 16), "b": 4},
+        ]
+        tbl = Table(data)
+        view = tbl.view(
+            filter=[["a", "!=", np.datetime64(datetime(2019, 7, 11, 8, 15))]]
+        )
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 11, () * 1000)), "b": 4}
+        ]
+
+    def test_view_filter_datetime_as_number_eq(self, util):
+        data = [
+            {"a": datetime(2019, 7, 11, 8, 15).timestamp(), "b": 2},
+            {"a": datetime(2019, 7, 11, 8, 16).timestamp(), "b": 4},
+        ]
+
+        tbl = Table({"a": "datetime", "b": "integer"})
+        tbl.update(data)
+        view = tbl.view(filter=[["a", "==", datetime(2019, 7, 11, 8, 15).timestamp()]])
+        assert view.to_records() == [
+            {"a": datetime(2019, 7, 11, 8, 15).timestamp(), "b": 2}
+        ]
+
+    def test_view_filter_datetime_as_number_neq(self, util):
+        data = [
+            {"a": datetime(2019, 7, 11, 8, 15).timestamp(), "b": 2},
+            {"a": datetime(2019, 7, 11, 8, 16).timestamp(), "b": 4},
+        ]
+        tbl = Table({"a": "datetime", "b": "integer"})
+        tbl.update(data)
+        view = tbl.view(filter=[["a", "!=", datetime(2019, 7, 11, 8, 15).timestamp()]])
+        assert view.to_records() == [
+            {"a": datetime(2019, 7, 11, 8, 16).timestamp(), "b": 4}
+        ]
+
+    def test_view_filter_datetime_str_eq(self, util):
+        data = [
+            {"a": datetime(2019, 7, 11, 8, 15), "b": 2},
+            {"a": datetime(2019, 7, 11, 8, 16), "b": 4},
+        ]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "==", "2019/7/11 8:15"]])
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 11, 8, 15)), "b": 2}
+        ]
+
+    def test_view_filter_datetime_str_neq(self, util):
+        data = [
+            {"a": datetime(2019, 7, 11, 8, 15), "b": 2},
+            {"a": datetime(2019, 7, 11, 8, 16), "b": 4},
+        ]
+        tbl = Table(data)
+        view = tbl.view(filter=[["a", "!=", "2019/7/11 8:15"]])
+        assert view.to_records() == [
+            {"a": util.to_timestamp(datetime(2019, 7, 11, 8, 16)), "b": 4}
+        ]
+
+    def test_view_filter_string_is_none(self):
+        data = [{"a": None, "b": 2}, {"a": "abc", "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(
+            filter=[["a", "is null", ""]]
+        )  # XXX: Having to add this "" is not soooo great.
+        assert view.to_records() == [{"a": None, "b": 2}]
+
+    def test_view_filter_string_is_not_none(self):
+        data = [{"a": None, "b": 2}, {"a": "abc", "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(
+            filter=[["a", "is not null", ""]]
+        )  # XXX: Having to add this "" is not soooo great.
