Skip to content

D
docs

MindSpore / docs

通知 4
Star 2
Fork 2
D
docs

前往新版Gitcode，体验更适合开发者的 AI 搜索 >>

提交 dd1dc944 编写于 作者: M mindspore-ci-bot 提交者： Gitee
浏览文件
操作

!318 add Mindinsight Notebook Doc for histogram,image and scalar to r0.5 

Merge pull request !318 from zhangyi/MindInsightDoc_Add_r0.5
上级 81878f1b fdb49bcb
隐藏空白更改
内联 并排
Showing with 602 addition and 1 deletion
tutorials/notebook/mindinsight/mindinsight_image_histogram_scalar.ipynb 0 → 100644
浏览文件 @ dd1dc944
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# MindInsight之标量、直方图和图像\n",
"\n",
"MindInsight可以将神经网络训练过程中的损失值标量、直方图、图像信息记录到日志文件中，通过可视化界面解析以供用户查看。\n",
"\n",
"整体流程:\n",
"\n",
"1. 下载MNIST数据集。\n",
"\n",
"2. 原始数据预处理。\n",
"\n",
"3. 初始化`lenet`网络。\n",
"\n",
"4. 执行主程序，使用`SummaryCollector`记录图像信息、损失值标量、权重梯度等参数，启动MindInsight服务。\n",
"\n",
"5. 在MindInsight可视化面板中查看结果。"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# 数据集操作\n",
"\n",
"本次流程用到MNIST数据集，MNIST数据集是一个手写的数据文件，数据库里的图像都是28x28的灰度图像，每个像素都是一个八位字节，包含了60000张训练图像和10000张测试图像，常被用作神经网络训练和测试任务的数据集。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/mnist_dataset.png)\n",
"\n",
"## 下载MNIST数据集\n",
"\n",
"下面一段代码分为两部分：\n",
"\n",
"1. 判断是否存在MNIST数据集目录，不存在则创建目录，存在则跳至[**数据预处理**](#数据预处理)。\n",
"\n",
"2. 判断是否存在MNIST数据集，不存在则下载MNIST数据集，存在则跳至[**数据预处理**](#数据预处理)。\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"import urllib.request\n",
"from urllib.parse import urlparse\n",
"import gzip\n",
"\n",
"def unzipfile(gzip_path):\n",
" \"\"\"unzip dataset file\n",
" Args:\n",
" gzip_path: dataset file path\n",
" \"\"\"\n",
" open_file = open(gzip_path.replace('.gz',''), 'wb')\n",
" gz_file = gzip.GzipFile(gzip_path)\n",
" open_file.write(gz_file.read())\n",
" gz_file.close()\n",
"\n",
"\n",
"def download_dataset():\n",
" \"\"\"Download the dataset from http://yann.lecun.com/exdb/mnist/.\"\"\"\n",
" print(\"******Downloading the MNIST dataset******\")\n",
" train_path = \"./MNIST_Data/train/\"\n",
" test_path = \"./MNIST_Data/test/\"\n",
" train_path_check = os.path.exists(train_path)\n",
" test_path_check = os.path.exists(test_path)\n",
" if train_path_check == False and test_path_check ==False:\n",
" os.makedirs(train_path)\n",
" os.makedirs(test_path)\n",
" train_url = {\"http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz\", \"http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz\"}\n",
" test_url = {\"http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz\", \"http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz\"}\n",
" for url in train_url:\n",
" url_parse = urlparse(url)\n",
" # split the file name from url\n",
" file_name = os.path.join(train_path,url_parse.path.split('/')[-1])\n",
" if not os.path.exists(file_name.replace('.gz','')):\n",
" file = urllib.request.urlretrieve(url, file_name)\n",
