def timeseries2seqs_3D(data, timestamps, length=3, T=48):
raw_ts = copy(timestamps)
if type(timestamps[0]) != pd.Timestamp:
timestamps = string2timestamp(timestamps, T=T)
offset = pd.DateOffset(minutes=24 * 60 // T)
breakpoints = [0]
for i in range(1, len(timestamps)):
if timestamps[i - 1] + offset != timestamps[i]:
print(timestamps[i - 1], timestamps[i], raw_ts[i - 1], raw_ts[i])
breakpoints.append(i)
breakpoints.append(len(timestamps))
X = []
Y = []
for b in range(1, len(breakpoints)):
print('breakpoints: ', breakpoints[b - 1], breakpoints[b])
idx = range(breakpoints[b - 1], breakpoints[b])
for i in range(len(idx) - length):
x = data[idx[i:i + length]].reshape(-1, length, 32, 32)
y = np.asarray([data[idx[i + length]]]).reshape(-1, 1, 32, 32)
X.append(x)
Y.append(y)
X = np.asarray(X)
Y = np.asarray(Y)
print("X shape: ", X.shape, "Y shape:", Y.shape)
return X, Y
def bug_timeseries2seqs(data, timestamps, length=3, T=48):
# have a bug
if type(timestamps[0]) != pd.Timestamp:
timestamps = string2timestamp(timestamps, T=T)
offset = pd.DateOffset(minutes=24 * 60 // T)
breakpoints = [0]
for i in range(1, len(timestamps)):
if timestamps[i - 1] + offset != timestamps[i]:
breakpoints.append(i)
X = []
Y = []
for b in range(1, len(breakpoints)):
print('breakpoints: ', breakpoints[b - 1], breakpoints[b])
idx = range(breakpoints[b - 1], breakpoints[b])
for i in range(len(idx) - 3):
x = np.vstack(data[idx[i:i + 3]])
y = data[idx[i + 3]]
X.append(x)
Y.append(y)
X = np.asarray(X)
Y = np.asarray(Y)
print("X shape: ", X.shape, "Y shape:", Y.shape)
return X, Y
def __init__(self, data, timestamps, data_air, T=48, CheckComplete=True):
super(STMatrix, self).__init__()
assert len(data) == len(timestamps) #时间长度和数据信息相等
self.data = data
self.timestamps = timestamps
self.data_air = data_air
self.T = T
self.pd_timestamps = string2timestamp(timestamps, T=self.T)
if CheckComplete:
self.check_complete()
# index
self.make_index()
def timeseries2seqs_peroid_trend(data,
timestamps,
length=3,
T=48,
peroid=pd.DateOffset(days=7),
peroid_len=2):
raw_ts = copy(timestamps)
if type(timestamps[0]) != pd.Timestamp:
timestamps = string2timestamp(timestamps, T=T)
# timestamps index
timestamp_idx = dict()
for i, t in enumerate(timestamps):
timestamp_idx[t] = i
offset = pd.DateOffset(minutes=24 * 60 // T)
breakpoints = [0]
for i in range(1, len(timestamps)):
if timestamps[i - 1] + offset != timestamps[i]:
print(timestamps[i - 1], timestamps[i], raw_ts[i - 1], raw_ts[i])
breakpoints.append(i)
breakpoints.append(len(timestamps))
X = []
Y = []
for b in range(1, len(breakpoints)):
print('breakpoints: ', breakpoints[b - 1], breakpoints[b])
idx = range(breakpoints[b - 1], breakpoints[b])
for i in range(len(idx) - length):
# period
target_timestamp = timestamps[i + length]
legal_idx = []
for pi in range(1, 1 + peroid_len):
if target_timestamp - peroid * pi not in timestamp_idx:
break
legal_idx.append(timestamp_idx[target_timestamp - peroid * pi])
# print("len: ", len(legal_idx), peroid_len)
if len(legal_idx) != peroid_len:
continue
legal_idx += idx[i:i + length]
# trend
x = np.vstack(data[legal_idx])
y = data[idx[i + length]]
X.append(x)
Y.append(y)
X = np.asarray(X)
Y = np.asarray(Y)
print("X shape: ", X.shape, "Y shape:", Y.shape)
return X, Y