+        assert view.to_records() == [{"a": "abc", "b": 4}]
+
+    # on_update
+    def test_view_on_update(self, sentinel):
+        s = sentinel(False)
+
+        def callback(port_id):
+            s.set(True)
+
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view()
+        view.on_update(callback)
+        tbl.update(data)
+        assert s.get() is True
+
+    def test_view_on_update_multiple_callback(self, sentinel):
+        s = sentinel(0)
+
+        def callback(port_id):
+            s.set(s.get() + 1)
+
+        def callback1(port_id):
+            s.set(s.get() - 1)
+
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view()
+        view.on_update(callback)
+        view.on_update(callback1)
+        tbl.update(data)
+        assert s.get() == 0
+
+    # on_delete
+
+    def test_view_on_delete(self, sentinel):
+        s = sentinel(False)
+
+        def callback():
+            s.set(True)
+
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view()
+        view.on_delete(callback)
+        view.delete()
+        assert s.get() is True
+
+    # delete
+
+    def test_view_delete(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view()
+        with raises(PerspectiveError):
+            tbl.delete()
+        view.delete()
+        tbl.delete()
+
+    def test_view_delete_multiple_callbacks(self, sentinel):
+        # make sure that callbacks on views get filtered
+        s1 = sentinel(0)
+        s2 = sentinel(0)
+
+        def cb1(port_id):
+            s1.set(s1.get() + 1)
+
+        def cb2(port_id):
+            s2.set(s2.get() + 1)
+
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        v1 = tbl.view()
+        v2 = tbl.view()
+        v1.on_update(cb1)
+        v2.on_update(cb2)
+        tbl.update(data)
+        assert s1.get() == 1
+        assert s2.get() == 1
+        v1.delete()
+        tbl.update(data)
+        assert s1.get() == 1
+        assert s2.get() == 2
+
+    def test_view_delete_full_cleanup(self, sentinel):
+        s = sentinel(0)
+
+        def cb1(port_id):
+            s.set(s.get() + 1)
+
+        def cb2(port_id):
+            s.set(s.get() + 2)
+
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        v1 = tbl.view()
+        v2 = tbl.view()
+        v1.on_update(cb1)
+        v2.on_update(cb2)
+        v1.delete()
+        v2.delete()
+        tbl.update(data)
+        assert s.get() == 0
+
+    # remove_update
+
+    def test_view_remove_update(self, sentinel):
+        s = sentinel(0)
+
+        def cb1(port_id):
+            s.set(s.get() + 1)
+
+        def cb2(port_id):
+            s.set(s.get() + 2)
+
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view()
+        t1 = view.on_update(cb1)
+        view.on_update(cb2)
+        view.remove_update(t1)
+        tbl.update(data)
+        assert s.get() == 2
+
+    def test_view_remove_multiple_update(self, sentinel):
+        s1 = sentinel(0)
+        s2 = sentinel(0)
+
+        def cb1(port_id):
+            s1.set(s1.get() + 1)
+
+        def cb2(port_id):
+            s2.set(s2.get() + 1)
+
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view()
+        t1 = view.on_update(cb1)
+        view.on_update(cb2)
+        view.on_update(cb1)
+        tbl.update(data)
+        assert s1.get() == 2
+        assert s2.get() == 1
+        view.remove_update(t1)
+        tbl.update(data)
+        assert s1.get() == 3
+        assert s2.get() == 2
+
+    # row delta
+
+    def test_view_row_delta_zero(self, util):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        update_data = {"a": [5], "b": [6]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, update_data)
+
+        tbl = Table(data)
+        view = tbl.view()
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_zero_column_subset(self, util):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        update_data = {"a": [5], "b": [6]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"b": [6]})
+
+        tbl = Table(data)
+        view = tbl.view(columns=["b"])
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_zero_from_schema(self, util):
+        update_data = {"a": [5], "b": [6]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, update_data)
+
+        tbl = Table({"a": "integer", "b": "integer"})
+        view = tbl.view()
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_zero_from_schema_column_subset(self, util):
+        update_data = {"a": [5], "b": [6]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"b": [6]})
+
+        tbl = Table({"a": "integer", "b": "integer"})
+
+        view = tbl.view(columns=["b"])
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_zero_from_schema_filtered(self, util):
+        update_data = {"a": [8, 9, 10, 11], "b": [1, 2, 3, 4]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"a": [11], "b": [4]})
+
