实用篇 | T-SNE可视化工具详情及代码示例

本文主要是为了快速的了解t-sne和如何快速使用！ 

简要了解TSNE

TSNE，降维方法之一。降维在机器学习中非常重要。这是因为如果使用高维数据创建模型，则很容易欠拟合。换句话说，有太多无用的数据需要学习。可以通过从各种数据中仅选择最重要的数据在模型中使用它，也可以使用多个数据创建新数据并使其低维。无论如何，有必要将此类高维数据转换为低维数据。这称为降维。（还有其他方法可以创建要素，例如“特征消除”和“特征选择”。降维方法有两种类型：线性方法（主成分分析（PCA），独立成分分析，线性判别分析等）和非线性方法（歧管，自动编码器等）。TSNE是多种方法之一。
它从SNE（随机邻居嵌入）演变为t-SNE（t分布随机邻居嵌入），然后发展到UMAP（均匀流形近似和投影）

• TSNE是由T和SNE组成，T分布和随机近邻嵌入(Stochastic neighbor Embedding).
• TSNE是一种可视化工具，将高位数据降到2-3维，然后画成图。
• t-SNE是目前效果最好的数据降维和可视化方法
• t-SNE的缺点是：占用内存大，运行时间长。

t-sne实现过程：How to Use t-SNE Effectively (distill.pub)

T-SNE算法步骤

• 找出高维空间中相邻点之间的成对相似性。
• 根据高维空间中点的成对相似性，将高维空间中的每个点映射到低维映射。
• 使用基于Kullback-Leibler散度(KL散度)的梯度下降找到最小化条件概率分布之间的不匹配的低维数据表示。
• 使用Student-t分布计算低维空间中两点之间的相似度。

案例1：随机生成数据并进行降维

类型1

import matplotlib.pyplot as plt
from sklearn.manifold import TSNE

tsne = TSNE(n_components=2)
x = [[1,2,2],[2,2,2],[3,3,3]]
y = [1,0,2]#y是x对应的标签
x_tsne = tsne.fit_transform(x)
plt.scatter(x_tsne[:,0],x_tsne[:,1],c=y)
plt.show()

 

 类型2

#pip install pandas
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

sns.set()


def get_data():
    """生成聚类数据"""
    from sklearn.datasets import make_blobs
    x_value, y_value = make_blobs(n_samples=1000, n_features=40, centers=3, )
    return x_value, y_value


def plot_xy(x_values, label, title):
    """绘图"""
    df = pd.DataFrame(x_values, columns=['x', 'y'])
    df['label'] = label
    sns.scatterplot(x="x", y="y", hue="label", data=df)
    plt.title(title)
    plt.show()


def main():
    x_value, y_value = get_data()
    # PCA 降维
    from sklearn.decomposition import PCA
    pca = PCA(n_components=2)
    x_pca = pca.fit_transform(x_value)
    plot_xy(x_pca, y_value, "PCA")
    # t-sne 降维
    from sklearn.manifold import TSNE
    tsne = TSNE(n_components=2)
    x_tsne = tsne.fit_transform(x_value)
    plot_xy(x_tsne, y_value, "t-sne")


if __name__ == '__main__':
    main()

 案例2：用t-SNE降维查看分类效果

读取数据集的特征及标签，并进行降维画图

2.1.计算机视觉类

2.1.1.MNIST数据集上实现t-SNE的Python代码

#! pip install sklearn
#! pip install seaborn
#! pip install matplotlib

from sklearn.datasets import load_digits
from sklearn.manifold import TSNE
import seaborn as sns
from matplotlib import pyplot as plt