" unzipfile(file_name)\n",
" os.remove(file_name)\n",
" for url in test_url:\n",
" url_parse = urlparse(url)\n",
" # split the file name from url\n",
" file_name = os.path.join(test_path,url_parse.path.split('/')[-1])\n",
" if not os.path.exists(file_name.replace('.gz','')):\n",
" file = urllib.request.urlretrieve(url, file_name)\n",
" unzipfile(file_name)\n",
" os.remove(file_name)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 数据预处理\n",
"\n",
"好的数据集可以有效提高训练精度和效率，在加载数据集前，会进行一些处理，增加数据的可用性和随机性。下面一段代码定义`create_dataset`函数进行数据处理操作。\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import mindspore.dataset as ds\n",
"import mindspore.dataset.transforms.vision.c_transforms as CV\n",
"import mindspore.dataset.transforms.c_transforms as C\n",
"from mindspore.dataset.transforms.vision import Inter\n",
"from mindspore.common import dtype as mstype\n",
"\n",
"\n",
"def create_dataset(data_path, batch_size=32, repeat_size=1,\n",
" num_parallel_workers=1):\n",
" \"\"\"create dataset for train or test.\"\"\"\n",
" # define dataset\n",
" mnist_ds = ds.MnistDataset(data_path)\n",
"\n",
" resize_height, resize_width = 32, 32\n",
" rescale = 1.0 / 255.0\n",
" shift = 0.0\n",
" rescale_nml = 1 / 0.3081\n",
" shift_nml = -1 * 0.1307 / 0.3081\n",
"\n",
" # define map operations\n",
" resize_op = CV.Resize((resize_height, resize_width), interpolation=Inter.LINEAR) # Bilinear mode\n",
" rescale_op = CV.Rescale(rescale, shift)\n",
" hwc2chw_op = CV.HWC2CHW()\n",
" type_cast_op = C.TypeCast(mstype.int32)\n",
"\n",
" # apply map operations on images\n",
" mnist_ds = mnist_ds.map(input_columns=\"label\", operations=type_cast_op, num_parallel_workers=num_parallel_workers)\n",
" mnist_ds = mnist_ds.map(input_columns=\"image\", operations=resize_op, num_parallel_workers=num_parallel_workers)\n",
" mnist_ds = mnist_ds.map(input_columns=\"image\", operations=rescale_op, num_parallel_workers=num_parallel_workers)\n",
" mnist_ds = mnist_ds.map(input_columns=\"image\", operations=hwc2chw_op, num_parallel_workers=num_parallel_workers)\n",
"\n",
" # apply DatasetOps\n",
" buffer_size = 10000\n",
" mnist_ds = mnist_ds.shuffle(buffer_size=buffer_size) # 10000 as in LeNet train script\n",
" mnist_ds = mnist_ds.batch(batch_size, drop_remainder=True)\n",
" mnist_ds = mnist_ds.repeat(repeat_size)\n",
"\n",
" return mnist_ds"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# 网络初始化\n",
"\n",
"在进行训练之前，需定义神经网络模型，本流程采用最简单的[LeNet卷积神经网络](http://yann.lecun.com/exdb/lenet/)。\n",
"\n",
"LeNet网络不包括输入层的情况下，共有7层：2个卷积层、2个下采样层（池化层）、3个全连接层。每层都包含不同数量的训练参数。"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import mindspore.nn as nn\n",
"from mindspore.common.initializer import TruncatedNormal\n",
"\n",
"\n",
"def conv(in_channels, out_channels, kernel_size, stride=1, padding=0):\n",