+        tbl = Table({"a": "integer", "b": "integer"})
+        view = tbl.view(filter=[["a", ">", 10]])
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_zero_from_schema_indexed(self, util):
+        update_data = {"a": ["a", "b", "a"], "b": [1, 2, 3]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"a": ["a", "b"], "b": [3, 2]})
+
+        tbl = Table({"a": "string", "b": "integer"}, index="a")
+
+        view = tbl.view()
+        view.on_update(cb1, mode="row")
+
+        tbl.update(update_data)
+
+    def test_view_row_delta_zero_from_schema_indexed_filtered(self, util):
+        update_data = {"a": [8, 9, 10, 11, 11], "b": [1, 2, 3, 4, 5]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"a": [11], "b": [5]})
+
+        tbl = Table({"a": "integer", "b": "integer"}, index="a")
+        view = tbl.view(filter=[["a", ">", 10]])
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_one(self, util):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        update_data = {"a": [5], "b": [6]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"a": [9, 5], "b": [12, 6]})
+
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"])
+        assert view.to_columns() == {
+            "__ROW_PATH__": [[], [1], [3]],
+            "a": [4, 1, 3],
+            "b": [6, 2, 4],
+        }
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_one_from_schema(self, util):
+        update_data = {"a": [1, 2, 3, 4, 5], "b": [6, 7, 8, 9, 10]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"a": [15, 1, 2, 3, 4, 5], "b": [40, 6, 7, 8, 9, 10]})
+
+        tbl = Table({"a": "integer", "b": "integer"})
+        view = tbl.view(group_by=["a"])
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_one_from_schema_sorted(self, util):
+        update_data = {"a": [1, 2, 3, 4, 5], "b": [6, 7, 8, 9, 10]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"a": [15, 5, 4, 3, 2, 1], "b": [40, 10, 9, 8, 7, 6]})
+
+        tbl = Table({"a": "integer", "b": "integer"})
+        view = tbl.view(group_by=["a"], sort=[["a", "desc"]])
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_one_from_schema_filtered(self, util):
+        update_data = {"a": [1, 2, 3, 4, 5], "b": [6, 7, 8, 9, 10]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"a": [9, 4, 5], "b": [19, 9, 10]})
+
+        tbl = Table({"a": "integer", "b": "integer"})
+        view = tbl.view(group_by=["a"], filter=[["a", ">", 3]])
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_one_from_schema_sorted_filtered(self, util):
+        update_data = {"a": [1, 2, 3, 4, 5], "b": [6, 7, 8, 9, 10]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"a": [9, 5, 4], "b": [19, 10, 9]})
+
+        tbl = Table({"a": "integer", "b": "integer"})
+        view = tbl.view(group_by=["a"], sort=[["a", "desc"]], filter=[["a", ">", 3]])
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_one_from_schema_indexed(self, util):
+        update_data = {"a": [1, 2, 3, 4, 5, 5, 4], "b": [6, 7, 8, 9, 10, 11, 12]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"a": [15, 1, 2, 3, 4, 5], "b": [44, 6, 7, 8, 12, 11]})
+
+        tbl = Table({"a": "integer", "b": "integer"}, index="a")
+
+        view = tbl.view(group_by=["a"])
+        view.on_update(cb1, mode="row")
+
+        tbl.update(update_data)
+
+    def test_view_row_delta_one_from_schema_sorted_indexed(self, util):
+        update_data = {"a": [1, 2, 3, 4, 5, 5, 4], "b": [6, 7, 8, 9, 10, 11, 12]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"a": [15, 4, 5, 3, 2, 1], "b": [44, 12, 11, 8, 7, 6]})
+
+        tbl = Table({"a": "integer", "b": "integer"}, index="a")
+
+        view = tbl.view(group_by=["a"], sort=[["b", "desc"]])
+        view.on_update(cb1, mode="row")
+
+        tbl.update(update_data)
+
+    def test_view_row_delta_one_from_schema_filtered_indexed(self, util):
+        update_data = {"a": [1, 2, 3, 4, 5, 5, 4], "b": [6, 7, 8, 9, 10, 11, 12]}
+
+        def cb1(port_id, delta):
+            compare_delta(delta, {"a": [9, 4, 5], "b": [23, 12, 11]})
+
+        tbl = Table({"a": "integer", "b": "integer"}, index="a")
+
+        view = tbl.view(group_by=["a"], filter=[["a", ">", 3]])
+        view.on_update(cb1, mode="row")
+
+        tbl.update(update_data)
+
+    def test_view_row_delta_two(self, util):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        update_data = {"a": [5], "b": [6]}
+
+        def cb1(port_id, delta):
+            compare_delta(
+                delta,
+                {
+                    "2|a": [1, None],
+                    "2|b": [2, None],
+                    "4|a": [3, None],
+                    "4|b": [4, None],
+                    "6|a": [5, 5],
+                    "6|b": [6, 6],
+                },
+            )
+
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"], split_by=["b"])
+        assert view.to_columns() == {
