{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# 数据集加载\n",
"\n",
"[](https://gitee.com/mindspore/docs/blob/r1.2/docs/programming_guide/source_zh_cn/dataset_loading.ipynb) [](https://obs.dualstack.cn-north-4.myhuaweicloud.com/mindspore-website/notebook/r1.2/programming_guide/mindspore_dataset_loading.ipynb) [](https://console.huaweicloud.com/modelarts/?region=cn-north-4#/notebook/loading?share-url-b64=aHR0cHM6Ly9vYnMuZHVhbHN0YWNrLmNuLW5vcnRoLTQubXlodWF3ZWljbG91ZC5jb20vbWluZHNwb3JlLXdlYnNpdGUvbm90ZWJvb2svbW9kZWxhcnRzL3Byb2dyYW1taW5nX2d1aWRlL21pbmRzcG9yZV9kYXRhc2V0X2xvYWRpbmcuaXB5bmI=&image_id=65f636a0-56cf-49df-b941-7d2a07ba8c8c)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 概述\n",
"\n",
"MindSpore支持加载图像领域常用的数据集，用户可以直接使用`mindspore.dataset`中对应的类实现数据集的加载。目前支持的常用数据集及对应的数据集类如下表所示。\n",
"\n",
"| 图像数据集 | 数据集类 | 数据集简介 |\n",
"| :--------- | :-------------- | :----------------------------------------------------------------------------------------------------------------------------------- |\n",
"| MNIST | MnistDataset | MNIST是一个大型手写数字图像数据集，拥有60,000张训练图像和10,000张测试图像，常用于训练各种图像处理系统。 |\n",
"| CIFAR-10 | Cifar10Dataset | CIFAR-10是一个微小图像数据集，包含10种类别下的60,000张32x32大小彩色图像，平均每种类别6,000张，其中5,000张为训练集，1,000张为测试集。 |\n",
"| CIFAR-100 | Cifar100Dataset | CIFAR-100与CIFAR-10类似，但拥有100种类别，平均每种类别600张，其中500张为训练集，100张为测试集。 |\n",
"| CelebA | CelebADataset | CelebA是一个大型人脸图像数据集，包含超过200,000张名人人脸图像，每张图像拥有40个特征标记。 |\n",
"| PASCAL-VOC | VOCDataset | PASCAL-VOC是一个常用图像数据集，被广泛用于目标检测、图像分割等计算机视觉领域。 |\n",
"| COCO | CocoDataset | COCO是一个大型目标检测、图像分割、姿态估计数据集。 |\n",
"| CLUE | CLUEDataset | CLUE是一个大型中文语义理解数据集。 |\n",
"\n",
"MindSpore还支持加载多种数据存储格式下的数据集，用户可以直接使用`mindspore.dataset`中对应的类加载磁盘中的数据文件。目前支持的数据格式及对应加载方式如下表所示。\n",
"\n",
"| 数据格式 | 数据集类 | 数据格式简介 |\n",
"| :--------- | :----------------- | :------------------------------------------------------------------------------------------------ |\n",
"| MindRecord | MindDataset | MindRecord是MindSpore的自研数据格式，具有读写高效、易于分布式处理等优势。 |\n",
"| Manifest | ManifestDataset | Manifest是华为ModelArts支持的一种数据格式，描述了原始文件和标注信息，可用于标注、训练、推理场景。 |\n",
"| TFRecord | TFRecordDataset | TFRecord是TensorFlow定义的一种二进制数据文件格式。 |\n",
"| NumPy | NumpySlicesDataset | NumPy数据源指的是已经读入内存中的NumPy arrays格式数据集。 |\n",
"| Text File | TextFileDataset | Text File指的是常见的文本格式数据。 |\n",
"| CSV File | CSVDataset | CSV指逗号分隔值，其文件以纯文本形式存储表格数据。 |\n",
"\n",
"MindSpore也同样支持使用`GeneratorDataset`自定义数据集的加载方式，用户可以根据需要实现自己的数据集类。\n",
"\n",
"| 数据集类 | 数据格式简介 |\n",
"| :----------------- | :------------------------------------ |\n",
"| GeneratorDataset | 用户自定义的数据集读取、处理的方式。 |\n",
"| NumpySlicesDataset | 用户自定义的由NumPy构建数据集的方式。 |\n",
"\n",
"> 更多详细的数据集加载接口说明，参见[API文档](https://www.mindspore.cn/doc/api_python/zh-CN/r1.2/mindspore/mindspore.dataset.html)。"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 常用数据集加载\n",
"\n",
"下面将介绍几种常用数据集的加载方式。\n",
"\n",
"### CIFAR-10/100数据集\n",
"\n",
"下载[CIFAR-10数据集](https://www.cs.toronto.edu/~kriz/cifar-10-binary.tar.gz)并解压到指定位置："
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"./datasets/cifar-10-batches-bin\n",
"├── readme.html\n",
"├── test\n",
"│ └── test_batch.bin\n",
"└── train\n",