" \"\"\"weight initial for conv layer\"\"\"\n",
" weight = weight_variable()\n",
" return nn.Conv2d(in_channels, out_channels,\n",
" kernel_size=kernel_size, stride=stride, padding=padding,\n",
" weight_init=weight, has_bias=False, pad_mode=\"valid\")\n",
"\n",
"\n",
"def fc_with_initialize(input_channels, out_channels):\n",
" \"\"\"weight initial for fc layer\"\"\"\n",
" weight = weight_variable()\n",
" bias = weight_variable()\n",
" return nn.Dense(input_channels, out_channels, weight, bias)\n",
"\n",
"\n",
"def weight_variable():\n",
" \"\"\"weight initial\"\"\"\n",
" return TruncatedNormal(0.02)\n",
"\n",
"\n",
"class LeNet5(nn.Cell):\n",
" \"\"\"\n",
" Lenet network\n",
"\n",
" Args:\n",
" num_class (int): Num classes. Default: 10.\n",
"\n",
" Returns:\n",
" Tensor, output tensor\n",
"\n",
" \"\"\"\n",
" def __init__(self, num_class=10, channel=1):\n",
" super(LeNet5, self).__init__()\n",
" self.num_class = num_class\n",
" self.conv1 = conv(channel, 6, 5)\n",
" self.conv2 = conv(6, 16, 5)\n",
" self.fc1 = fc_with_initialize(16 * 5 * 5, 120)\n",
" self.fc2 = fc_with_initialize(120, 84)\n",
" self.fc3 = fc_with_initialize(84, self.num_class)\n",
" self.relu = nn.ReLU()\n",
" self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)\n",
" self.flatten = nn.Flatten()\n",
"\n",
" def construct(self, x):\n",
" x = self.conv1(x)\n",
" x = self.relu(x)\n",
" x = self.max_pool2d(x)\n",
" x = self.conv2(x)\n",
" x = self.relu(x)\n",
" x = self.max_pool2d(x)\n",
" x = self.flatten(x)\n",
" x = self.fc1(x)\n",
" x = self.relu(x)\n",
" x = self.fc2(x)\n",
" x = self.relu(x)\n",
" x = self.fc3(x)\n",
" return x"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# 记录标量、直方图、图像\n",
"\n",
"在主程序中应用`SummaryCollector`来记录标量、直方图、图像信息。\n",
"\n",
"## 运行主程序\n",
"\n",
"在MindSpore中通过`Callback`机制提供支持快速简易地收集损失值、参数权重、梯度等信息的`Callback`, 叫做`SummaryCollector`。详细的用法可以参考API文档中`mindspore.train.callback.SummaryCollector`。 \n",
"\n",
"1. 为了记录损失值标量、直方图、图像信息，在主程序代码中需要在`specified`参数中指定需要记录的信息。\n",
"\n",
" ```python\n",
" specified={\"collect_metric\": True, \"histogram_regular\": \"^conv1.*|^conv2.*\", \"collect_input_data\": True}\n",
" ```\n",
"\n",
" 其中：\n",
" - `\"collect_metric\"`为记录损失值标量信息。\n",
" - `\"histogram_regular\"`为记录`conv1`层和`conv2`层直方图信息。\n",
" - `\"collect_input_data\"`为记录图像信息。\n",
"\n",
"2. 实例化`SummaryCollector`，并将其应用到`model.train`或者`model.eval`中。\n",
"\n",
"程序运行过程中将启动MindInsight服务并自动遍历读取当前notebook目录下`summary_dir`子目录下所有日志文件、解析进行可视化展示。"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import mindspore.nn as nn\n",
"from mindspore import context\n",
"from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor, TimeMonitor\n",
"from mindspore.train import Model\n",
"from mindspore.nn.metrics import Accuracy\n",