+            "__ROW_PATH__": [[], [1], [3]],
+            "2|a": [1, 1, None],
+            "2|b": [2, 2, None],
+            "4|a": [3, None, 3],
+            "4|b": [4, None, 4],
+        }
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_two_from_schema(self, util):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+
+        def cb1(port_id, delta):
+            compare_delta(
+                delta,
+                {
+                    "2|a": [1, 1, None],
+                    "2|b": [2, 2, None],
+                    "4|a": [3, None, 3],
+                    "4|b": [4, None, 4],
+                },
+            )
+
+        tbl = Table({"a": "integer", "b": "integer"})
+        view = tbl.view(group_by=["a"], split_by=["b"])
+        view.on_update(cb1, mode="row")
+        tbl.update(data)
+
+    def test_view_row_delta_two_from_schema_indexed(self, util):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}, {"a": 3, "b": 5}]
+
+        def cb1(port_id, delta):
+            compare_delta(
+                delta,
+                {
+                    "2|a": [1, 1, None],
+                    "2|b": [2, 2, None],
+                    "5|a": [3, None, 3],
+                    "5|b": [5, None, 5],
+                },
+            )
+
+        tbl = Table({"a": "integer", "b": "integer"}, index="a")
+        view = tbl.view(group_by=["a"], split_by=["b"])
+        view.on_update(cb1, mode="row")
+        tbl.update(data)
+
+    def test_view_row_delta_two_column_only(self, util):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        update_data = {"a": [5], "b": [6]}
+
+        def cb1(port_id, delta):
+            compare_delta(
+                delta,
+                {
+                    "2|a": [1, None],
+                    "2|b": [2, None],
+                    "4|a": [3, None],
+                    "4|b": [4, None],
+                    "6|a": [5, 5],
+                    "6|b": [6, 6],
+                },
+            )
+
+        tbl = Table(data)
+        view = tbl.view(split_by=["b"])
+        assert view.to_columns() == {
+            "2|a": [1, None],
+            "2|b": [2, None],
+            "4|a": [None, 3],
+            "4|b": [None, 4],
+        }
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_two_column_only_indexed(self, util):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}, {"a": 3, "b": 5}]
+        update_data = {"a": [5], "b": [6]}
+
+        def cb1(port_id, delta):
+            compare_delta(
+                delta,
+                {
+                    "2|a": [1, None],
+                    "2|b": [2, None],
+                    "5|a": [3, None],
+                    "5|b": [5, None],
+                    "6|a": [5, 5],
+                    "6|b": [6, 6],
+                },
+            )
+
+        tbl = Table(data, index="a")
+        view = tbl.view(split_by=["b"])
+        assert view.to_columns() == {
+            "2|a": [1, None],
+            "2|b": [2, None],
+            "5|a": [None, 3],
+            "5|b": [None, 5],
+        }
+        view.on_update(cb1, mode="row")
+        tbl.update(update_data)
+
+    def test_view_row_delta_two_column_only_from_schema(self, util):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+
+        def cb1(port_id, delta):
+            compare_delta(
+                delta,
+                {
+                    "2|a": [1, 1, None],
+                    "2|b": [2, 2, None],
+                    "4|a": [3, None, 3],
+                    "4|b": [4, None, 4],
+                },
+            )
+
+        tbl = Table({"a": "integer", "b": "integer"})
+        view = tbl.view(split_by=["b"])
+        view.on_update(cb1, mode="row")
+        tbl.update(data)
+
+    def test_view_row_delta_two_column_only_from_schema_indexed(self, util):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}, {"a": 3, "b": 5}]
+
+        def cb1(port_id, delta):
+            compare_delta(
+                delta,
+                {
+                    "2|a": [1, 1, None],
+                    "2|b": [2, 2, None],
+                    "5|a": [3, None, 3],
+                    "5|b": [5, None, 5],
+                },
+            )
+
+        tbl = Table({"a": "integer", "b": "integer"}, index="a")
+        view = tbl.view(split_by=["b"])
+        view.on_update(cb1, mode="row")
+        tbl.update(data)
+
+    # hidden cols
+
+    def test_view_num_hidden_cols(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(columns=["a"], sort=[["b", "desc"]])
+        cols = view.to_columns()
+        assert cols == {"a": [3, 1]}
+
+    def test_view_context_two_update_clears_column_regression(self, util):
+        """Tests that, when a 2-sided View() is updated to a state where one of
+        the column groups is empty, an infinite loop is not encountered.