" ├── batches.meta.txt\n",
" ├── data_batch_1.bin\n",
" ├── data_batch_2.bin\n",
" ├── data_batch_3.bin\n",
" ├── data_batch_4.bin\n",
" └── data_batch_5.bin\n",
"\n",
"2 directories, 8 files\n"
]
}
],
"source": [
"!wget -N https://mindspore-website.obs.cn-north-4.myhuaweicloud.com/notebook/datasets/cifar-10-binary.tar.gz\n",
"!mkdir -p datasets\n",
"!tar -xzf cifar-10-binary.tar.gz -C datasets\n",
"!mkdir -p datasets/cifar-10-batches-bin/train datasets/cifar-10-batches-bin/test\n",
"!mv -f datasets/cifar-10-batches-bin/test_batch.bin datasets/cifar-10-batches-bin/test\n",
"!mv -f datasets/cifar-10-batches-bin/data_batch*.bin datasets/cifar-10-batches-bin/batches.meta.txt datasets/cifar-10-batches-bin/train\n",
"!tree ./datasets/cifar-10-batches-bin"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"下面的样例通过`Cifar10Dataset`接口加载CIFAR-10数据集，使用顺序采样器获取其中5个样本，然后展示了对应图片的形状和标签。\n",
"\n",
"CIFAR-100数据集和MNIST数据集的加载方式也与之类似。"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Image shape: (32, 32, 3) , Label: 6\n",
"Image shape: (32, 32, 3) , Label: 9\n",
"Image shape: (32, 32, 3) , Label: 9\n",
"Image shape: (32, 32, 3) , Label: 4\n",
"Image shape: (32, 32, 3) , Label: 1\n"
]
}
],
"source": [
"import mindspore.dataset as ds\n",
"\n",
"DATA_DIR = \"./datasets/cifar-10-batches-bin/train/\"\n",
"\n",
"sampler = ds.SequentialSampler(num_samples=5)\n",
"dataset = ds.Cifar10Dataset(DATA_DIR, sampler=sampler)\n",
"\n",
"for data in dataset.create_dict_iterator():\n",
" print(\"Image shape:\", data['image'].shape, \", Label:\", data['label'])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### VOC数据集\n",
"\n",
"VOC数据集有多个版本，此处以VOC2012为例。下载[VOC2012数据集](http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar)并解压，目录结构如下。\n",
"\n",
"```\n",
"└─ VOCtrainval_11-May-2012\n",
" └── VOCdevkit\n",
" └── VOC2012\n",
" ├── Annotations\n",
" ├── ImageSets\n",
" ├── JPEGImages\n",
" ├── SegmentationClass\n",
" └── SegmentationObject\n",
"```\n",
"\n",
"下面的样例通过`VOCDataset`接口加载VOC2012数据集，分别演示了将任务指定为分割（Segmentation）和检测（Detection）时的原始图像形状和目标形状。\n",
"\n",
"```python\n",
"import mindspore.dataset as ds\n",
"\n",
"DATA_DIR = \"VOCtrainval_11-May-2012/VOCdevkit/VOC2012/\"\n",
"\n",
"dataset = ds.VOCDataset(DATA_DIR, task=\"Segmentation\", usage=\"train\", num_samples=2, decode=True, shuffle=False)\n",
"\n",
"print(\"[Segmentation]:\")\n",
"for data in dataset.create_dict_iterator():\n",
" print(\"image shape:\", data[\"image\"].shape)\n",
" print(\"target shape:\", data[\"target\"].shape)\n",
"\n",
"dataset = ds.VOCDataset(DATA_DIR, task=\"Detection\", usage=\"train\", num_samples=1, decode=True, shuffle=False)\n",
"\n",
"print(\"[Detection]:\")\n",
"for data in dataset.create_dict_iterator():\n",
" print(\"image shape:\", data[\"image\"].shape)\n",
" print(\"bbox shape:\", data[\"bbox\"].shape)\n",
"```\n",
"\n",
"输出结果：\n",
"\n",
"```text\n",
"[Segmentation]:\n",
"image shape: (281, 500, 3)\n",
"target shape: (281, 500, 3)\n",
"image shape: (375, 500, 3)\n",
"target shape: (375, 500, 3)\n",
"[Detection]:\n",
"image shape: (442, 500, 3)\n",
"bbox shape: (2, 4)\n",