"from mindspore.train.callback import SummaryCollector\n",
"from mindspore.train.serialization import load_checkpoint, load_param_into_net\n",
"\n",
"\n",
"if __name__ == \"__main__\":\n",
" device_target = \"GPU\"\n",
" summary_base_dir = \"./summary_dir\"\n",
" context.set_context(mode=context.GRAPH_MODE, device_target=device_target)\n",
" download_dataset()\n",
" ds_train = create_dataset(data_path=\"./MNIST_Data/train/\")\n",
" network = LeNet5()\n",
" net_loss = nn.SoftmaxCrossEntropyWithLogits(is_grad=False, sparse=True, reduction=\"mean\")\n",
" net_opt = nn.Momentum(network.trainable_params(), learning_rate=0.01, momentum=0.9)\n",
" time_cb = TimeMonitor(data_size=ds_train.get_dataset_size())\n",
" config_ck = CheckpointConfig(save_checkpoint_steps=1875, keep_checkpoint_max=10)\n",
" ckpoint_cb = ModelCheckpoint(prefix=\"checkpoint_lenet\", config=config_ck)\n",
" model = Model(network, net_loss, net_opt, metrics={\"Accuracy\": Accuracy()})\n",
" os.system(f\"mindinsight start --summary-base-dir {summary_base_dir} --port=8080\")\n",
" # Init a SummaryCollector callback instance, and use it in model.train or model.eval\n",
" specified = {\"collect_metric\": True, \"histogram_regular\": \"^conv1.*|^conv2.*\", \"collect_input_data\": True}\n",
" summary_collector = SummaryCollector(summary_dir=\"./summary_dir/summary_01\", collect_specified_data=specified, collect_freq=1, keep_default_action=False)\n",
" print(\"============== Starting Training ==============\")\n",
" # Note: dataset_sink_mode should be set to False, else you should modify collect freq in SummaryCollector\n",
" model.train(epoch=3, train_dataset=ds_train, callbacks=[time_cb, ckpoint_cb, LossMonitor(), summary_collector], dataset_sink_mode=False)\n",
" print(\"============== Starting Testing ==============\")\n",
" param_dict = load_checkpoint(\"checkpoint_lenet-3_1875.ckpt\")\n",
" load_param_into_net(network, param_dict)\n",
" ds_eval = create_dataset(\"./MNIST_Data/test/\")\n",
" acc = model.eval(ds_eval, callbacks=summary_collector, dataset_sink_mode=False)\n",
" print(\"============== {} ==============\".format(acc))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# MindInsight看板\n",
"\n",
"在本地浏览器中打开地址：`127.0.0.1:8080`，进入到可视化面板。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/mindinsight_panel.png)\n",
"\n",
"在上图所示面板中可以看到`summary_01`日志文件目录，点击训练看板进入到下图所示的训练数据展示面板，该面板展示了标量数据、直方图和图像信息，并随着训练、测试的进行实时刷新数据，实时显示训练过程参数的变化情况。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/mindinsight_panel2.png)\n",
"\n",
"## 标量可视化\n",
"\n",
"标量可视化用于展示训练过程中标量的变化趋势情况，点击打开标量信息展示面板，该面板记录了迭代计算过程中的学习率（下图左侧所示）和损失值（下图右侧所示）标量信息。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/scalar_panel.png)\n",
"\n",
