refactor(pynet): 修复PyNet库和示例的路径引用

zjZSTU · zjZSTU · commit 0608f3571682 · 2019-06-28T15:17:34.000+08:00
diff --git a/examples/2_nn_xor.py b/examples/2_nn_xor.py
@@ -3,9 +3,8 @@
 # @Time    : 19-5-27 下午6:39
 # @Author  : zj
 
-import pynet.models.utils
-import pynet.vision.data
 from pynet import nn, models, vision
+from pynet.vision.data import xor
 import numpy as np
 import os
 
@@ -17,7 +16,7 @@
 def two_layer_train():
     net = models.two_layer_net(num_in=2, num_hidden=6, num_out=2)
 
-    input_array, xor_array = pynet.vision.data.load_xor()
+    input_array, xor_array = xor.load_xor()
 
     criterion = nn.CrossEntropyLoss()
 
@@ -51,16 +50,16 @@ def two_layer_train():
         params = net.get_params()
         print('FC1: {}'.format(params['fc1']))
         print('FC2: {}'.format(params['fc2']))
-        pynet.models.utils.save_params(params, path=os.path.join(os.getcwd(), 'two_layer_net.pkl'))
+        # models.utils.save_params(params, path=os.path.join('/home/zj/deeplearning/pkl/', 'two_layer_net.pkl'))
 
 
 def two_layer_test():
-    params = pynet.models.utils.load_params(os.path.join(os.getcwd(), 'two_layer_net.pkl'))
+    params = models.utils.load_params(os.path.join(os.getcwd(), 'two_layer_net.pkl'))
     print(params)
     net = models.two_layer_net(num_in=2, num_hidden=6, num_out=2)
     net.set_params(params)
 
-    input_array, xor_array = pynet.vision.data.load_xor()
+    input_array, xor_array = xor.load_xor()
 
     for item in input_array:
         print(net.forward(np.atleast_2d(item)))
diff --git a/examples/3_nn_cifar10.py b/examples/3_nn_cifar10.py
@@ -4,8 +4,8 @@
 # @Author  : zj
 
 
-from pynet import nn, models
-import pynet.vision.data
+from pynet import nn, models, vision
+from pynet.vision.data import cifar
 import time
 import numpy as np
 
@@ -31,7 +31,7 @@
 data_path = '/home/lab305/Documents/data/decompress_cifar_10'
 
 if __name__ == '__main__':
-    x_train, x_test, y_train, y_test = pynet.vision.data.load_cifar10(data_path, shuffle=True, is_flatten=True)
+    x_train, x_test, y_train, y_test = cifar.load_cifar10(data_path, shuffle=True, is_flatten=True)
 
     x_train = x_train / 255 - 0.5
     x_test = x_test / 255 - 0.5
diff --git a/examples/3_nn_iris.py b/examples/3_nn_iris.py
@@ -4,9 +4,9 @@
 # @Author  : zj
 
 import os
-import pynet.vision.data
+from pynet.vision.data import iris
 import pynet.models.utils as utils
-from pynet import nn, models
+from pynet import nn, models, vision
 
 """
 iris分类
@@ -34,7 +34,7 @@
 iris_path = '/home/lab305/Documents/data/iris-species/Iris.csv'
 
 if __name__ == '__main__':
-    x_train, x_test, y_train, y_test = pynet.vision.data.load_iris(iris_path, shuffle=True, tsize=0.8)
+    x_train, x_test, y_train, y_test = iris.load_iris(iris_path, shuffle=True, tsize=0.8)
 
     net = models.three_layer_net(num_in=D, num_h1=H1, num_h2=H2, num_out=K)
     criterion = nn.CrossEntropyLoss()
@@ -63,7 +63,7 @@
             train_list.append(train_accuracy)
             test_list.append(test_accuracy)
             if train_accuracy >= 0.9999 and test_accuracy >= 0.9999:
-                utils.save_params(net.get_params(), path=os.path.join(os.getcwd(), 'three_layer_net_iris.pkl'))
+                # utils.save_params(net.get_params(), path=os.path.join(os.getcwd(), 'three_layer_net_iris.pkl'))
                 break
         if (i % 10000) == 9999:
             # 每隔10000次降低学习率
diff --git a/examples/3_nn_orl.py b/examples/3_nn_orl.py
@@ -6,9 +6,9 @@
 import numpy as np
 import time
 
-import pynet.vision.data
+from pynet.vision.data import orl
 import pynet.models.utils as utils
-from pynet import nn, models
+from pynet import nn, models, vision
 
 # 批量大小
 batch_size = 4
@@ -28,7 +28,7 @@
 data_path = '/home/lab305/Documents/data/att_faces_png'
 
 if __name__ == '__main__':
-    x_train, x_test, y_train, y_test = pynet.vision.data.load_orl(data_path, shuffle=True)
+    x_train, x_test, y_train, y_test = orl.load_orl(data_path, shuffle=True)
 
     x_train = x_train / 255 - 0.5
     x_test = x_test / 255 - 0.5
@@ -71,7 +71,7 @@
             test_accuracy = accuracy.compute_v2(x_test, y_test, net, batch_size=batch_size)
             if best_test_accuracy < test_accuracy:
                 best_test_accuracy = test_accuracy
-                utils.save_params(net.get_params(), path='three-dropout-nn-epochs-%d.pkl' % (i + 1))
+                # utils.save_params(net.get_params(), path='three-dropout-nn-epochs-%d.pkl' % (i + 1))
 
         print('best train accuracy: %.2f %%   best test accuracy: %.2f %%' % (
             best_train_accuracy * 100, best_test_accuracy * 100))
diff --git a/examples/lenet5_mnist.py b/examples/lenet5_mnist.py
@@ -3,9 +3,9 @@
 # @Time    : 19-5-27 下午6:40
 # @Author  : zj
 
