李宏毅hw-9:Explainable ML

——欲速则不达，我已经很幸运了，只要珍惜这份幸运就好了，不必患得患失，慢慢来。

----查漏补缺：

1.关于这个os.listdir的使用

2.从‘num_文件名.jpg’中提取出数值：

3.slic图像分割标记函数的作用：

4.zip这个函数，用来讲2个数组“一一对应”的合成1个数组：

5.关于astype的这个用来类型转换的东西：

6.关于 利用[]合并之后，再进行enumerate:

PART1:11个food的分类问题的explainable部分：

一、对于这个cnn的代码部分的回顾:

1.前期准备：库的引入，参数的设置

import os
import sys
import argparse
import numpy as np
from PIL import Image
import matplotlib.pyplot as plt
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Adam
from torch.utils.data import Dataset
import torchvision.transforms as transforms
from skimage.segmentation import slic
from lime import lime_image
from pdb import set_trace
from torch.autograd import Variable

args = {
      'ckptpath': './checkpoint.pth',
      'dataset_dir': './food/'
}
args = argparse.Namespace(**args)

2.模型结构的定义：

（1）cnn是一系列的卷积层最终得到4**4*512的图像

（2）flatten展平之后，再经过一系列的linear层得到11种的向量

# Model definition——分析这个model的结构：

class Classifier(nn.Module):
  def __init__(self):
    super(Classifier, self).__init__()

    def building_block(indim, outdim):
      return [
        nn.Conv2d(indim, outdim, 3, 1, 1),
        nn.BatchNorm2d(outdim),
        nn.ReLU(),
      ]
    def stack_blocks(indim, outdim, block_num):
      layers = building_block(indim, outdim)
      for i in range(block_num - 1):
        layers += building_block(outdim, outdim)
      layers.append(nn.MaxPool2d(2, 2, 0))
      return layers

    cnn_list = []
    cnn_list += stack_blocks(3, 128, 3)
    cnn_list += stack_blocks(128, 128, 3)
    cnn_list += stack_blocks(128, 256, 3)
    cnn_list += stack_blocks(256, 512, 1)
    cnn_list += stack_blocks(512, 512, 1)
    self.cnn = nn.Sequential( * cnn_list)   #上面所有的函数，都是为了这个cnn的过程的设计

    dnn_list = [
      nn.Linear(512 * 4 * 4, 1024),
      nn.ReLU(),
      nn.Dropout(p = 0.3),
      nn.Linear(1024, 11),
    ]
    self.fc = nn.Sequential( * dnn_list)

  def forward(self, x):
    out = self.cnn(x)
    out = out.reshape(out.size()[0], -1)
    return self.fc(out)

模型对象的实例化：

# Load trained model
model = Classifier().cuda()
checkpoint = torch.load(args.ckptpath)
model.load_state_dict(checkpoint['model_state_dict'])
# It should display:  

3.定义food_dataset，虽然实例的部分使用eval不是很确定是不是已经把model已经train好了，还是说，只是使用eval版本的eval：

# It might take some time, if it is too long, try to reload it.
# Dataset definition
#定义这个dataset了
class FoodDataset(Dataset):
    def __init__(self, paths, labels, mode):
        # mode: 'train' or 'eval'
        
        self.paths = paths
        self.labels = labels
        trainTransform = transforms.Compose([
            transforms.Resize(size=(128, 128)),
            transforms.RandomHorizontalFlip(),
            transforms.RandomRotation(15),
            transforms.ToTensor(),
        ])
        evalTransform = transforms.Compose([
            transforms.Resize(size=(128, 128)),
            transforms.ToTensor(),
        ])
        self.transform = trainTransform if mode == 'train' else evalTransform

    # pytorch dataset class
    def __len__(self):
        return len(self.paths)

    def __getitem__(self, index):
        X = Image.open(self.paths[index])
        X = self.transform(X)
        Y = self.labels[index]
        return X, Y

