Update recurrent.py (#676)

* Update recurrent.py

* Update CHANGELOG.md

* Update recurrent.py

* YAPF Cleaning

Author: dengyueyun666

CHANGELOG.md

@@ -144,6 +144,7 @@ To release a new version, please update the changelog as followed:
 - Issue #498 - Deprecation Warning Fix in `tl.files` with truth value of an empty array is ambiguous (by @DEKHTIARJonathan in #575)
 - Issue #565 related to `tl.utils.predict` fixed - `np.hstack` problem in which the results for multiple batches are stacked along `axis=1` (by @2wins in #566)
 - Issue #572 with `tl.layers.DeformableConv2d` fixed (by @DEKHTIARJonathan in #573)
+- Issue #664 with `tl.layers.ConvLSTMLayer` fixed (by @dengyueyun666 in #676)
 - Typo of the document of ElementwiseLambdaLayer (by @zsdonghao in #588)
 - Error in `tl.layers.TernaryConv2d` fixed - self.inputs not defined (by @DEKHTIARJonathan in #658)
 - Deprecation warning fixed in `tl.layers.binary._compute_threshold()` (by @DEKHTIARJonathan in #658)

tensorlayer/layers/recurrent.py

@@ -763,9 +763,14 @@ def __init__(
         self.batch_size = batch_size
         outputs = []
         self.cell = cell = cell_fn(shape=cell_shape, filter_size=filter_size, num_features=feature_map)
+
         if initial_state is None:
-            self.initial_state = cell.zero_state(batch_size, dtype=LayersConfig.tf_dtype)  # dtype=tf.float32)  # 1.2.3
+            self.initial_state = cell.zero_state(batch_size, dtype=LayersConfig.tf_dtype)
+        else:
+            self.initial_state = initial_state
+
         state = self.initial_state
+
         # with tf.variable_scope("model", reuse=None, initializer=initializer):
         with tf.variable_scope(name, initializer=initializer) as vs:
             for time_step in range(n_steps):