From 81f6e72da02239f1f9ead5b67433bfc18ad3d367 Mon Sep 17 00:00:00 2001
From: Connor Olding <cloningdonor@gmail.com>
Date: Thu, 16 Feb 2017 19:21:24 -0800
Subject: [PATCH] .

---
 optim_nn.py      | 25 ++++++++++++++++++++++---
 optim_nn_core.py | 19 -------------------
 2 files changed, 22 insertions(+), 22 deletions(-)

diff --git a/optim_nn.py b/optim_nn.py
index 3a4d330..b445227 100644
--- a/optim_nn.py
+++ b/optim_nn.py
@@ -1,11 +1,11 @@
 #!/usr/bin/env python3
 
-# BIG TODO: ensure numpy isn't upcasting to float64 *anywhere*.
-#           this is gonna take some work.
-
 # external packages required for full functionality:
 # numpy scipy h5py sklearn dotmap
 
+# BIG TODO: ensure numpy isn't upcasting to float64 *anywhere*.
+#           this is gonna take some work.
+
 from optim_nn_core import *
 from optim_nn_core import _check, _f
 
@@ -48,6 +48,25 @@ class SomethingElse(ResidualLoss):
 
 # Parametric Layers {{{1
 
+class DenseOneLess(Dense):
+    def init(self, W, dW):
+        super().init(W, dW)
+        ins, outs = self.input_shape[0], self.output_shape[0]
+        assert ins == outs, (ins, outs)
+
+    def F(self, X):
+        np.fill_diagonal(self.coeffs, 0)
+        self.X = X
+        Y = X.dot(self.coeffs) + self.biases
+        return Y
+
+    def dF(self, dY):
+        dX = dY.dot(self.coeffs.T)
+        self.dcoeffs[:] = self.X.T.dot(dY)
+        self.dbiases[:] = dY.sum(0, keepdims=True)
+        np.fill_diagonal(self.dcoeffs, 0)
+        return dX
+
 class LayerNorm(Layer):
     # paper: https://arxiv.org/abs/1607.06450
     # my implementation may be incorrect.
diff --git a/optim_nn_core.py b/optim_nn_core.py
index 3beacc6..3a42042 100644
--- a/optim_nn_core.py
+++ b/optim_nn_core.py
@@ -455,25 +455,6 @@ class Dense(Layer):
         self.dbiases[:] = dY.sum(0, keepdims=True)
         return dX
 
-class DenseOneLess(Dense):
-    def init(self, W, dW):
-        super().init(W, dW)
-        ins, outs = self.input_shape[0], self.output_shape[0]
-        assert ins == outs, (ins, outs)
-
-    def F(self, X):
-        np.fill_diagonal(self.coeffs, 0)
-        self.X = X
-        Y = X.dot(self.coeffs) + self.biases
-        return Y
-
-    def dF(self, dY):
-        dX = dY.dot(self.coeffs.T)
-        self.dcoeffs[:] = self.X.T.dot(dY)
-        self.dbiases[:] = dY.sum(0, keepdims=True)
-        np.fill_diagonal(self.dcoeffs, 0)
-        return dX
-
 # Models {{{1
 
 class Model: