Renames trainable to learnable.

Author: lucasb-eyer

DeepFried2/Module.py

@@ -41,13 +41,13 @@ def _addparam_optional(self, shape, init, *a, **kw):
 
 
     def zero_grad_parameters(self):
-        for p in self.parameters(trainable_only=True):
+        for p in self.parameters(learnable_only=True):
             p.zero_grad()
 
-    def parameters(self, trainable_only=False):
+    def parameters(self, learnable_only=False):
         params = getattr(self, '_params', [])
-        if trainable_only:
-            params = [p for p in params if p.trainable()]
+        if learnable_only:
+            params = [p for p in params if p.learnable()]
         return params
 
     def evaluate(self):
@@ -88,7 +88,7 @@ def accumulate_gradients(self, data_in, data_tgt, crit):
             symb_cost = crit(symb_out, symb_tgt)
             extra_out = self.get_extra_outputs() + crit.get_extra_outputs()
 
-            params = self.parameters(trainable_only=True)
+            params = self.parameters(learnable_only=True)
             symb_grads = df.th.grad(cost=symb_cost, wrt=[p.param for p in params])
             grads_updates = [(p.grad, p.grad + symb_grad) for p, symb_grad in zip(params, symb_grads)]
 

DeepFried2/Optimizer.py

@@ -12,7 +12,7 @@ def update_parameters(self, model):
         if model not in self.states:
             # TODO: Not only scalar, e.g. Adam might profit from integer t
             hyperparams = {name: df.T.scalar(name) for name in self.hyperparams}
-            params, grads = zip(*[(p.param, p.grad) for p in model.parameters(trainable_only=True)])
+            params, grads = zip(*[(p.param, p.grad) for p in model.parameters(learnable_only=True)])
             updates = self.get_updates(params, grads, **hyperparams)
             self.states[model] = df.th.function(
                 inputs=list(hyperparams.values()),

DeepFried2/Param.py

@@ -41,5 +41,5 @@ def zero_grad(self):
     def may_decay(self):
         return self.grad is not None and self.decay
 
-    def trainable(self):
+    def learnable(self):
         return self.grad is not None

DeepFried2/utils.py

@@ -37,8 +37,8 @@ def tensors_for_ndarrays(datas, basename):
     raise TypeError("I only understand lists or tuples of numpy arrays! (possibly nested)")
 
 
-def count_params(module, trainable_only=True):
-    return sum(p.get_value().size for p in module.parameters(trainable_only=trainable_only))
+def count_params(module, learnable_only=True):
+    return sum(p.get_value().size for p in module.parameters(learnable_only=learnable_only))
 
 
 def flatten(what, types=(list, tuple), none_to_empty=False):