    # help to get images for visualizing
    def getbatch(self, indices):
        images = []
        labels = []
        for index in indices:
          image, label = self.__getitem__(index)
          images.append(image)
          labels.append(label)
        return torch.stack(images), torch.tensor(labels)

# help to get data path and label
#先分析这个函数，再分析上面的dataset
def get_paths_labels(path):
    #定义1个lambda函数
    def my_key(name):
      return int(name.replace(".jpg",""))+1000000*int(name.split("_")[0])
    imgnames = os.listdir(path)
    imgnames.sort(key=my_key) #使用这个lambda函数进行sort排序
    imgpaths = []
    labels = []
    for name in imgnames:
        imgpaths.append(os.path.join(path, name))
        labels.append(int(name.split('_')[0]))
    return imgpaths, labels
train_paths, train_labels = get_paths_labels(args.dataset_dir)  #没问题，只是key处理了，但是name本身没改变

train_set = FoodDataset(train_paths, train_labels, mode='eval') #可能这里用到的model是已经train好的model

从这个dataset中抽出11张图像进行人工观察：

img_indices = [i for i in range(10)]
images, labels = train_set.getbatch(img_indices)
fig, axs = plt.subplots(1, len(img_indices), figsize=(15, 8))
for i, img in enumerate(images):
  axs[i].imshow(img.cpu().permute(1, 2, 0))
# print(labels)

二、使用Lime对图像中的

1.Local Interpretable Model-Agnostic Explanations的定义：

2.具体使用这个lime：

#调用model的eval对整个batch的input进行预测得到1个batch的predicts结果
def predict(input):
    # input: numpy array, (batches, height, width, channels)                                                                                                                                                     
    
    model.eval()                                                                                                                                                             
    input = torch.FloatTensor(input).permute(0, 3, 1, 2)                                                                                                            
    # pytorch tensor, (batches, channels, height, width)

    output = model(input.cuda())                                                                                                                                             
    return output.detach().cpu().numpy()                                                                                                                              

#对输入的图像进行分割后标记                                                                                                                                                                  
def segmentation(input):
    # split the image into 200 pieces with the help of segmentaion from skimage                                                                                                                   
    return slic(input, n_segments=200, compactness=1, sigma=1)                                                                                                              
                                                                                                                                                                             

#设置画布参数
fig, axs = plt.subplots(1, len(img_indices), figsize=(15, 8))                                                                                                                                                                 
# fix the random seed to make it reproducible
np.random.seed(16)    



for idx, (image, label) in enumerate(zip(images.permute(0, 2, 3, 1).numpy(), labels)):                                                                                                                                             
    x = image.astype(np.double)
    # numpy array for lime

    #调用explainer的explain_instance，传递对应图像x，predict函数，segmentation函数作为它的参数
    explainer = lime_image.LimeImageExplainer()                                                                                                                              
    explaination = explainer.explain_instance(image=x, classifier_fn=predict, segmentation_fn=segmentation)

    # doc: https://lime-ml.readthedocs.io/en/latest/lime.html?highlight=explain_instance#lime.lime_image.LimeImageExplainer.explain_instance
    
    #调用上面的这个explaination,传递的参数主要是label值 和 num_features种类，其他的就是说是否显示不是重要的地方等。。
    lime_img, mask = explaination.get_image_and_mask(                                                                                                                         
                                label=label.item(),                                                                                                                           
                                positive_only=False,                                                                                                                         
                                hide_rest=False,                                                                                                                             
                                num_features=11,                                                                                                                              
                                min_weight=0.05                                                                                                                              
                            )
    # turn the result from explainer to the image
    # doc: https://lime-ml.readthedocs.io/en/latest/lime.html?highlight=get_image_and_mask#lime.lime_image.ImageExplanation.get_image_and_mask
    
    axs[idx].imshow(lime_img) #axs的第idx位置的图像，就放置这个lime_img了

#show出这些用lime标记的图像咯
plt.show()
plt.close()

三、使用saliency map：显著性标注出这个图像中贡献这个类型特征最多的地方

(其实就是普通的gradient的方法)

The heatmaps that highlight pixels of the input image that contribute the most in the classification task.

