def test_from_gluonts_list_dataset():
number_of_ts = 10 # number of time series
ts_length = 100 # number of timesteps
prediction_length = 24
freq = "D"
custom_dataset = np.random.normal(size=(number_of_ts, ts_length))
start = pd.Timestamp("01-01-2019", freq=freq) # type: ignore
gluonts_list_dataset = ListDataset(
[{
"target": x,
"start": start
} for x in custom_dataset[:, :-prediction_length]],
freq=freq,
)
TimeSeriesDataFrame(gluonts_list_dataset)
ts_df = TimeSeriesDataFrame(ListDataset(SAMPLE_ITERABLE, freq=freq))
pd.testing.assert_frame_equal(ts_df,
SAMPLE_TS_DATAFRAME,
check_dtype=False)
empty_list_dataset = ListDataset([], freq=freq)
with pytest.raises(ValueError):
TimeSeriesDataFrame(empty_list_dataset)
def test_validate_data_frame():
item_ids = pd.Series(np.repeat(ITEM_IDS, 3))
datetimes = pd.Series(np.tile(DATETIME_INDEX, 3))
targets = pd.Series(TARGETS)
df = pd.concat([item_ids, datetimes, targets], axis=1)
with pytest.raises(ValueError):
TimeSeriesDataFrame(df)
df.columns = ["item_id", "timestamp", "target"]
TimeSeriesDataFrame(df)
def test_when_dataset_constructed_with_irregular_timestamps_then_constructor_raises(
list_of_timestamps):
df_tuples = []
for i, ts in enumerate(list_of_timestamps):
for t in ts:
df_tuples.append((i, pd.Timestamp(t), np.random.rand()))
df = pd.DataFrame(df_tuples, columns=[ITEMID, TIMESTAMP, "target"])
with pytest.raises(ValueError, match="uniformly sampled"):
TimeSeriesDataFrame.from_data_frame(df)
def test_validate_multi_index_data_frame():
TimeSeriesDataFrame(SAMPLE_TS_DATAFRAME)
target = list(range(4))
item_ids = (1, 2, 3, 4)
with pytest.raises(ValueError):
TimeSeriesDataFrame(np.array([item_ids, target]).T, freq="D")
ts_df = pd.Series(target, name="target", index=item_ids).to_frame()
with pytest.raises(ValueError):
TimeSeriesDataFrame(ts_df, freq="D")
def test_from_iterable():
ts_df = TimeSeriesDataFrame(SAMPLE_ITERABLE)
pd.testing.assert_frame_equal(ts_df, SAMPLE_TS_DATAFRAME, check_dtype=True)
with pytest.raises(ValueError):
TimeSeriesDataFrame([])
sample_iter = [{"target": [0, 1, 2]}]
with pytest.raises(ValueError):
TimeSeriesDataFrame(sample_iter)
sample_iter = [{
"target": [0, 1, 2],
"start": pd.Timestamp("01-01-2019")
}] # type: ignore
with pytest.raises(ValueError):
TimeSeriesDataFrame(sample_iter)
def test_when_dataset_constructed_from_dataframe_without_freq_then_freq_is_inferred(
timestamps, expected_freq):
df = pd.DataFrame({
"item_id": [0, 0, 0],
"target": [1, 2, 3],
"timestamp": map(pd.Timestamp, timestamps), # noqa
})
ts_df = TimeSeriesDataFrame.from_data_frame(df)
assert ts_df.freq == expected_freq
def test_when_dataframe_class_rename_called_then_output_correct(
input_df, inplace):
renamed_df = TimeSeriesDataFrame.rename(input_df,
columns={"target": "mytarget"},
inplace=inplace)
if inplace:
renamed_df = input_df
assert isinstance(renamed_df, TimeSeriesDataFrame)
assert "mytarget" in renamed_df.columns
assert "target" not in renamed_df.columns
if inplace:
assert renamed_df._data is input_df._data
def test_when_dataset_sliced_by_step_then_output_times_and_values_correct(
input_iterable, input_slice, expected_times, expected_values):
df = TimeSeriesDataFrame.from_iterable_dataset(input_iterable)
dfv = df.slice_by_timestep(input_slice)
if not expected_times:
assert len(dfv) == 0
assert np.allclose(dfv["target"], expected_values)
assert isinstance(dfv, TimeSeriesDataFrame)
assert all(ixval[1] == pd.Timestamp(expected_times[i])
for i, ixval in enumerate(dfv.index.values)) # type: ignore
def test_when_dataset_constructed_via_constructor_with_freq_then_freq_is_inferred(
start_time, freq):
item_list = ListDataset(
[{
"target": [1, 2, 3],
"start": pd.Timestamp(start_time, freq=freq)
} for _ in range(3)], # type: ignore
freq=freq,
)
ts_df = TimeSeriesDataFrame(item_list)
assert ts_df.freq == freq
def test_when_dataframe_class_copy_called_then_output_correct(input_df):
copied_df = TimeSeriesDataFrame.copy(input_df, deep=True)
assert isinstance(copied_df, TimeSeriesDataFrame)
assert copied_df._data is not input_df._data
def _build_ts_dataframe(item_ids, datetime_index, target):
multi_inds = pd.MultiIndex.from_product([item_ids, datetime_index],
names=["item_id", "timestamp"])
return TimeSeriesDataFrame(
pd.Series(target, name="target", index=multi_inds).to_frame())
def test_from_data_frame():
tsdf_from_data_frame = TimeSeriesDataFrame(SAMPLE_DATAFRAME)
pd.testing.assert_frame_equal(tsdf_from_data_frame,
SAMPLE_TS_DATAFRAME,
check_dtype=True)