"如下图的loss值标量可视化信息——标量趋势图。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/scalar.png)\n",
"\n",
"上图展示了神经网络在训练过程中loss值的变化过程。横坐标是训练步骤，纵坐标是loss值。\n",
"\n",
"图中右上角有几个按钮功能，从左到右功能分别是全屏展示，切换Y轴比例，开启/关闭框选，分步回退和还原图形。\n",
"\n",
"- 全屏展示即全屏展示该标量曲线，再点击一次即可恢复。\n",
"\n",
"- 切换Y轴比例是指可以将Y轴坐标进行对数转换。\n",
"\n",
"- 开启/关闭框选是指可以框选图中部分区域，并放大查看该区域， 可以在已放大的图形上叠加框选。\n",
"\n",
"- 分步回退是指对同一个区域连续框选并放大查看时，可以逐步撤销操作。\n",
"\n",
"- 还原图形是指进行了多次框选后，点击此按钮可以将图还原回原始状态。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/scalar_select.png)\n",
"\n",
"上图展示的标量可视化的功能区，提供了根据选择不同标签，水平轴的不同维度和平滑度来查看标量信息的功能。\n",
"\n",
"- 标签：提供了对所有标签进行多项选择的功能，用户可以通过勾选所需的标签，查看对应的标量信息。\n",
"\n",
"- 水平轴：可以选择“步骤”、“相对时间”、“绝对时间”中的任意一项，来作为标量曲线的水平轴。\n",
"\n",
"- 平滑度：可以通过调整平滑度，对标量曲线进行平滑处理。\n",
"\n",
"- 标量合成：可以选中两条标量曲线进行合成并展示在一个图中，以方便对两条曲线进行对比或者查看合成后的图。\n",
" 标量合成的功能区与标量可视化的功能区相似。其中与标量可视化功能区不一样的地方，在于标签选择时，标量合成功能最多只能同时选择两个标签，将其曲线合成并展示。\n",
"\n",
"## 直方图可视化\n",
"\n",
"\n",
"直方图用于将用户所指定的张量以直方图的形式展示。点击打开直方图展示面板，以直方图的形式记录了在迭代过程中所有层参数分布信息。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/histogram_panel.png)\n",
"\n",
"如下图为`conv1`层参数分布信息，点击图中右上角，可以将图放大。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/histogram.png)\n",
"\n",
"上图展示直方图的功能区，包含以下内容：\n",
"\n",
"- 标签选择：提供了对所有标签进行多项选择的功能，用户可以通过勾选所需的标签，查看对应的直方图。\n",
"\n",
"- 纵轴：可以选择步骤、相对时间、绝对时间中的任意一项，来作为直方图纵轴显示的数据。\n",
"\n",
"- 视角：可以选择正视和俯视中的一种。正视是指从正面的角度查看直方图，此时不同步骤之间的数据会覆盖在一起。俯视是指偏移以45度角俯视直方图区域，这时可以呈现不同步骤之间数据的差异。\n",
"\n",
"## 图像可视化\n",
"\n",
"图像可视化用于展示用户所指定的图片。点击图像展示面板，展示了每个step进行处理的图像信息。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/image_panel.png)\n",
"\n",
"下图为展示`summary_01`记录的图像信息。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/image_vi.png)\n",
"\n",
"通过滑动上图中的\"步骤\"滑条，查看不同步骤的图片。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/image_function.png)\n",
"\n",
"上图展示图像可视化的功能区，提供了选择查看不同标签，不同亮度和不同对比度来查看图片信息。\n",
"\n",
"- 标签：提供了对所有标签进行多项选择的功能，用户可以通过勾选所需的标签，查看对应的图片信息。\n",
"\n",
"- 亮度调整：可以调整所展示的所有图片亮度。\n",
"\n",
"- 对比度调整：可以调整所展示的所有图片对比度。"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 对比看板\n",
"\n",
"对比看板可视用于多个训练之间的标量数据对比，为了展示对比看板，执行以下代码，在可视化面板中可以得到`summary_02`日志记录信息。"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import mindspore.nn as nn\n",
"from mindspore import context\n",
"from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor, TimeMonitor\n",
"from mindspore.train import Model\n",
"from mindspore.nn.metrics import Accuracy\n",
"from mindspore.train.callback import SummaryCollector\n",
"from mindspore.train.serialization import load_checkpoint, load_param_into_net\n",
"\n",
"\n",
"if __name__ == \"__main__\":\n",
" device_target = \"GPU\"\n",
" context.set_context(mode=context.GRAPH_MODE, device_target=device_target)\n",
" download_dataset()\n",
" ds_train = create_dataset(data_path=\"./MNIST_Data/train/\")\n",
