# Reshaping a 2D matrix of Time series vectors into a 3D matrix of sequences (frames) – overlapping windows

## Question:

I have a matrix (shape: **m** by 51) of 51 time series vectors **m** samples each. I want to train two autoencoders one using CNN and another using LSTM network. I want to reshape the 2D matrix into a 3D matrix such that it contains **m_new** sequences for each of the 51 variables and each sequence is **w** long with overlapping of **lap** samples.

I managed to pull this off but without the overlapping part. Is there an efficient way to do it?

```
W = 20 #window size
m_new = int(np.floor(m/W))
m_trct = int(m_new*W)
X_raw_trct = X_raw[0:m_trct,:]
X = np.reshape(X_raw_trct,(m_new,W,X_raw_trct.shape[1]))
```

As demonstrated below, the sequences are generated with overlapping of **lap = w-1**.

** **UPDATE** **

In reference to the answer in Split Python sequence (time series/array) into subsequences with overlap,

using the function **sub-sequences** which splits the 1D array into **w** long sub-sequences with overlap of **w-1** (stride of 1) resulting in a 2D array of shape (**m_new, w**) . As in **code 2**

below, I had to use a loop to work every vector of the 51 variables as a 1D array then appending the results of the 2D arrays to produce my final 3D array of shape (**m_new, w, 51**). However, the loop takes so long to execute.

```
**code 2**
def subsequences(ts, window):
## ts is of shape (m,)
shape = (ts.size - window + 1, window)
strides = ts.strides * 2
return np.lib.stride_tricks.as_strided(ts, shape=shape, strides=strides)
# rescaledX_raw.shape is (m,51)
n = rescaledX_raw.shape[1]
# n = 51
a = rescaledX_raw[:,0]
# a.shape is (m,)
Xaa = subsequences(a,W)
X = ones(Xaa.shape)*-1
# X.shape is (m_new, W)
for kk in range(n):
## a is of shape (m,)
a = rescaledX_raw[:,kk]
Xaa = subsequences(a,W)
X = np.dstack((X, Xaa))
X_nn = np.delete(X, 0, axis=2)
# X_nn.shape is (m_new, W, 51)
```

In addition, I tried to work it out as a full 2D array of shape (**m** by 51) to the 3D array of shape (**m_new,w**,51) using the function in **code 3**

```
**code 3**
def rolling_window(a, window):
## a is of shape (51,m)
shape = (a.shape[-1] - window + 1,window,a.shape[0])
strides = a.strides + (a.strides[-1],)
return np.lib.stride_tricks.as_strided(a, shape=shape, strides=strides)
```

But the resulted 3D matrix is not the correct one. Kindly refer to the demonstration below. In addition, how can I add the stride as a variable I can change. In scripts above the stride is 1 (meaning the overlap is **w-1**)

## Answers:

I found a helpful post to get this done using **TimeseriesGenerator**. Custom Data Generator for Keras LSTM with TimeSeriesGenerator

```
class CustomGenFit(TimeseriesGenerator):
def __getitem__(self, idx):
x, y = super().__getitem__(idx)
return x, x
```

```
Xsequences = CustomGenPredict(X, X, length=W, stride = s,sampling_rate=1, batch_size=m)
```

```
def lstm_data_transform(x_data, y_data, num_steps=10):
""" Changes data to the format for LSTM training
for sliding window approach """
# Prepare the list for the transformed data
X, y = list(), list()
# Loop of the entire data set
for i in range(x_data.shape[0]):
# compute a new (sliding window) index
end_ix = i + num_steps
# if index is larger than the size of the dataset, we stop
if end_ix >= x_data.shape[0]:
break
# Get a sequence of data for x
seq_X = x_data[i:end_ix]
# Get only the last element of the sequency for y
#seq_y = y_data[end_ix]#ori end-----fking somw wrong
seq_y = y_data[i]#first correct wtf
# Append the list with sequencies
X.append(seq_X)
y.append(seq_y)
# Make final arrays
x_array = np.array(X)
y_array = np.array(y)
return x_array, y_array
```