"```"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### COCO数据集\n",
"\n",
"COCO数据集有多个版本，此处以COCO2017的验证数据集为例。下载COCO2017的[验证集](http://images.cocodataset.org/zips/val2017.zip)、[检测任务标注](http://images.cocodataset.org/annotations/annotations_trainval2017.zip)和[全景分割任务标注](http://images.cocodataset.org/annotations/panoptic_annotations_trainval2017.zip)并解压，只取其中的验证集部分，按以下目录结构存放。\n",
"\n",
"```\n",
"└─ COCO\n",
" ├── val2017\n",
" └── annotations\n",
" ├── instances_val2017.json\n",
" ├── panoptic_val2017.json\n",
" └── person_keypoints_val2017.json\n",
"```\n",
"\n",
"下面的样例通过`CocoDataset`接口加载COCO2017数据集，分别演示了将任务指定为目标检测（Detection）、背景分割（Stuff）、关键点检测（Keypoint）和全景分割（Panoptic）时获取到的不同数据。\n",
"\n",
"```python\n",
"import mindspore.dataset as ds\n",
"\n",
"DATA_DIR = \"COCO/val2017/\"\n",
"ANNOTATION_FILE = \"COCO/annotations/instances_val2017.json\"\n",
"KEYPOINT_FILE = \"COCO/annotations/person_keypoints_val2017.json\"\n",
"PANOPTIC_FILE = \"COCO/annotations/panoptic_val2017.json\"\n",
"\n",
"dataset = ds.CocoDataset(DATA_DIR, annotation_file=ANNOTATION_FILE, task=\"Detection\", num_samples=1)\n",
"for data in dataset.create_dict_iterator():\n",
" print(\"Detection:\", data.keys())\n",
"\n",
"dataset = ds.CocoDataset(DATA_DIR, annotation_file=ANNOTATION_FILE, task=\"Stuff\", num_samples=1)\n",
"for data in dataset.create_dict_iterator():\n",
" print(\"Stuff:\", data.keys())\n",
"\n",
"dataset = ds.CocoDataset(DATA_DIR, annotation_file=KEYPOINT_FILE, task=\"Keypoint\", num_samples=1)\n",
"for data in dataset.create_dict_iterator():\n",
" print(\"Keypoint:\", data.keys())\n",
"\n",
"dataset = ds.CocoDataset(DATA_DIR, annotation_file=PANOPTIC_FILE, task=\"Panoptic\", num_samples=1)\n",
"for data in dataset.create_dict_iterator():\n",
" print(\"Panoptic:\", data.keys())\n",
"```\n",
"\n",
"输出结果：\n",
"\n",
"```text\n",
"Detection: dict_keys(['image', 'bbox', 'category_id', 'iscrowd'])\n",
"Stuff: dict_keys(['image', 'segmentation', 'iscrowd'])\n",
"Keypoint: dict_keys(['image', 'keypoints', 'num_keypoints'])\n",
"Panoptic: dict_keys(['image', 'bbox', 'category_id', 'iscrowd', 'area'])\n",
"```"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 特定格式数据集加载\n",
"\n",
"下面将介绍几种特定格式数据集文件的加载方式。\n",
"\n",
"### MindRecord数据格式\n",
"\n",
"MindRecord是MindSpore定义的一种数据格式，使用MindRecord能够获得更好的性能提升。\n",
"\n",
"> 阅读[数据格式转换](https://www.mindspore.cn/doc/programming_guide/zh-CN/r1.2/dataset_conversion.html)章节，了解如何将数据集转化为MindSpore数据格式。\n",
"\n",
"执行本例之前需下载对应的测试数据`test_mindrecord.zip`并解压到指定位置，执行如下命令："
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"./datasets/mindspore_dataset_loading/\n",
"├── test.mindrecord\n",
"└── test.mindrecord.db\n",
"\n",
"0 directories, 2 files\n"
]
}
],
"source": [
"!wget -N https://obs.dualstack.cn-north-4.myhuaweicloud.com/mindspore-website/notebook/datasets/test_mindrecord.zip\n",
"!unzip -o ./test_mindrecord.zip -d ./datasets/mindspore_dataset_loading/\n",
"!tree ./datasets/mindspore_dataset_loading/"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"下面的样例通过`MindDataset`接口加载MindRecord文件，并展示已加载数据的标签。"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"dict_keys(['chinese', 'english'])\n",