总的来说，就是通过计算每个pixel对于整个loss的gradient，这个gradient就是新的图像的pixel数值

#对图像中的每个pixel的数值进行normalize
def normalize(image):
  return (image - image.min()) / (image.max() - image.min())
  # return torch.log(image)/torch.log(image.max())

#用于计算saliency的函数
def compute_saliency_maps(x, y, model): #x就是图像， y就是label， model就是分类器
  model.eval()
  x = x.cuda()

  # we want the gradient of the input x
  x.requires_grad_()
  
  y_pred = model(x)
  loss_func = torch.nn.CrossEntropyLoss()
  loss = loss_func(y_pred, y.cuda())
  loss.backward()

  # saliencies = x.grad.abs().detach().cpu()
  saliencies, _ = torch.max(x.grad.data.abs().detach().cpu(),dim=1) #这一步，就是将每个像素的位置的gradient梯度（3个通道中的最大的那个）作为新的图像位置的 像素值

  # We need to normalize each image, because their gradients might vary in scale, but we only care about the relation in each image
  saliencies = torch.stack([normalize(item) for item in saliencies])
  return saliencies

# images, labels = train_set.getbatch(img_indices)
saliencies = compute_saliency_maps(images, labels, model)

# visualize
fig, axs = plt.subplots(2, len(img_indices), figsize=(15, 8))
for row, target in enumerate([images, saliencies]):
  for column, img in enumerate(target):
    if row==0:
      axs[row][column].imshow(img.permute(1, 2, 0).numpy()) #第一行：正常图像显示
      # What is permute?
      # In pytorch, the meaning of each dimension of image tensor is (channels, height, width)
      # In matplotlib, the meaning of each dimension of image tensor is (height, width, channels)
      # permute is a tool for permuting dimensions of tensors
      # For example, img.permute(1, 2, 0) means that,
      # - 0 dimension is the 1 dimension of the original tensor, which is height
      # - 1 dimension is the 2 dimension of the original tensor, which is width
      # - 2 dimension is the 0 dimension of the original tensor, which is channels
    else:
      axs[row][column].imshow(img.numpy(), cmap=plt.cm.hot)   #第二行：热成像图
    
plt.show()
plt.close()

四、smooth grad的方法查看heat 图像

Smooth grad

Smooth grad 的方法是，在圖片中隨機地加入 noise，然後得到不同的 heatmap，把這些 heatmap 平均起來就得到一個比較能抵抗 noisy gradient 的結果。

# Smooth grad
#一样的normalize函数
def normalize(image):
  return (image - image.min()) / (image.max() - image.min())

#计算出类似于saliency map中的saliencies图像的东西:
def smooth_grad(x, y, model, epoch, param_sigma_multiplier):         #总共epoch数，一个常量sigma
  model.eval()
  #x = x.cuda().unsqueeze(0)

  mean = 0
  sigma = param_sigma_multiplier / (torch.max(x) - torch.min(x)).item()  #sigma就是1个数值
  smooth = np.zeros(x.cuda().unsqueeze(0).size())               #一个和x相同大小zero变量
  for i in range(epoch):
    # call Variable to generate random noise
    noise = Variable(x.data.new(x.size()).normal_(mean, sigma**2))     #sigma用作正太分布的标准差参数，抽取noise的抽样,和x一样大
    x_mod = (x+noise).unsqueeze(0).cuda()
    x_mod.requires_grad_()

    y_pred = model(x_mod)
    loss_func = torch.nn.CrossEntropyLoss()
    loss = loss_func(y_pred, y.cuda().unsqueeze(0))
    loss.backward()

