modify : Pretraining_VGG_from_scratch.rst

apply suggestion

pull request #2971

Author: woongjoonchoi

beginner_source/Pretraining_Vgg_from_scratch.rst

@@ -1,23 +1,4 @@
 
-.. DO NOT EDIT.
-.. THIS FILE WAS AUTOMATICALLY GENERATED BY SPHINX-GALLERY.
-.. TO MAKE CHANGES, EDIT THE SOURCE PYTHON FILE:
-.. "beginner/Pretraining_Vgg_from_scratch.py"
-.. LINE NUMBERS ARE GIVEN BELOW.
-
-.. only:: html
-
-    .. note::
-        :class: sphx-glr-download-link-note
-
-        Click :ref:`here <sphx_glr_download_beginner_Pretraining_Vgg_from_scratch.py>`
-        to download the full example code
-
-.. rst-class:: sphx-glr-example-title
-
-.. _sphx_glr_beginner_Pretraining_Vgg_from_scratch.py:
-
-
 Pre-training VGG from scratch
 ============================
 
@@ -73,7 +54,7 @@ VGG within the training time suggested in the paper.
 Setup
 --------
 
-.. note:: if you are running this in Google Colab, install ``albumentations`` by running:
+.. note:: If you are running this in Google Colab, install ``albumentations`` by running:
 
    .. code-block:: python
    
@@ -82,7 +63,6 @@ Setup
 
 First, let's import the required dependencies:
 
-.. GENERATED FROM PYTHON SOURCE LINES 67-92
 
 .. code-block:: default
 
@@ -110,22 +90,6 @@ First, let's import the required dependencies:
 
     device = 'cuda' if torch.cuda.is_available() else 'cpu'
 
-
-
-
-
-
-.. rst-class:: sphx-glr-script-out
-
- .. code-block:: none
-
-    albumentations are already installed
-
-
-
-
-.. GENERATED FROM PYTHON SOURCE LINES 93-100
-
 VGG Configuration
 -----------------
 
@@ -134,7 +98,7 @@ We use the CIFAR100 dataset. The authors of the VGG paper scale images ``isotrop
 which means increasing the size of an image while maintaining its proportions,
 preventing distortion and maintaining the consistency of the object.
 
-.. GENERATED FROM PYTHON SOURCE LINES 100-140
+
 
 .. code-block:: default
 
@@ -185,7 +149,7 @@ preventing distortion and maintaining the consistency of the object.
 
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 141-147
+
 
 .. note:: In the code above, we have defined the batch size as 32,
    which is recommended for Google Colab. However, if you are
@@ -194,7 +158,7 @@ preventing distortion and maintaining the consistency of the object.
    size according to your preference and hardware capabilities.
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 149-174
+
 
 Defining the dataset
 --------------------
@@ -222,7 +186,7 @@ To apply preprocessing, we need to override the CIFAR100 class that we have impo
 ``torchvision.datasets`` with a custom class:
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 174-227
+
 
 .. code-block:: default
 
@@ -286,7 +250,7 @@ To apply preprocessing, we need to override the CIFAR100 class that we have impo
 
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 228-238
+
 
 Define Model
 ------------
@@ -299,7 +263,7 @@ We will use two main components to define the model:
 * ``Config_channels``: This refers to the number of output channels for each layer.
 * ``Config_kernels``: This refers to the kernel size (or filter size) for each layer.
 
-.. GENERATED FROM PYTHON SOURCE LINES 238-266
+
 
 .. code-block:: default
 
@@ -338,12 +302,10 @@ We will use two main components to define the model:
 
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 267-269
 
-Next, we define a model class that generates a model with a choice of six versions.
 
+Next, we define a model class that generates a model with a choice of six versions.
 
-.. GENERATED FROM PYTHON SOURCE LINES 269-363
 
 .. code-block:: default
 
@@ -448,7 +410,7 @@ Next, we define a model class that generates a model with a choice of six versio
 
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 364-377
+
 
 Initializing Model Weights
 ----------------------------
@@ -464,7 +426,7 @@ to initialize the model weights. Specifically, we will apply Xavier
 initialization to the first few layers and the last few layers, while using
 random initialization for the remaining layers.
 
-.. GENERATED FROM PYTHON SOURCE LINES 377-399
+
 
 .. code-block:: default
 
@@ -497,15 +459,15 @@ random initialization for the remaining layers.
 
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 400-405
+
 
 Training the Model
 ------------------
 
 First, let's define top-k error.
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 405-422
+
 
 .. code-block:: default
 
@@ -533,13 +495,13 @@ First, let's define top-k error.
 
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 423-426
+
 
 Next, we initiate the model and loss function, optimizer and schedulers. In the VGG model,
 they use a softmax output, Momentum Optimizer, and scheduling based on accuracy.
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 426-434
+
 
 .. code-block:: default
 
@@ -630,12 +592,12 @@ they use a softmax output, Momentum Optimizer, and scheduling based on accuracy.
 
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 435-437
+
 
 As mentioned above, we are using the ``CIFAR100`` dataset and set gradient
 clipping to 1.0 to prevent gradient exploding.
 
-.. GENERATED FROM PYTHON SOURCE LINES 437-570
+
 
 .. code-block:: default
 
@@ -3432,14 +3394,14 @@ clipping to 1.0 to prevent gradient exploding.
 
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 571-575
+
 
 (Optional) Additional Exercise: ImageNet
 --------------------------------------------
 
 You can apply the same model that we have trained above with another popular dataset called ImageNet:  
 
-.. GENERATED FROM PYTHON SOURCE LINES 575-644
+
 
 .. code-block:: default
 
@@ -3519,7 +3481,7 @@ You can apply the same model that we have trained above with another popular dat
 
 
 
-.. GENERATED FROM PYTHON SOURCE LINES 645-660
+
 
 Conclusion
 ----------