def test_trend_scanning_labels(self):
"""
Test trend scanning labels
"""
t_events = self.eem_close.index
tr_scan_labels = trend_scanning_labels(self.eem_close, t_events, 20)
self.assertEqual(tr_scan_labels.shape[0],
len(t_events)) # we have label value for all t events
# Before 2008/5/12 we had a strong positive trend
self.assertTrue(
set(
tr_scan_labels.reindex(
pd.date_range(pd.Timestamp(2008, 1, 1),
pd.Timestamp(2008, 5, 9))).dropna().bin) ==
set([1]))
self.assertEqual(tr_scan_labels.bin.value_counts()[-1],
70) # Number of -1 labels check
self.assertEqual(tr_scan_labels.bin.value_counts()[1], 40)
# Approx. 20 latest labels are nan because we don't have data for look-forward period (20)
self.assertEqual(
tr_scan_labels.shape[0] - tr_scan_labels.dropna().shape[0], 19)
tr_scan_labels.dropna(inplace=True) # Drop na values
# Index should be < t1
self.assertTrue((tr_scan_labels.t1 > tr_scan_labels.index).all())
for int_index, (ret_v, bin_v) in zip([0, 2, 10, 20, 50],
[(0.05037, 1), (0.0350, 1),
(0.07508, 1), (0.05219, 1),
(0.02447, 1)]):
self.assertAlmostEqual(tr_scan_labels.iloc[int_index]['ret'],
ret_v,
delta=1e-4)
self.assertEqual(tr_scan_labels.iloc[int_index]['bin'], bin_v)
tr_scan_labels_none = trend_scanning_labels(self.eem_close,
t_events=None,
look_forward_window=20)
tr_scan_labels_none.dropna(inplace=True)
self.assertTrue((tr_scan_labels == tr_scan_labels_none).all().all())
def test_trend_scanning_labels_backward(self):
"""
Test trend scanning labels with backward-looking window
"""
t_events = self.eem_close.index
tr_scan_labels = trend_scanning_labels(self.eem_close, t_events, 20,
False)
self.assertEqual(tr_scan_labels.shape[0],
len(t_events)) # we have label value for all t events
# Looking backwards, before 2008/5/22 was a strong positive trend
self.assertTrue(
set(tr_scan_labels.loc[pd.Timestamp(2008, 4, 29):pd.
Timestamp(2008, 5, 22)].bin) == set([1]))
self.assertEqual(tr_scan_labels.bin.value_counts()[-1],
86) # Number of -1 labels check
self.assertEqual(tr_scan_labels.bin.value_counts()[1], 24)
# First 20 labels are nan because we don't have data for look-backward period (20)
self.assertEqual(
tr_scan_labels.shape[0] - tr_scan_labels.dropna().shape[0], 19)
tr_scan_labels.dropna(inplace=True) # Drop na values
# Index should be > t1
self.assertTrue((tr_scan_labels.t1 < tr_scan_labels.index).all())
for int_index, (ret_v, bin_v) in zip([0, 2, 10, 20, 50],
[(-0.0025, 1), (-0.0201, 1),
(-0.0126, 1), (0.0185, -1),
(0.0147, -1)]):
self.assertAlmostEqual(tr_scan_labels.iloc[int_index]['ret'],
ret_v,
delta=1e-4)
self.assertEqual(tr_scan_labels.iloc[int_index]['bin'], bin_v)
tr_scan_labels_none = trend_scanning_labels(self.eem_close,
t_events=None,
observation_window=20,
look_forward=False)
tr_scan_labels_none.dropna(inplace=True)
self.assertTrue((tr_scan_labels == tr_scan_labels_none).all().all())
max_abs_t_val = -np.inf
max_t_val = None
max_t_index = None
for fwd in range(min_obs, len(lookForward)):
y = lookForward.iloc[:fwd]
tVal = tValOLS(y)
if abs(tVal) > max_abs_t_val:
max_abs_t_val = abs(tVal)
max_t_val = tVal
max_t_index = y.index[-1]
ser.loc[curr_t, "t_val"] = max_t_val
ser.loc[curr_t, "st"] = max_t_index
# print(f"curr_t: {curr_t}, max_abs_t_val: {max_abs_t_val}, max_t_index: {max_t_index}")
# print(ser[name].loc[max_t_index])
# print(ser[name].loc[curr_t])
ser.loc[curr_t,
"ret"] = ser[name].loc[max_t_index] / ser[name].loc[curr_t] - 1
return ser
df = pd.DataFrame({"close": 100 * np.sin(range(0, 90))})
print(df["close"])
trend_labelling(df["close"], span=20).to_csv("own.csv")
trend_scanning_labels(price_series=df["close"],
t_events=df.index,
look_forward_window=20,
min_sample_length=5).to_csv("mfinlab.csv")