# 0-9的数字数据
data = load_digits()
embeddings = TSNE().fit_transform(digits.data)#t-SNE降维，默认降为二维
vis_x = embeddings[:, 0]#0维
vis_y = embeddings[:, 1]#1维

index0 = [i for i in range(len(digits.target)) if digits.target == 0]
index1 = [i for i in range(len(digits.target)) if digits.target == 1]
index2 = [i for i in range(len(digits.target)) if digits.target == 2]
index3 = [i for i in range(len(digits.target)) if digits.target == 3]
index4 = [i for i in range(len(digits.target)) if digits.target == 4]
index5 = [i for i in range(len(digits.target)) if digits.target == 5]
index6 = [i for i in range(len(digits.target)) if digits.target == 6]
index7 = [i for i in range(len(digits.target)) if digits.target == 7]
index8 = [i for i in range(len(digits.target)) if digits.target == 8]
index9 = [i for i in range(len(digits.target)) if digits.target == 9]

colors=['b', 'c', 'y', 'm', 'r', 'g', 'k','yellow','yellowgreen','wheat']
plt.scatter(vis_x[index0], vis_y[index0], c=colors[0], cmap='brg', marker='h',label='0')
plt.scatter(vis_x[index1], vis_y[index1], c=colors[1], cmap='brg',marker='<',label='1')
plt.scatter(vis_x[index2], vis_y[index2], c=colors[2], cmap='brg',marker='x',label='2')
plt.scatter(vis_x[index3], vis_y[index3], c=colors[3], cmap='brg',marker='.',label='3')
plt.scatter(vis_x[index4], vis_y[index4], c=colors[4], cmap='brg',marker='p',label='4')
plt.scatter(vis_x[index5], vis_y[index5], c=colors[5], cmap='brg',marker='>',label='5')
plt.scatter(vis_x[index6], vis_y[index6], c=colors[6], cmap='brg',marker='^',label='6')
plt.scatter(vis_x[index7], vis_y[index7], c=colors[7], cmap='brg',marker='d',label='7')
plt.scatter(vis_x[index8], vis_y[index8], c=colors[8], cmap='brg',marker='s',label='8')
plt.scatter(vis_x[index9], vis_y[index9], c=colors[9], cmap='brg',marker='o',label='9')

plt.title(u't-SNE')
plt.legend()
plt.show()

2.1.2.动物(Animals10)数据集上的可视化

使用 Animals10 数据集。它包含 10 种不同动物的图片：猫、狗、鸡、牛、马、羊、松鼠、大象、蝴蝶和蜘蛛。 

数据集下载地址：Animals-10 (kaggle.com)

代码还在优化中。。。 

import 
#设置随机种子
seed = 10
random.seed(seed)
torch.manual_seed(seed)
np.random.seed(seed)

# ResNet101 网络提取图片特征

 # Define the architecture by modifying resnet.
# Original code is here http://tiny.cc/8zpmmz 
class ResNet101(models.ResNet):
    def __init__(self, num_classes=1000, pretrained=True, **kwargs):
        # Start with the standard resnet101
        super().__init__(
            block=models.resnet.Bottleneck,
            layers=[3, 4, 23, 3],
            num_classes=num_classes,
            **kwargs
        )
        if pretrained:
            state_dict = load_state_dict_from_url(
                models.resnet.model_urls['resnet101'],
                progress=True
            )
            self.load_state_dict(state_dict)
 
    # Reimplementing forward pass.
    # Replacing the forward inference defined here 
    # http://tiny.cc/23pmmz
    def _forward_impl(self, x):
        # Standard forward for resnet
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)
 
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
 
        # Notice there is no forward pass through the original classifier.
        x = self.avgpool(x)
        x = torch.flatten(x, 1)
 
        return x

#存储推按2048特征
# initialize our implementation of ResNet
model = ResNet101(pretrained=True)
model.eval()
 
for batch in tqdm(dataloader, desc='Running the model inference'):
    images = batch['image'].to(device)
    labels += batch['label']
    image_paths += batch['image_path']
 
    output = model.forward(images)
 
    current_outputs = output.cpu().numpy()
    features = np.concatenate((outputs, current_outputs))

	
tsne = TSNE(n_components=2).fit_transform(features)


# scale and move the coordinates so they fit [0; 1] range
def scale_to_01_range(x):
    # compute the distribution range
    value_range = (np.max(x) - np.min(x))
 
    # move the distribution so that it starts from zero
    # by extracting the minimal value from all its values
    starts_from_zero = x - np.min(x)
 
    # make the distribution fit [0; 1] by dividing by its range
    return starts_from_zero / value_range
 
# extract x and y coordinates representing the positions of the images on T-SNE plot
tx = tsne[:, 0]
ty = tsne[:, 1]
 
tx = scale_to_01_range(tx)
ty = scale_to_01_range(ty)