    # like the method in saliency map
    smooth += x_mod.grad.abs().detach().cpu().data.numpy()      #smooth用于累计每一个epoch的和
  smooth = normalize(smooth / epoch) # don't forget to normalize，取个均值就可以了
  # smooth = smooth / epoch
  return smooth


# images, labels = train_set.getbatch(img_indices)
smooth = []
for i, l in zip(images, labels):
  smooth.append(smooth_grad(i, l, model, 500, 0.4))
smooth = np.stack(smooth)
print(smooth.shape)

fig, axs = plt.subplots(2, len(img_indices), figsize=(15, 8)) #2行喔！
for row, target in enumerate([images, smooth]):
  for column, img in enumerate(target):
    axs[row][column].imshow(np.transpose(img.reshape(3,128,128), (1,2,0)))

五、Filter Explanation,透过卷积的中间层进行观察：

1.hook钩子函数的作用：

2.只输出指定filterid的那个滤波器的输出：

3.具体的代码部分

#定义正规化
def normalize(image):
  return (image - image.min()) / (image.max() - image.min())

layer_activations = None
#filter的观察函数，返回的是 activation 和 visulization
def filter_explanation(x, model, cnnid, filterid, iteration=100, lr=1):#cnnid是对应的卷积层的id，filterid是对应的过滤器的id
  # x: input image
  # cnnid, filterid: cnn layer id, which filter
  model.eval()

  def hook(model, input, output): #定义hook函数，就是将output给到全局的layer_activations
    global layer_activations
    layer_activations = output
  
  hook_handle = model.cnn[cnnid].register_forward_hook(hook)  #hook的handle句柄，下面有解释这行代码的含义
  # When the model forward through the layer[cnnid], need to call the hook function first
  # The hook function save the output of the layer[cnnid]
  # After forwarding, we'll have the loss and the layer activation

  # Filter activation: x passing the filter will generate the activation map
  model(x.cuda()) # forward

  # Based on the filterid given by the function argument, pick up the specific filter's activation map
  # We just need to plot it, so we can detach from graph and save as cpu tensor
  filter_activations = layer_activations[:, filterid, :, :].detach().cpu()
  
  # Filter visualization: find the image that can activate the filter the most
  x = x.cuda()
  x.requires_grad_()
  # input image gradient
  optimizer = Adam([x], lr=lr)
  # Use optimizer to modify the input image to amplify filter activation
  for iter in range(iteration): #iteration==100
    optimizer.zero_grad()
    model(x)
    
    objective = -layer_activations[:, filterid, :, :].sum()
    # We want to maximize the filter activation's summation
    # So we add a negative sign
    
    objective.backward()
    # Calculate the partial differential value of filter activation to input image
    optimizer.step()
    # Modify input image to maximize filter activation
  filter_visualizations = x.detach().cpu().squeeze()

  # Don't forget to remove the hook
  hook_handle.remove()
  # The hook will exist after the model register it, so you have to remove it after used
  # Just register a new hook if you want to use it

  return filter_activations, filter_visualizations

images, labels = train_set.getbatch(img_indices)
#下面的这个函数的参数可以看出，是获取第cnnid==6第6个卷积层的第0个过滤器的activation和visulization
filter_activations, filter_visualizations = filter_explanation(images, model, cnnid=6, filterid=0, iteration=100, lr=0.1)


#以下总共进行了3组图片的绘制，分别是原始图片、activation图片，visulation图片
fig, axs = plt.subplots(3, len(img_indices), figsize=(15, 8))
for i, img in enumerate(images):
  axs[0][i].imshow(img.permute(1, 2, 0))
# Plot filter activations
for i, img in enumerate(filter_activations):
  axs[1][i].imshow(normalize(img))
# Plot filter visualization
for i, img in enumerate(filter_visualizations):
  axs[2][i].imshow(normalize(img.permute(1, 2, 0)))
plt.show()
plt.close()

