__init__.py 文件源码

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