Update ConvLSTMCell to accept shape as list or tuple

PiperOrigin-RevId: 228852641

Author: tensorflower-gardener

tensorflow/contrib/rnn/python/ops/rnn_cell.py

@@ -2071,7 +2071,7 @@ def __init__(self,
       conv_ndims: Convolution dimensionality (1, 2 or 3).
       input_shape: Shape of the input as int tuple, excluding the batch size.
       output_channels: int, number of output channels of the conv LSTM.
-      kernel_shape: Shape of kernel as in tuple (of size 1,2 or 3).
+      kernel_shape: Shape of kernel as an int tuple (of size 1, 2 or 3).
       use_bias: (bool) Use bias in convolutions.
       skip_connection: If set to `True`, concatenate the input to the
         output of the conv LSTM. Default: `False`.
@@ -2092,7 +2092,7 @@ def __init__(self,
     self._conv_ndims = conv_ndims
     self._input_shape = input_shape
     self._output_channels = output_channels
-    self._kernel_shape = kernel_shape
+    self._kernel_shape = list(kernel_shape)
     self._use_bias = use_bias
     self._forget_bias = forget_bias
     self._skip_connection = skip_connection
@@ -2172,7 +2172,7 @@ def _conv(args, filter_size, num_features, bias, bias_start=0.0):
   Args:
     args: a Tensor or a list of Tensors of dimension 3D, 4D or 5D,
     batch x n, Tensors.
-    filter_size: int tuple of filter height and width.
+    filter_size: int tuple of filter shape (of size 1, 2 or 3).
     num_features: int, number of features.
     bias: Whether to use biases in the convolution layer.
     bias_start: starting value to initialize the bias; 0 by default.