Merge pull request #163 from danielhers/birnn_save

Fix #161: support save/load BiRNNBuilder

python/dynet.pyx

@@ -312,9 +312,14 @@ cdef class Model: # {{{
         elif isinstance(c, SimpleRNNBuilder):
             saver.add_srnn_builder((<CSimpleRNNBuilder*>(<SimpleRNNBuilder>c).thisptr)[0])
             fh.write("srnn_builder ")
+        elif isinstance(c, BiRNNBuilder):
+            fh.write("birnn_builder~%d " % (2 * len(c.builder_layers)))
+            for (f,b) in c.builder_layers:
+                self._save_one(f,saver,fh,pfh)
+                self._save_one(b,saver,fh,pfh)
         elif isinstance(c, Saveable):
             cs = c.get_components()
-            fh.write("user~%s " % len(cs))
+            fh.write("user~%d " % len(cs))
             pickle.dump(c,pfh)
             for subc in cs:
                 self._save_one(subc,saver,fh,pfh)
@@ -362,6 +367,11 @@ cdef class Model: # {{{
             sb_ = SimpleRNNBuilder(0,0,0,self) # empty builder
             loader.fill_srnn_builder((<CSimpleRNNBuilder *>sb_.thisptr)[0])
             return sb_
+        elif tp.startswith("birnn_builder~"):
+            tp,num = tp.split("~",1)
+            num = int(num)
+            items = [self._load_one(itypes, loader, pfh) for _ in xrange(num)]
+            return BiRNNBuilder(None, None, None, None, None, zip(items[0::2], items[1::2]))
         elif tp.startswith("user~"):
             # user defiend type
             tp,num = tp.split("~",1)
@@ -1126,24 +1136,28 @@ class BiRNNBuilder(object):
         builder = BiRNNBuilder(1, 128, 100, model, LSTMBuilder)
         [o1,o2,o3] = builder.transduce([i1,i2,i3])
     """
-    def __init__(self, num_layers, input_dim, hidden_dim, model, rnn_builder_factory):
+    def __init__(self, num_layers, input_dim, hidden_dim, model, rnn_builder_factory, builder_layers=None):
         """
         @param num_layers: depth of the BiRNN
         @param input_dim: size of the inputs
         @param hidden_dim: size of the outputs (and intermediate layer representations)
         @param model
         @param rnn_builder_factory: RNNBuilder subclass, e.g. LSTMBuilder
+        @param builder_layers: list of (forward, backward) pairs of RNNBuilder instances to directly initialize layers
         """
-        assert num_layers > 0
-        assert hidden_dim % 2 == 0
-        self.builder_layers = []
-        f = rnn_builder_factory(1, input_dim, hidden_dim/2, model)
-        b = rnn_builder_factory(1, input_dim, hidden_dim/2, model)
-        self.builder_layers.append((f,b))
-        for _ in xrange(num_layers-1):
-            f = rnn_builder_factory(1, hidden_dim, hidden_dim/2, model)
-            b = rnn_builder_factory(1, hidden_dim, hidden_dim/2, model)
+        if builder_layers is None:
+            assert num_layers > 0
+            assert hidden_dim % 2 == 0
+            self.builder_layers = []
+            f = rnn_builder_factory(1, input_dim, hidden_dim/2, model)
+            b = rnn_builder_factory(1, input_dim, hidden_dim/2, model)
             self.builder_layers.append((f,b))
+            for _ in xrange(num_layers-1):
+                f = rnn_builder_factory(1, hidden_dim, hidden_dim/2, model)
+                b = rnn_builder_factory(1, hidden_dim, hidden_dim/2, model)
+                self.builder_layers.append((f,b))
+        else:
+            self.builder_layers = builder_layers
 
     def whoami(self): return "BiRNNBuilder"