1.visdom
visdom是Facebook专门为Pytorch开发的一款可视化工具。在 pytorch可视化之visdom使用中已经对如何使用visdom做了简要介绍。
下面主要从代码方面介绍下。
(1)先看下代码和结果:
import time
from visdom import Visdom
import requests
import os
import numpy as np
import torch
viz = Visdom(server='http://127.0.0.1', port=8097)
assert viz.check_connection()
# video
video_file = 'demo.ogv'
if not os.path.exists(video_file):
video_url = 'http://media.w3.org/2010/05/dintel/trailer.ogv'
res = requests.get(video_url)
with open(video_file, 'wb') as f:
f.write(res.content)
##---------picture-------------##
# single picture
viz.image(
np.random.rand(3, 512, 256),
opts = {
'title': 'Random',
}
)
# multiple picture
viz.images(
#np.random.rand(20, 3, 50, 50),
#torch.randn(3,500,500),
torch.randn(4,3,300,300),
opts = {
'title':'tulti-images',
}
)
# scatter point
Y = np.random.rand(100)
Y = (Y[Y>0] + 1.5).astype(int)
old_scatter = viz.scatter(
X=np.random.rand(100, 2)*100,
Y=Y,
opts={
'title': 'Scatter',
'legend': ['A', 'B'],
'xtickmin': 0,
'xtickstep': 10,
'ytickmax': 100,
'ytickstep': 10,
'matkersymbolk': 'cross-thin-open',
'width': 800,
'height': 600
},
)
# time.sleep(5)
# 更新样式
viz.update_window_opts(
win=old_scatter,
opts={
'tittle': 'New Scatter',
'legend': ['Apple', 'Banana'],
'markersymbol': 'dot'
}
)
# 3D scatter map
viz.scatter(
X=np.random.rand(100, 3),
Y=Y,
opts={
'title': '3D Scatter',
'legend': ['Men', 'Women'],
'makersize': 2
}
)
# histogram
viz.bar(X=np.random.rand(20))
viz.bar(
X=np.abs(np.random.rand(5,3)), # five column, every column has three component
opts={
'stacked': True,
'lengend': ['A', 'B', 'C'],
'rownames': ['2012', '2013', '2014', '2015', '2016']
}
)
# heatmap
viz.heatmap(
X=np.outer(np.arange(1,6), np.arange(1,11)),
opts={
'columnnames': ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j'],
'rownames': ['y1', 'y2', 'y3', 'y4', 'y5'],
'colormap': 'Electric'
}
)
# surface
x = np.tile(np.arange(1,101), (100,1))
y = x.transpose()
X = np.exp((((x - 50) ** 2) + ((y - 50) ** 2)) / -(20.**2))
viz.contour(X=X, opts=dict(colormap='Viridis'))
#
viz.surf(X=X, opts={'colormap': 'Hot'})
(2)绘图命令
单张图片:(image)
viz = Visdom(server='http://127.0.0.1', port=8097)
# 可以接受ndarray数据和torch.tensor 数据
# image 只能绘制一个图片
viz.image(
np.random.rand(3, 512, 256), # 输入的图片矩阵(二维或者三维),可以是[1, h, w](灰度图),或者[3, h, w](rgb图), 或者[h, w](和[1, h, w]相同)
opts = {
'title': 'Random',
)
多张图片:(images)
# images可以绘制多张图片
viz.images(
# 输入的矩阵为[batch, 1或3, h, w](batch张图片,1 为二值图像,3为彩色图像),也可以[1或3,h, w](显示一张图片),[h, w]显示一张二值图片。
#np.random.rand(20, 3, 50, 50),
#torch.randn(3,500,500),
torch.randn(4,3,300,300),
opts = {
'title':'tulti-images',
}
)
散点图:(scatter)
Y = np.random.rand(100) # 01之间均匀分布
Y = (Y[Y>0] + 1.5).astype(int) # 取整到1或者2
old_scatter = viz.scatter(
X=np.random.rand(100, 2)*100, # 二维或者三维矩阵,指定每一个点坐标位置
Y=Y, # 指定每个点的标签
opts={
'title': 'Scatter',
'legend': ['A', 'B'],
'xtickmin': 0,
'xtickstep': 10,
'ytickmax': 100,
'ytickstep': 10,
'matkersymbolk': 'cross-thin-open',
'width': 800,
'height': 600
},
)