froog 0.3.1__tar.gz → 0.3.2__tar.gz

{froog-0.3.1 → froog-0.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: froog
-Version: 0.3.1
+Version: 0.3.2
 Summary: a beautifully simplistic tensor library
 Author: Kevin Buhler
 License: MIT

{froog-0.3.1 → froog-0.3.2}/froog/nn.py

@@ -10,8 +10,8 @@ from froog.tensor import Tensor
 import numpy as np
 
 def Linear(*x):
-  # TODO: why dividing by sqrt?
-  ret = np.random.uniform(-1., 1., size=x)/np.sqrt(np.prod(x)) # random init weights
+  # random Glorot initialization
+  ret = np.random.uniform(-1., 1., size=x)/np.sqrt(np.prod(x))
   return ret.astype(np.float32)
 
 def swish(x):
@@ -55,6 +55,6 @@ class BatchNorm2D:
   def __call__(self, x):
     x = x.sub(self.running_mean.reshape(shape=[1, -1, 1, 1]))
     x = x.mul(self.weight.reshape(shape=[1, -1, 1, 1]))
-    x = x.div(self.running_var.add(Tensor([self.eps], gpu=x.gpu)).reshape(shape=[1, -1, 1, 1]).sqrt()) # TODO: shouldn't div go first?
+    x = x.div(self.running_var.add(Tensor([self.eps], gpu=x.gpu)).reshape(shape=[1, -1, 1, 1]).sqrt())
     x = x.add(self.bias.reshape(shape=[1, -1, 1, 1]))
     return x

{froog-0.3.1 → froog-0.3.2}/froog/ops.py

@@ -340,7 +340,6 @@ class MaxPool2D(Function):
                             *ctx.kernel_size)
 register('max_pool2d', MaxPool2D)
 
-
 class AvgPool2D(Function):
   @staticmethod
   def forward(ctx, x, kernel_size=(2,2)):
@@ -351,7 +350,7 @@ class AvgPool2D(Function):
   @staticmethod
   def backward(ctx, grad_output):
     s, = ctx.saved_tensors
-    py, px = ctx.kernel_size                                  # TODO: where does kernel_size come from?
+    py, px = ctx.kernel_size                                  # kernel_size passed from forward context
     my, mx = (s[2]//py)*py, (s[3]//px)*px
     ret = np.zeros(s, dtype=grad_output.dtype)
     for Y in range(py):

{froog-0.3.1 → froog-0.3.2}/froog/ops_gpu.py

@@ -5,6 +5,8 @@
 # |    ___||    __  ||  |_|  ||  |_|  ||   ||  |
 # |   |    |   |  | ||       ||       ||   |_| |
 # |___|    |___|  |_||_______||_______||_______|
+#
+# OpenCL kernels
 
 import numpy as np
 from .tensor import Function, register
@@ -71,7 +73,6 @@ def unary_op(ctx, code, x):
   prg.unop(ctx.cl_queue, [np.prod(ret.shape)], None, x, ret)
   return ret
 
-# ???
 @functools.lru_cache
 def cl_pooling_krnl_build(cl_ctx, iter_op, result_op, init_val=0):
   prg = """

{froog-0.3.1 → froog-0.3.2}/froog/tensor.py

@@ -62,7 +62,7 @@ class Tensor:
       self.gpu = False
 
     self.data = data
-    self.grad = None # TODO: why self.grad.data instead of self.grad?
+    self.grad = None
 
     if gpu:
       self.gpu_()

{froog-0.3.1 → froog-0.3.2}/froog/utils.py

@@ -67,7 +67,8 @@ def im2col(x, H, W):
   tx = x.reshape(bs, -1)[:, idx]
 
   # all the time is spent here
-  tx = tx.ravel() # TODO: whats the purpose of ravel ???
+  # np.ravel() flattens the array into a 1-dimensional shape
+  tx = tx.ravel()
   return tx.reshape(-1, cin*W*H)
 
 def col2im(tx, H, W, OY, OX):

{froog-0.3.1 → froog-0.3.2}/froog.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: froog
-Version: 0.3.1
+Version: 0.3.2
 Summary: a beautifully simplistic tensor library
 Author: Kevin Buhler
 License: MIT

{froog-0.3.1 → froog-0.3.2}/setup.py

@@ -10,7 +10,7 @@ with open(os.path.join(directory, 'README.md'), encoding='utf-8') as f:
   long_description = f.read()
 
 setup(name='froog',
-      version='0.3.1',
+      version='0.3.2',
       description='a beautifully simplistic tensor library',
       author='Kevin Buhler',
       license='MIT',

{froog-0.3.1 → froog-0.3.2}/tests/test_models.py

@@ -16,7 +16,6 @@ X_train, Y_train, X_test, Y_test = fetch_mnist()
 class SimpleMLP:
   def __init__(self):
     # 784 pixel inputs -> 128 -> 10 output
-    # TODO: why down to 128?
     self.l1 = Tensor(Linear(784, 128))
     self.l2 = Tensor(Linear(128, 10))
 
@@ -73,7 +72,7 @@ def train(model, optimizer, steps, BS=128, gpu=False):
     model_outputs = model.forward(x)
 
     # ********* backward pass *********
-    loss = model_outputs.mul(y).mean() # TODO: what exactly is NLL loss function?
+    loss = model_outputs.mul(y).mean()
     loss.backward()
     optimizer.step()