# 從下面四張圖可以看到，activate 的區域對應到一些物品的邊界，尤其是顏色對比較深的邊界

images, labels = train_set.getbatch(img_indices)
#下面的这个函数的参数可以看出，是获取第cnnid==23第23个卷积层的第0个过滤器的activation和visulization
filter_activations, filter_visualizations = filter_explanation(images, model, cnnid=23, filterid=0, iteration=100, lr=0.1)

# Plot filter activations
fig, axs = plt.subplots(3, len(img_indices), figsize=(15, 8))
for i, img in enumerate(images):
  axs[0][i].imshow(img.permute(1, 2, 0))
for i, img in enumerate(filter_activations):
  axs[1][i].imshow(normalize(img))
for i, img in enumerate(filter_visualizations):
  axs[2][i].imshow(normalize(img.permute(1, 2, 0)))
plt.show()
plt.close()

六、使用XAI中的Integrated gradient技术：

#什么都别说，5点45去西园吃点清淡的，就出去玩——看电影，或者其他的，好吧！

class IntegratedGradients():
    def __init__(self, model): #初始化这个类
        self.model = model
        self.gradients = None
        # Put model in evaluation mode
        self.model.eval()

    def generate_images_on_linear_path(self, input_image, steps):
        # Generate scaled xbar images
        xbar_list = [input_image*step/steps for step in range(steps)]
        return xbar_list

    def generate_gradients(self, input_image, target_class):  #计算一张图像的gradient
        # We want to get the gradients of the input image
        input_image.requires_grad=True
        # Forward
        model_output = self.model(input_image)
        # Zero grads
        self.model.zero_grad()
        # Target for backprop
        one_hot_output = torch.FloatTensor(1, model_output.size()[-1]).zero_().cuda()
        one_hot_output[0][target_class] = 1
        # Backward
        model_output.backward(gradient=one_hot_output)
        self.gradients = input_image.grad
        # Convert Pytorch variable to numpy array
        # [0] to get rid of the first channel (1,3,128,128)
        gradients_as_arr = self.gradients.data.cpu().numpy()[0]
        return gradients_as_arr

    def generate_integrated_gradients(self, input_image, target_class, steps): #计算img_list的图像的gradient的integrate
        # Generate xbar images
        xbar_list = self.generate_images_on_linear_path(input_image, steps)
        # Initialize an iamge composed of zeros
        integrated_grads = np.zeros(input_image.size())
        for xbar_image in xbar_list:
            # Generate gradients from xbar images
            single_integrated_grad = self.generate_gradients(xbar_image, target_class)
            # Add rescaled grads from xbar images
            integrated_grads = integrated_grads + single_integrated_grad/steps
        # [0] to get rid of the first channel (1,3,128,128)
        return integrated_grads[0]

def normalize(image):
  return (image - image.min()) / (image.max() - image.min())

# put the image to cuda
images, labels = train_set.getbatch(img_indices)
images = images.cuda()

IG = IntegratedGradients(model)
integrated_grads = []
for i, img in enumerate(images):
  img = img.unsqueeze(0)
  integrated_grads.append(IG.generate_integrated_gradients(img, labels[i], 10))
fig, axs = plt.subplots(2, len(img_indices), figsize=(15, 8))
for i, img in enumerate(images):      #输出一组正常的图像
  axs[0][i].imshow(img.cpu().permute(1, 2, 0))
for i, img in enumerate(integrated_grads): #输出integrate的图像
  axs[1][i].imshow(np.moveaxis(normalize(img),0,-1))
plt.show()
plt.close()