"dict_keys(['chinese', 'english'])\n",
"dict_keys(['chinese', 'english'])\n"
]
}
],
"source": [
"import mindspore.dataset as ds\n",
"\n",
"DATA_FILE = [\"./datasets/mindspore_dataset_loading/test.mindrecord\"]\n",
"mindrecord_dataset = ds.MindDataset(DATA_FILE)\n",
"\n",
"for data in mindrecord_dataset.create_dict_iterator(output_numpy=True):\n",
" print(data.keys())"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Manifest数据格式\n",
"\n",
"Manifest是华为ModelArts支持的数据格式文件，详细说明请参见[Manifest文档](https://support.huaweicloud.com/engineers-modelarts/modelarts_23_0009.html)。\n",
"\n",
"本次示例需下载测试数据`test_manifest.zip`并将其解压到指定位置，执行如下命令："
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"./datasets/mindspore_dataset_loading/test_manifest/\n",
"├── eval\n",
"│ ├── 1.JPEG\n",
"│ └── 2.JPEG\n",
"├── test_manifest.json\n",
"└── train\n",
" ├── 1.JPEG\n",
" └── 2.JPEG\n",
"\n",
"2 directories, 5 files\n"
]
}
],
"source": [
"!wget -N https://obs.dualstack.cn-north-4.myhuaweicloud.com/mindspore-website/notebook/datasets/test_manifest.zip\n",
"!unzip -o ./test_manifest.zip -d ./datasets/mindspore_dataset_loading/test_manifest/\n",
"!tree ./datasets/mindspore_dataset_loading/test_manifest/"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"下面的样例通过`ManifestDataset`接口加载Manifest文件`test_manifest.json`，并展示已加载数据的标签。"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"0\n",
"1\n"
]
}
],
"source": [
"import mindspore.dataset as ds\n",
"\n",
"DATA_FILE = \"./datasets/mindspore_dataset_loading/test_manifest/test_manifest.json\"\n",
"manifest_dataset = ds.ManifestDataset(DATA_FILE)\n",
"\n",
"for data in manifest_dataset.create_dict_iterator():\n",
" print(data[\"label\"])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### TFRecord数据格式\n",
"\n",
"TFRecord是TensorFlow定义的一种二进制数据文件格式。\n",
"\n",
"下面的样例通过`TFRecordDataset`接口加载TFRecord文件，并介绍了两种不同的数据集格式设定方案。"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"下载`tfrecord`测试数据`test_tftext.zip`并解压到指定位置，执行如下命令："
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"./datasets/mindspore_dataset_loading/test_tfrecord/\n",
"└── test_tftext.tfrecord\n",
"\n",
"0 directories, 1 file\n"
]
}
],
"source": [
"!wget -N https://obs.dualstack.cn-north-4.myhuaweicloud.com/mindspore-website/notebook/datasets/test_tftext.zip\n",
"!unzip -o ./test_tftext.zip -d ./datasets/mindspore_dataset_loading/test_tfrecord/\n",
"!tree ./datasets/mindspore_dataset_loading/test_tfrecord/"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"1. 传入数据集路径或TFRecord文件列表，本例使用`test_tftext.tfrecord`，创建`TFRecordDataset`对象。"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"dict_keys(['chinese', 'line', 'words'])\n",
"dict_keys(['chinese', 'line', 'words'])\n",
"dict_keys(['chinese', 'line', 'words'])\n"
]
}
],
"source": [
"import mindspore.dataset as ds\n",
"\n",
"DATA_FILE = \"./datasets/mindspore_dataset_loading/test_tfrecord/test_tftext.tfrecord\"\n",
"tfrecord_dataset = ds.TFRecordDataset(DATA_FILE)\n",
"\n",
"for tf_data in tfrecord_dataset.create_dict_iterator():\n",
" print(tf_data.keys())"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"2. 用户可以通过编写Schema文件或创建Schema对象，设定数据集格式及特征。\n",
