tensorflow - Most efficient way to reshape tensor into sequences -

i working audio in tensorflow, , obtain series of sequences obtained sliding window on data, speak. examples illustrate situation:

current data format:

shape = [batch_size, num_features]

example = [   [1, 2, 3],   [4, 5, 6],   [7, 8, 9],   [10, 11, 12],   [13, 14, 15] ] 

what want:

shape = [batch_size - window_length + 1, window_length, num_features]

example = [   [     [1, 2, 3],     [4, 5, 6],     [7, 8, 9]   ],   [     [4, 5, 6],     [7, 8, 9],     [10, 11, 12]   ],   [     [7, 8, 9],     [10, 11, 12],     [13, 14, 15]   ], ] 

my current solution this:

list_of_windows_of_data = [] x in range(batch_size - window_length + 1):     list_of_windows_of_data.append(tf.slice(data, [x, 0], [window_length,             num_features])) windowed_data = tf.squeeze(tf.stack(list_of_windows_of_data, axis=0)) 

and transform. however, creates 20,000 operations slows tensorflow down lot when creating graph. if else has fun , more efficient way this, please share.

you can using tf.map_fn follows:

example = tf.constant([   [1, 2, 3],   [4, 5, 6],   [7, 8, 9],   [10, 11, 12],   [13, 14, 15] ] ) res = tf.map_fn(lambda i: example[i:i+3], tf.range(example.shape[0]-2), dtype=tf.int32) sess=tf.interactivesession() res.eval() 

this prints

array([[[ 1,  2,  3],         [ 4,  5,  6],         [ 7,  8,  9]],         [[ 4,  5,  6],         [ 7,  8,  9],         [10, 11, 12]],         [[ 7,  8,  9],         [10, 11, 12],         [13, 14, 15]]])