" network = LeNet5()\n",
" net_loss = nn.SoftmaxCrossEntropyWithLogits(is_grad=False, sparse=True, reduction=\"mean\")\n",
" net_opt = nn.Momentum(network.trainable_params(), learning_rate=0.01, momentum=0.9)\n",
" time_cb = TimeMonitor(data_size=ds_train.get_dataset_size())\n",
" config_ck = CheckpointConfig(save_checkpoint_steps=1875, keep_checkpoint_max=10)\n",
" ckpoint_cb = ModelCheckpoint(prefix=\"checkpoint_lenet\", config=config_ck)\n",
" model = Model(network, net_loss, net_opt, metrics={\"Accuracy\": Accuracy()})\n",
" # Init a SummaryCollector callback instance, and use it in model.train or model.eval\n",
" specified = {\"collect_metric\": True, \"histogram_regular\": \"^conv1.*|^conv2.*\", \"collect_input_data\": True}\n",
" summary_collector = SummaryCollector(summary_dir=\"./summary_dir/summary_02\", collect_specified_data=specified, collect_freq=1, keep_default_action=False)\n",
" print(\"============== Starting Training ==============\")\n",
" # Note: dataset_sink_mode should be set to False, else you should modify collect freq in SummaryCollector\n",
" model.train(epoch=3, train_dataset=ds_train, callbacks=[time_cb, ckpoint_cb, LossMonitor(), summary_collector], dataset_sink_mode=False)\n",
" print(\"============== Starting Testing ==============\")\n",
" param_dict = load_checkpoint(\"checkpoint_lenet_1-3_1875.ckpt\")\n",
" load_param_into_net(network, param_dict)\n",
" ds_eval = create_dataset(\"./MNIST_Data/test/\")\n",
" acc = model.eval(ds_eval, callbacks=summary_collector, dataset_sink_mode=False)\n",
" print(\"============== {} ==============\".format(acc))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"打开对比看板，可以得到`summary_01`和`summary_02`标量对比信息。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/multi_scalars.png)\n",
"\n",
"上图展示了多个训练之间的标量曲线对比效果，横坐标是训练步骤，纵坐标是标量值。\n",
"\n",
"图中右上角有几个按钮功能，从左到右功能分别是全屏展示，切换Y轴比例，开启/关闭框选，分步回退和还原图形。\n",
"\n",
"- 全屏展示即全屏展示该标量曲线，再点击一次即可恢复。\n",
"\n",
"- 切换Y轴比例是指可以将Y轴坐标进行对数转换。\n",
"\n",
"- 开启/关闭框选是指可以框选图中部分区域，并放大查看该区域， 可以在已放大的图形上叠加框选。\n",
"\n",
"- 分步回退是指对同一个区域连续框选并放大查看时，可以逐步撤销操作。\n",
"\n",
"- 还原图形是指进行了多次框选后，点击此按钮可以将图还原回原始状态。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/multi_scalars_select.png)\n",
"\n",
"上图展示的对比看板可视的功能区，提供了根据选择不同训练或标签，水平轴的不同维度和平滑度来进行标量对比的功能。\n",
"\n",
"- 训练: 提供了对所有训练进行多项选择的功能，用户可以通过勾选或关键字筛选所需的训练。\n",
"\n",
"- 标签：提供了对所有标签进行多项选择的功能，用户可以通过勾选所需的标签，查看对应的标量信息。\n",
"\n",
"- 水平轴：可以选择“步骤”、“相对时间”、“绝对时间”中的任意一项，来作为标量曲线的水平轴。\n",
"\n",
"- 平滑度：可以通过调整平滑度，对标量曲线进行平滑处理。"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# 单独记录数据\n",
"\n",
"以上流程为整体展示Summary算子能记录到的所有数据，也可以单独记录关心的数据，以降低性能开销和日志文件大小。\n",
"\n",
"> 为了展示运行的效果，进行以下每个步骤之前先删除当前notebook根目录下的`summary_dir/summary_02`目录，配置完`specified`参数后执行[**对比看板**](#对比看板)中的代码。\n",
"\n",
"## 单独记录损失值标量\n",
"\n",