PART2:有关BERT的可解释行的model

（一）、在这个网站上感受bert的各个层的过程：

exBERT

这个模型可以用于查看注意力头部等信息，这里我就先不管了，后期慢慢摸索吧。。。。

（二）、visualizing bert's embedding：

湯姆有 3 個預訓練模型，但他忘記每一個模型是否有微調在閱讀理解的任務上了

通过观察各个token的embedding的位置，分析这个model是否具有阅读理解的fine_tune,

、。。。。我没做出来，有点难，

不过，它的代码就是从bert的每一层中取出embedding结果，再将每个token投射到二维坐标中进行分析

（三）、分析 吃的苹果 和 苹果手机的苹果词汇的embedding的距离

# Sentences for visualization
sentences = []
sentences += ["今天買了蘋果來吃"]
sentences += ["進口蘋果（富士)平均每公斤下跌12.3%"]
sentences += ["蘋果茶真難喝"]
sentences += ["老饕都知道智利的蘋果季節即將到來"]
sentences += ["進口蘋果因防止水分流失故添加人工果糖"]
sentences += ["蘋果即將於下月發振新款iPhone"]
sentences += ["蘋果獲新Face ID專利"]
sentences += ["今天買了蘋果手機"]
sentences += ["蘋果的股價又跌了"]
sentences += ["蘋果押寶指紋辨識技術"]

# Index of word selected for embedding comparison. E.g. For sentence "蘋果茶真難喝", if index is 0, "蘋 is selected"
select_word_index = [4, 2, 0, 8, 2, 0, 0, 4, 0, 0] #设置上面的词汇数组中的"苹果"二字的index位置

#计算向量a 和 向量b的欧式距离
def euclidean_distance(a, b):
    # Compute euclidean distance (L2 norm) between two numpy vectors a and b
    return np.linalg.norm(a-b)

#计算a向量和b向量的余弦相似度cosine_similarity = (A · B) / (||A|| * ||B||)
def cosine_similarity(a, b):
    # Compute cosine similarity between two numpy vectors a and b
    return 0

# Metric for comparison. Choose from euclidean_distance, cosine_similarity
#METRIC有2个选择，要么用欧式距离 要么用余弦相似度
METRIC = euclidean_distance

def get_select_embedding(output, tokenized_sentence, select_word_index):
    # The layer to visualize, choose from 0 to 12
    LAYER = 12
    # Get selected layer's hidden state
    hidden_state = output.hidden_states[LAYER][0]
    # Convert select_word_index in sentence to select_token_index in tokenized sentence
    select_token_index = tokenized_sentence.word_to_tokens(select_word_index).start
    # Return embedding of selected word
    return hidden_state[select_token_index].numpy()

# Tokenize and encode sentences into model's input format
tokenized_sentences = [tokenizer(sentence, return_tensors='pt') for sentence in sentences]

# Input encoded sentences into model and get outputs 
with torch.no_grad():
    outputs = [model(**tokenized_sentence) for tokenized_sentence in tokenized_sentences]

#得到词汇"苹果"在各个句子中的embedding
# Get embedding of selected word(s) in sentences. "embeddings" has shape (len(sentences), 768), where 768 is the dimension of BERT's hidden state
embeddings = [get_select_embedding(outputs[i], tokenized_sentences[i], select_word_index[i]) for i in range(len(outputs))]

#计算 对应 "苹果"二字的 词汇的距离
# Pairwse comparsion of sentences' embeddings using the metirc defined. "similarity_matrix" has shape [len(sentences), len(sentences)]
similarity_matrix = pairwise_distances(embeddings, metric=METRIC) 

#绘制这个词汇的距离
##### Plot the similarity matrix #####
plt.rcParams['figure.figsize'] = [12, 10] # Change figure size of the plot
plt.imshow(similarity_matrix) # Display an image in the plot
plt.colorbar() # Add colorbar to the plot
plt.yticks(ticks=range(len(sentences)), labels=sentences, fontproperties=myfont) # Set tick locations and labels (sentences) of y-axis
plt.title('Comparison of BERT Word Embeddings') # Add title to the plot
for (i,j), label in np.ndenumerate(similarity_matrix): # np.ndenumerate is 2D version of enumerate
    plt.text(i, j, '{:.2f}'.format(label), ha='center', va='center') # Add values in similarity_matrix to the corresponding position in the plot
plt.show() # Show the plot