+        """
+        data = [
+            {"a": "a", "b": 1, "c": 1.5, "i": 0},
+            {"a": "a", "b": 2, "c": 2.5, "i": 1},
+            {"a": "a", "b": 3, "c": 3.5, "i": 2},
+            {"a": "b", "b": 1, "c": 4.5, "i": 3},
+            {"a": "b", "b": 2, "c": 5.5, "i": 4},
+            {"a": "b", "b": 3, "c": 6.5, "i": 5},
+        ]
+
+        tbl = Table(data, index="i")
+        view = tbl.view(
+            group_by=["b"],
+            split_by=["a"],
+            columns=["c"],
+            filter=[["c", ">", 0]],
+            sort=[["c", "asc"], ["a", "col asc"]],
+        )
+
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "a|c": 7.5, "b|c": 16.5},
+            {"__ROW_PATH__": [1], "a|c": 1.5, "b|c": 4.5},
+            {"__ROW_PATH__": [2], "a|c": 2.5, "b|c": 5.5},
+            {"__ROW_PATH__": [3], "a|c": 3.5, "b|c": 6.5},
+        ]
+
+        tbl.update(
+            [
+                {"c": -1, "i": 0},
+                {"c": -1, "i": 1},
+                {"c": -1, "i": 2},
+            ]
+        )
+
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "b|c": 16.5},
+            {"__ROW_PATH__": [1], "b|c": 4.5},
+            {"__ROW_PATH__": [2], "b|c": 5.5},
+            {"__ROW_PATH__": [3], "b|c": 6.5},
+        ]
+
+        tbl.update(
+            [
+                {"a": "a", "b": 1, "c": 1.5, "i": 6},
+                {"a": "a", "b": 2, "c": 2.5, "i": 7},
+                {"a": "a", "b": 3, "c": 3.5, "i": 8},
+            ]
+        )
+
+        assert view.to_records() == [
+            {"__ROW_PATH__": [], "a|c": 7.5, "b|c": 16.5},
+            {"__ROW_PATH__": [1], "a|c": 1.5, "b|c": 4.5},
+            {"__ROW_PATH__": [2], "a|c": 2.5, "b|c": 5.5},
+            {"__ROW_PATH__": [3], "a|c": 3.5, "b|c": 6.5},
+        ]
+
+        assert tbl.size() == 9
+
+    # expand/collapse
+
+    def test_view_collapse_one(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"])
+        assert view.collapse(0) == 2
+
+    def test_view_collapse_two(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"], split_by=["c"])
+        assert view.collapse(0) == 2
+
+    # TODO collapse/espand should be no-ops on column only contexts, but
+    # the concept of "column only" is not yet implemented in C++
+    @mark.skip
+    def test_view_collapse_two_column_only(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        view = tbl.view(split_by=["c"])
+        assert view.collapse(0) == 0
+
+    def test_view_expand_one(self):
+        data = [{"a": 1, "b": 2}, {"a": 3, "b": 4}]
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"])
+        assert view.expand(0) == 0
+
+    def test_view_expand_two(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        view = tbl.view(group_by=["a"], split_by=["c"])
+        assert view.expand(1) == 1
+
+    def test_view_expand_two_column_only(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        view = tbl.view(split_by=["c"])
+        assert view.expand(0) == 0
+
+    # view config validation
+
+    def test_invalid_column_should_throw(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        with raises(PerspectiveError) as ex:
+            tbl.view(columns=["x"])
+        assert str(ex.value) == "Abort(): Invalid column 'x' found in View columns.\n"
+
+    def test_invalid_column_should_throw_and_updates_should_work(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+
+        with raises(PerspectiveError) as ex:
+            tbl.view(columns=["x"])
+        assert str(ex.value) == "Abort(): Invalid column 'x' found in View columns.\n"
+
+        for i in range(100):
+            tbl.update(data)