"在主程序中配置`specified`参数为：\n",
"\n",
"```python\n",
"specified={\"collect_metric\": True}\n",
"```\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/loss_scalar_only.png)\n",
"\n",
"在MindInsight面板中，如上图所示，只展示了损失值标量信息。\n",
"\n",
"## 单独记录参数分布直方图\n",
"\n",
"在主程序中配置`specified`参数为：\n",
"\n",
"```python\n",
"specified = {\"histogram_regular\": \"^conv1.*|^conv2.*|fc1.*|fc2.*|fc3.*\"}\n",
"```\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/histogram_only.png)\n",
"\n",
"在MindInsight面板中，如上图所示，只展示了参数直方图信息。\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/histogram_only_all.png)\n",
"\n",
"点击进入直方图面板，如上图所示，展示了`conv1`、`conv2`、`fc1`、`fc2`、`fc3`等各层的权重、参数信息。\n",
"\n",
"## 单独记录图像\n",
"\n",
"在主程序中配置`specified`参数为：\n",
"\n",
"```python\n",
"specified = {\"collect_input_data\": True}\n",
"```\n",
"\n",
"![](https://gitee.com/mindspore/docs/raw/master/tutorials/notebook/mindinsight/images/image_only.png)\n",
"\n",
"在MindInsight面板中，如上图所示，只展示了输入图像信息。"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 关闭MindInsight服务\n",
"\n",
"执行以下代码关闭MindInsight服务。"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"\n",
"\n",
"os.system(\"mindinsight stop --port 8080\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 注意事项和规格\n",
"\n",
"- 为了控制列出summary列表的用时，MindInsight最多支持发现999个summary列表条目。\n",
"\n",
"- 为了控制内存占用，MindInsight对标签（tag）数目和步骤（step）数目进行了限制：\n",
"\n",
" - 每个训练看板的最大标签数量为300个标签。标量标签、图片标签、计算图标签、参数分布图（直方图）标签的数量总和不得超过300个。特别地，每个训练看板最多有10个计算图标签。当实际标签数量超过这一限制时，将依照MindInsight的处理顺序，保留最近处理的300个标签。\n",
"\n",
" - 每个训练看板的每个标量标签最多有1000个步骤的数据。当实际步骤的数目超过这一限制时，将对数据进行随机采样，以满足这一限制。\n",
"\n",
" - 每个训练看板的每个图片标签最多有10个步骤的数据。当实际步骤的数目超过这一限制时，将对数据进行随机采样，以满足这一限制。\n",
" \n",
" - 每个训练看板的每个参数分布图（直方图）标签最多有50个步骤的数据。当实际步骤的数目超过这一限制时，将对数据进行随机采样，以满足这一限制。\n",
" \n",
"- 出于性能上的考虑，MindInsight对比看板使用缓存机制加载训练的标量曲线数据，并进行以下限制：\n",
" \n",
" - 对比看板只支持在缓存中的训练进行比较标量曲线对比。\n",
" \n",
" - 缓存最多保留最新（按修改时间排列）的15个训练。\n",
" \n",
" - 用户最多同时对比5个训练的标量曲线。"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.7.5"
}
},
"nbformat": 4,
"nbformat_minor": 4
}
\ No newline at end of file
tutorials/source_zh_cn/advanced_use/dashboard_and_lineage.md
浏览文件 @ dd1dc944
......@@ -21,7 +21,8 @@
<!-- /TOC -->
<a href="https://gitee.com/mindspore/docs/blob/r0.5/tutorials/source_zh_cn/advanced_use/dashboard_and_lineage.md" target="_blank"><img src="../_static/logo_source.png"></a>
<a href="https://gitee.com/mindspore/docs/blob/r0.5/tutorials/source_zh_cn/advanced_use/dashboard_and_lineage.md" target="_blank"><img src="../_static/logo_source.png"></a>&nbsp;&nbsp;
<a href="https://gitee.com/mindspore/docs/tree/r0.5/tutorials/notebook/mindinsight" target="_blank"><img src="../_static/logo_source.png"></a>
## 概述
训练过程中的标量、图像、计算图以及模型超参等信息记录到文件中，通过可视化界面供用户查看。
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或