-from pynet import nn, models
+from pynet import nn, models, vision
 import pynet.models.utils as utils
-import pynet.vision.data
+from pynet.vision.data import mnist
 import numpy as np
 import time
 
@@ -19,7 +19,7 @@
 
 
 def lenet5_train():
-    x_train, x_test, y_train, y_test = pynet.vision.data.load_mnist(data_path, shuffle=True)
+    x_train, x_test, y_train, y_test = mnist.load_mnist(data_path, shuffle=True)
 
     # 标准化
     x_train = x_train / 255.0 - 0.5
@@ -87,7 +87,7 @@ def lenet5_test(model_path):
     net = models.lenet5()
     net.set_params(params)
 
-    x_train, x_test, y_train, y_test = pynet.vision.data.load_mnist(data_path, shuffle=True)
+    x_train, x_test, y_train, y_test = mnist.load_mnist(data_path, shuffle=True)
 
     # 标准化
     x_train = x_train / 255.0 - 0.5
diff --git a/examples/nin_cifar10.py b/examples/nin_cifar10.py
@@ -3,9 +3,9 @@
 # @Time    : 19-6-21 下午2:45
 # @Author  : zj
 
-from pynet import nn, models
+from pynet import nn, models, vision
 import pynet.models.utils as utils
-import pynet.vision.data
+from pynet.vision.data import cifar
 import numpy as np
 import time
 
@@ -20,7 +20,7 @@
 
 
 def nin_train():
-    x_train, x_test, y_train, y_test = pynet.vision.data.load_cifar10(data_path, shuffle=True)
+    x_train, x_test, y_train, y_test = cifar.load_cifar10(data_path, shuffle=True)
 
     # 标准化
     x_train = x_train / 255.0 - 0.5
diff --git a/pynet/models/NIN.py b/pynet/models/NIN.py
@@ -4,7 +4,7 @@
 # @Author  : zj
 
 
-import nn
+import pynet.nn as nn
 from .Net import Net
 from .utils import load_params
 
diff --git a/pynet/models/__init__.py b/pynet/models/__init__.py
@@ -1,7 +1,7 @@
-# __all__ = ['TwoLayerNet', 'ThreeLayerNet', 'LeNet5', 'AlexNet']
-
 from .TwoLayerNet import *
 from .ThreeLayerNet import *
 from .LeNet5 import *
 from .AlexNet import *
 from .NIN import *
+
+from .utils import *
diff --git a/pynet/nn/GAP.py b/pynet/nn/GAP.py
@@ -5,7 +5,7 @@
 
 
 import numpy as np
-from pynet.nn import *
+from .Layer import *
 
 __all__ = ['GAP']
 
diff --git a/pynet/nn/__init__.py b/pynet/nn/__init__.py
@@ -1,6 +1,4 @@
-# __all__ = ['FC', 'ReLU', 'Dropout', 'Conv2d', 'MaxPool', 'CrossEntropyLoss']
-
-
+from .Layer import Layer
 from .FC import FC
 from .ReLU import ReLU
 from .Dropout import Dropout
diff --git a/pynet/vision/Accuracy.py b/pynet/vision/Accuracy.py
@@ -5,7 +5,6 @@
 
 
 import numpy as np
-import torch
 
 
 class Accuracy(object):
@@ -35,17 +34,3 @@ def compute_v2(self, data_array, labels_array, net, batch_size=128):
             num += 1
 
         return total_accuracy / num
-
-    def compute_pytorch(self, loader, net, device):
-        total_accuracy = 0
-        num = 0
-        for item in loader:
-            data, labels = item
-            data = data.to(device)
-            labels = labels.to(device)
-
-            scores = net.forward(data)
-            predicted = torch.argmax(scores, dim=1)
-            total_accuracy += torch.mean((predicted == labels).float()).item()
-            num += 1
-        return total_accuracy / num
diff --git a/pynet/vision/data/__init__.py b/pynet/vision/data/__init__.py
@@ -1,8 +1,6 @@
-# __all__ = ['xor', 'iris']
-
 
 from .xor import *
 from .iris import *
-from .cifar10 import *
+from .cifar import *
 from .orl import *
 from .mnist import *
diff --git a/pynet/vision/data/cifar.py b/pynet/vision/data/cifar.py
@@ -3,10 +3,6 @@
 # @Time    : 19-6-20 下午7:22
 # @Author  : zj
 
-import torchvision.transforms as transforms
-import torchvision.datasets as datasets
-from torch.utils.data import DataLoader
-
 import numpy as np
 import os
 from .utils import *
@@ -66,19 +62,3 @@ def load_cifar10(cifar10_path, shuffle=True, is_flatten=False):
                 x_train.append(np.transpose(img, (2, 0, 1)))
 
     return np.array(x_train), np.array(x_test), np.array(y_train), np.array(y_test)
-
-
-def load_cifar10_pytorch(cifar10_path, batch_size=128, shuffle=False):
-    transform = transforms.Compose([
-        transforms.Resize((227, 227)),
-        transforms.ToTensor(),
-        transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5))
-    ])
-
-    train_data_set = datasets.CIFAR10(root=cifar10_path, train=True, download=True, transform=transform)
-    test_data_set = datasets.CIFAR10(root=cifar10_path, train=False, download=True, transform=transform)
-
-    train_loader = DataLoader(train_data_set, batch_size=batch_size, shuffle=shuffle, num_workers=2)
-    test_loader = DataLoader(test_data_set, batch_size=batch_size, shuffle=shuffle, num_workers=2)
-
-    return train_loader, test_loader
diff --git a/pytorch/examples/nin_cifar10_pytorch.py b/pytorch/examples/nin_cifar10_pytorch.py
