关于SimGCL中的分布图画法请教

sunshinelium commented 2 years ago

非常抱歉打扰大佬，我最近关注到您在SIGIR22‘发表的SimGCL的工作且对您里边画的分布图非常感兴趣，不知能否获得关于这段画图的代码，提前谢谢大佬了。

Coder-Yu commented 1 year ago

核心代码如下：

f, axs = plt.subplots(2, len(embs), figsize=(12,3.5),gridspec_kw={'height_ratios': [3, 1]})
kwargs = {'levels': np.arange(0, 5.5, 0.5)}
for i,name in enumerate(models):
    sns.kdeplot(data=data[name], bw=0.05, shade=True, cmap="GnBu", legend=True, ax=axs[0][i], **kwargs)
    axs[0][i].set_title(name, fontsize=9, fontweight="bold")
    x = [p[0] for p in data[name]]
    y = [p[1] for p in data[name]]
    angles = np.arctan2(y, x)
    sns.kdeplot(data=angles, bw=0.15, shade=True, legend=True, ax=axs[1][i], color='green')

data[name] 为normalize之后的2d向量

seaborn的版本需要低于0.10.1

Coder-Yu commented 1 year ago

embedding降维代码为

from sklearn import manifold
from sklearn.preprocessing import normalize

tsne = manifold.TSNE(n_components=2, init='pca', random_state=501)
user_emb_2d = tsne.fit_transform(user_emb)
user_emb_2d = normalize(user_emb_2d, axis=1, norm='l2')

sunshinelium commented 1 year ago

非常感谢啦

Gaozzzz commented 1 year ago

大佬你好，想要一下这里tsne分布图的完整代码。实在是复现不出来。希望可以发到邮箱527043469@qq.com 谢谢大佬！！！！！

Coder-Yu commented 1 year ago

import pickle
import numpy as np
import matplotlib.pyplot as plt
import matplotlib
from sklearn import manifold
from sklearn.preprocessing import normalize
import seaborn as sns
from math import pi
matplotlib.rcParams['pdf.fonttype'] = 42
matplotlib.rcParams['ps.fonttype'] = 42

ue = open('user', 'rb')
user = pickle.load(ue)
udx = np.random.choice(len(user), 2000)
embs = ['user_lgcn.emb', 'user_sgl.emb', 'user_simgcl.emb']
models = ['LightGCN','SGL-ED','SimGCL']
data = {}

for emb,name in zip(embs,models):
    ue = open(emb, 'rb')
    user_emb = pickle.load(ue)
    selected_user_emb = user_emb[udx]  # np.concatenate([item_emb[idx],user_emb[udx]],axis=0)
    tsne = manifold.TSNE(n_components=2, init='pca', random_state=501)
    user_emb_2d = tsne.fit_transform(selected_user_emb)
    user_emb_2d = normalize(user_emb_2d, axis=1, norm='l2')
    data[name]=user_emb_2d

f, axs = plt.subplots(2, len(embs), figsize=(12,3.5),gridspec_kw={'height_ratios': [3, 1]})
kwargs = {'levels': np.arange(0, 5.5, 0.5)}
for i,name in enumerate(models):
    sns.kdeplot(data=data[name], bw=0.05, shade=True, cmap="GnBu", legend=True, ax=axs[0][i], **kwargs)
    axs[0][i].set_title(name, fontsize=9, fontweight="bold")
    x = [p[0] for p in data[name]]
    y = [p[1] for p in data[name]]
    angles = np.arctan2(y, x)
    sns.kdeplot(data=angles, bw=0.15, shade=True, legend=True, ax=axs[1][i], color='green')

for ax in axs[0]:
    ax.tick_params(axis='x', labelsize=8)
    ax.tick_params(axis='y', labelsize=8)
    ax.patch.set_facecolor('white')
    ax.collections[0].set_alpha(0)
    ax.set_xlim(-1.2, 1.2)
    ax.set_ylim(-1.2, 1.2)
    ax.set_xlabel('Features', fontsize=9)
axs[0][0].set_ylabel('Features', fontsize=9)

for ax in axs[1]:
    ax.tick_params(axis='x', labelsize=8)
    ax.tick_params(axis='y', labelsize=8)
    ax.set_xlabel('Angles', fontsize=9)
    ax.set_ylim(0, 0.5)
    ax.set_xlim(-pi, pi)
axs[1][0].set_ylabel('Density', fontsize=9)

plt.show()

Coder-Yu commented 1 year ago

seaborn版本为 0.10

Gaozzzz commented 1 year ago

感谢！