+            # force call to _process which should shake out invalid column ptrs
+            tbl.size()
+
+        view2 = tbl.view()
+        assert view2.num_rows() == 202
+
+    def test_invalid_column_aggregate_should_throw(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+
+        with raises(PerspectiveError) as ex:
+            tbl.view(columns=["x"], aggregates={"x": "sum"})
+
+        assert str(ex.value) == "Abort(): Invalid column 'x' found in View columns.\n"
+
+    def test_invalid_column_aggregate_should_throw_and_updates_should_work(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+
+        with raises(PerspectiveError) as ex:
+            tbl.view(columns=["x"], aggregates={"x": "sum"})
+
+        assert str(ex.value) == "Abort(): Invalid column 'x' found in View columns.\n"
+
+        for i in range(100):
+            tbl.update(data)
+            # force call to _process which should shake out invalid column ptrs
+            tbl.size()
+
+        view2 = tbl.view()
+        assert view2.num_rows() == 202
+
+    def test_invalid_group_by_should_throw(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        with raises(PerspectiveError) as ex:
+            tbl.view(group_by=["x"])
+        assert str(ex.value) == "Abort(): Invalid column 'x' found in View group_by.\n"
+
+    def test_invalid_split_by_should_throw(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        with raises(PerspectiveError) as ex:
+            tbl.view(split_by=["x"])
+        assert str(ex.value) == "Abort(): Invalid column 'x' found in View split_by.\n"
+
+    def test_invalid_filters_should_throw(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        with raises(PerspectiveError) as ex:
+            tbl.view(filter=[["x", "==", "abc"]])
+        assert str(ex.value) == "Abort(): Filter column not in schema: x"
+
+    def test_invalid_sorts_should_throw(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        with raises(PerspectiveError) as ex:
+            tbl.view(sort=[["x", "desc"]])
+        assert str(ex.value) == "Abort(): Invalid column 'x' found in View sorts.\n"
+
+    def test_should_throw_on_first_invalid(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        with raises(PerspectiveError) as ex:
+            tbl.view(
+                group_by=["a"],
+                split_by=["c"],
+                filter=[["a", ">", 1]],
+                aggregates={"a": "avg"},
+                sort=[["x", "desc"]],
+            )
+        assert str(ex.value) == "Abort(): Invalid column 'x' found in View sorts.\n"
+
+    def test_invalid_columns_not_in_expression_should_throw(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        with raises(PerspectiveError) as ex:
+            tbl.view(columns=["abc", "x"], expressions={"abc": "1 + 2"})
+        assert str(ex.value) == "Abort(): Invalid column 'x' found in View columns.\n"
+
+    def test_should_not_throw_valid_expression(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        view = tbl.view(columns=["abc"], expressions={"abc": "'hello!'"})
+
+        assert view.schema() == {"abc": "string"}
+
+    def test_should_not_throw_valid_expression_config(self):
+        data = [{"a": 1, "b": 2, "c": "a"}, {"a": 3, "b": 4, "c": "b"}]
+        tbl = Table(data)
+        view = tbl.view(
+            aggregates={"abc": "dominant"},
+            columns=["abc"],
+            sort=[["abc", "desc"]],
+            filter=[["abc", "==", "A"]],
+            group_by=["abc"],
+            split_by=["abc"],
+            expressions={"abc": "'hello!'"},
+        )
+
+        assert view.schema() == {"abc": "string"}