# 绘制2D点，每个点的颜色与类标签对应

# initialize a matplotlib plot
fig = plt.figure()
ax = fig.add_subplot(111)
 
# for every class, we'll add a scatter plot separately
for label in colors_per_class:
    # find the samples of the current class in the data
    indices = [i for i, l in enumerate(labels) if l == label]
 
    # extract the coordinates of the points of this class only
    current_tx = np.take(tx, indices)
    current_ty = np.take(ty, indices)
 
    # convert the class color to matplotlib format
    color = np.array(colors_per_class[label], dtype=np.float) / 255
 
    # add a scatter plot with the corresponding color and label
    ax.scatter(current_tx, current_ty, c=color, label=label)
 
# build a legend using the labels we set previously
ax.legend(loc='best')
 
# finally, show the plot
plt.show()

2.2.语音类

2.2.1在语音数据集上利用tsne进行可视化

要求数据类别和数据是相同的，需要修改部分数据

from matplotlib import pyplot as plt
import matplotlib.cm as cm
import fnmatch
import os
import numpy as np
import librosa
import matplotlib.pyplot as plt
import librosa.display
from sklearn.manifold import TSNE
import json
# Importing library 
import csv 
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import glob

path = "/workspace/emo-vits/dataset/p225"

files = []
for root, dirnames, filenames in os.walk(path):
    for filename in fnmatch.filter(filenames, '*.wav'):
        files.append(os.path.join(root, filename))

print("found %d .wav files"%(len(files)))

def get_features(y, sr):
    y = y[0:sr]  # analyze just first second
    S = librosa.feature.melspectrogram(y, sr=sr, n_mels=128)
    log_S = librosa.amplitude_to_db(S, ref=np.max)
    mfcc = librosa.feature.mfcc(S=log_S, n_mfcc=13)
    delta_mfcc = librosa.feature.delta(mfcc, mode='nearest')
    delta2_mfcc = librosa.feature.delta(mfcc, order=2, mode='nearest')
    feature_vector = np.concatenate((np.mean(mfcc,1), np.mean(delta_mfcc,1), np.mean(delta2_mfcc,1)))
    feature_vector = (feature_vector-np.mean(feature_vector)) / np.std(feature_vector)
    return feature_vector

feature_vectors = []
sound_paths = []
for i,f in enumerate(files):
    if i % 100 == 0:
        print("get %d of %d = %s"%(i+1, len(files), f))
    y, sr = librosa.load(f)
    feat = get_features(y, sr)
    feature_vectors.append(feat)
    sound_paths.append(f)
        
print("calculated %d feature vectors"%len(feature_vectors))

model = TSNE(n_components=2, learning_rate=150, perplexity=30, verbose=2, angle=0.1).fit_transform(feature_vectors)

symbol=[]
symbol=[1]*1400
x=[2]*1400
symbol.extend(x)

# classes
file=[1,2,3]
color=[]
for i in file:
    x=[i]*20
    color.extend(x)
color.extend(color)
print(len(color))

x_axis=model[:,0]
y_axis=model[:,1]
import plotly.express as px
fig = px.scatter(x=x_axis, y=y_axis,color=color,symbol=symbol,opacity=0.7)
fig.show()

'''
# load 2D vector
x = np.load(path)

#使用np.ravel将向量展平为一维数组。
#x = np.ravel(x) 
print(x.shape)

# 2D vector -> high-dis
tsne= TSNE(n_components=2).fit_transform(x)
X_embedded = tsne.fit_transform(X)

# high-dis -> 2D vector
line_vector = 0
for i in range(0, 222):
    line_vector = x[i]
    print(line_vector)'''

参考文献

【1】Clustering with KMeans, PCA, TSNE | Kaggle

【2】[차원축소/시각화 방법] TSNE - Python 에서 T-SNE를 이용하는 방법 :: The Yellow Lion King 데이터와 함께 살아가기 (tistory.com) 【3】t-SNE 개념과 사용법 - gaussian37

【4】 Audio Dataset Analysis-4 | Kaggle