projektAI/venv/Lib/site-packages/pandas/tests/series/methods/test_convert.py
2021-06-06 22:13:05 +02:00

136 lines
4.8 KiB
Python

from datetime import datetime
import numpy as np
import pytest
from pandas import Series, Timestamp
import pandas._testing as tm
class TestConvert:
def test_convert(self):
# GH#10265
dt = datetime(2001, 1, 1, 0, 0)
td = dt - datetime(2000, 1, 1, 0, 0)
# Test coercion with mixed types
ser = Series(["a", "3.1415", dt, td])
results = ser._convert(numeric=True)
expected = Series([np.nan, 3.1415, np.nan, np.nan])
tm.assert_series_equal(results, expected)
# Test standard conversion returns original
results = ser._convert(datetime=True)
tm.assert_series_equal(results, ser)
results = ser._convert(numeric=True)
expected = Series([np.nan, 3.1415, np.nan, np.nan])
tm.assert_series_equal(results, expected)
results = ser._convert(timedelta=True)
tm.assert_series_equal(results, ser)
# test pass-through and non-conversion when other types selected
ser = Series(["1.0", "2.0", "3.0"])
results = ser._convert(datetime=True, numeric=True, timedelta=True)
expected = Series([1.0, 2.0, 3.0])
tm.assert_series_equal(results, expected)
results = ser._convert(True, False, True)
tm.assert_series_equal(results, ser)
ser = Series(
[datetime(2001, 1, 1, 0, 0), datetime(2001, 1, 1, 0, 0)], dtype="O"
)
results = ser._convert(datetime=True, numeric=True, timedelta=True)
expected = Series([datetime(2001, 1, 1, 0, 0), datetime(2001, 1, 1, 0, 0)])
tm.assert_series_equal(results, expected)
results = ser._convert(datetime=False, numeric=True, timedelta=True)
tm.assert_series_equal(results, ser)
td = datetime(2001, 1, 1, 0, 0) - datetime(2000, 1, 1, 0, 0)
ser = Series([td, td], dtype="O")
results = ser._convert(datetime=True, numeric=True, timedelta=True)
expected = Series([td, td])
tm.assert_series_equal(results, expected)
results = ser._convert(True, True, False)
tm.assert_series_equal(results, ser)
ser = Series([1.0, 2, 3], index=["a", "b", "c"])
result = ser._convert(numeric=True)
tm.assert_series_equal(result, ser)
# force numeric conversion
res = ser.copy().astype("O")
res["a"] = "1"
result = res._convert(numeric=True)
tm.assert_series_equal(result, ser)
res = ser.copy().astype("O")
res["a"] = "1."
result = res._convert(numeric=True)
tm.assert_series_equal(result, ser)
res = ser.copy().astype("O")
res["a"] = "garbled"
result = res._convert(numeric=True)
expected = ser.copy()
expected["a"] = np.nan
tm.assert_series_equal(result, expected)
# GH 4119, not converting a mixed type (e.g.floats and object)
ser = Series([1, "na", 3, 4])
result = ser._convert(datetime=True, numeric=True)
expected = Series([1, np.nan, 3, 4])
tm.assert_series_equal(result, expected)
ser = Series([1, "", 3, 4])
result = ser._convert(datetime=True, numeric=True)
tm.assert_series_equal(result, expected)
# dates
ser = Series(
[
datetime(2001, 1, 1, 0, 0),
datetime(2001, 1, 2, 0, 0),
datetime(2001, 1, 3, 0, 0),
]
)
result = ser._convert(datetime=True)
expected = Series(
[Timestamp("20010101"), Timestamp("20010102"), Timestamp("20010103")],
dtype="M8[ns]",
)
tm.assert_series_equal(result, expected)
result = ser._convert(datetime=True)
tm.assert_series_equal(result, expected)
# preserver if non-object
ser = Series([1], dtype="float32")
result = ser._convert(datetime=True)
tm.assert_series_equal(result, ser)
# FIXME: dont leave commented-out
# res = ser.copy()
# r[0] = np.nan
# result = res._convert(convert_dates=True,convert_numeric=False)
# assert result.dtype == 'M8[ns]'
def test_convert_no_arg_error(self):
ser = Series(["1.0", "2"])
msg = r"At least one of datetime, numeric or timedelta must be True\."
with pytest.raises(ValueError, match=msg):
ser._convert()
def test_convert_preserve_bool(self):
ser = Series([1, True, 3, 5], dtype=object)
res = ser._convert(datetime=True, numeric=True)
expected = Series([1, 1, 3, 5], dtype="i8")
tm.assert_series_equal(res, expected)
def test_convert_preserve_all_bool(self):
ser = Series([False, True, False, False], dtype=object)
res = ser._convert(datetime=True, numeric=True)
expected = Series([False, True, False, False], dtype=bool)
tm.assert_series_equal(res, expected)