fixed docstrings and kwargs

salad/solver/da/coral.py

@@ -138,9 +138,10 @@ def __init__(self, model, dataset, *args, **kwargs):
 
 class CentroidDistanceLossSolver(CorrelationDistanceSolver):
     """
-    Minimal Entropy Correlation Alignment for Unsupervised Domain Adaptation
-    Paper: https://openreview.net/pdf?id=rJWechg0Z
-    and: https://arxiv.org/pdf/1705.08180.pdf
+    Notes
+    -----
+
+    Needs work.
     """
 
     def __init__(self, model, dataset, *args, **kwargs):

salad/solver/gan.py

@@ -92,6 +92,11 @@ def derive_losses(self, batch):
 
 class ConditionalGANSolver(Solver):
 
+    """ Train a class conditional GAN model
+
+
+    """
+
     names     = ['D_GAN', 'D_CL', 'D_CON']
     n_classes = [1, 11, 3]
 
@@ -125,12 +130,12 @@ def _init_models(self, **kwargs):
             D.weight_init(mean=0.0, std=0.02)
             self.register_model(D, name)
 
-    def _init_optims(self, **kwargs):
-        opt = torch.optim.Adam(self.model.parameters(),lr = learningrate, betas = (.5,.999))
+    def _init_optims(self, lr = 2e-4, beta1 = .5, beta2 = .999, **kwargs):
+        opt = torch.optim.Adam(self.model.parameters(),lr = lr, betas = (beta1, beta2))
         self.register_optimizer(opt, CGANLoss(), name="Generator")
 
         for D in self.discriminators:
-            optim = torch.optim.Adam(D.parameters(), lr = learningrate, betas = (.5,.999))
+            optim = torch.optim.Adam(D.parameters(), lr = lr, betas = (beta1, beta2))
             self.register_optim(optim, CGANLoss())
 
     def format_train_report(self, losses):