"\n",
" - 编写Schema文件\n",
"\n",
" 将数据集格式和特征按JSON格式写入Schema文件。\n",
" \n",
" - `columns`：列信息字段，需要根据数据集的实际列名定义。上面的示例中，数据集有三组数据，其列均为`chinese`、`line`和`words`。\n",
"\n",
" 然后在创建`TFRecordDataset`时将Schema文件路径传入。"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"dict_values([Tensor(shape=[57], dtype=UInt8, value= [230, 177, 159, 229, 183, 158, 229, 184, 130, 233, 149, 191, 230, 177, 159, 229, 164, 167, 230, 161, 165, 229, 143, 130, \n",
" 229, 138, 160, 228, 186, 134, 233, 149, 191, 230, 177, 159, 229, 164, 167, 230, 161, 165, 231, 154, 132, 233, 128, 154, \n",
" 232, 189, 166, 228, 187, 170, 229, 188, 143]), Tensor(shape=[22], dtype=Int8, value= [ 71, 111, 111, 100, 32, 108, 117, 99, 107, 32, 116, 111, 32, 101, 118, 101, 114, 121, 111, 110, 101, 46]), Tensor(shape=[32], dtype=UInt8, value= [229, 165, 179, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 101, 118, 101, 114, 121, 111, 110, 101, \n",
" 99, 32, 32, 32, 32, 32, 32, 32])])\n",
"dict_values([Tensor(shape=[12], dtype=UInt8, value= [231, 148, 183, 233, 187, 152, 229, 165, 179, 230, 179, 170]), Tensor(shape=[19], dtype=Int8, value= [ 66, 101, 32, 104, 97, 112, 112, 121, 32, 101, 118, 101, 114, 121, 32, 100, 97, 121, 46]), Tensor(shape=[20], dtype=UInt8, value= [ 66, 101, 32, 32, 32, 104, 97, 112, 112, 121, 100, 97, 121, 32, 32, 98, 32, 32, 32, 32])])\n",
"dict_values([Tensor(shape=[48], dtype=UInt8, value= [228, 187, 138, 229, 164, 169, 229, 164, 169, 230, 176, 148, 229, 164, 170, 229, 165, 189, 228, 186, 134, 230, 136, 145, \n",
" 228, 187, 172, 228, 184, 128, 232, 181, 183, 229, 142, 187, 229, 164, 150, 233, 157, 162, 231, 142, 169, 229, 144, 167\n",
" ]), Tensor(shape=[20], dtype=Int8, value= [ 84, 104, 105, 115, 32, 105, 115, 32, 97, 32, 116, 101, 120, 116, 32, 102, 105, 108, 101, 46]), Tensor(shape=[16], dtype=UInt8, value= [ 84, 104, 105, 115, 116, 101, 120, 116, 102, 105, 108, 101, 97, 32, 32, 32])])\n"
]
}
],
"source": [
"import os\n",
"import json\n",
"\n",
"data_json = {\n",
" \"columns\": {\n",
" \"chinese\": {\n",
" \"type\": \"uint8\",\n",
" \"rank\": 1\n",
" },\n",
" \"line\" : {\n",
" \"type\": \"int8\",\n",
" \"rank\": 1\n",
" },\n",
" \"words\" : {\n",
" \"type\": \"uint8\",\n",
" \"rank\": 0\n",
" }\n",
" }\n",
" }\n",
"\n",
"if not os.path.exists(\"dataset_schema_path\"):\n",
" os.mkdir(\"dataset_schema_path\")\n",
"SCHEMA_DIR = \"dataset_schema_path/schema.json\"\n",
"with open(SCHEMA_DIR, \"w\") as f:\n",
" json.dump(data_json,f,indent=4)\n",
" \n",
"tfrecord_dataset = ds.TFRecordDataset(DATA_FILE, schema=SCHEMA_DIR)\n",
"\n",
"for tf_data in tfrecord_dataset.create_dict_iterator():\n",
" print(tf_data.values())"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"3. 创建Schema对象\n",
"\n",
"创建Schema对象，为其添加自定义字段，然后在创建数据集对象时传入。"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'chinese': Tensor(shape=[12], dtype=UInt8, value= [231, 148, 183, 233, 187, 152, 229, 165, 179, 230, 179, 170]), 'line': Tensor(shape=[19], dtype=UInt8, value= [ 66, 101, 32, 104, 97, 112, 112, 121, 32, 101, 118, 101, 114, 121, 32, 100, 97, 121, 46])}\n",
"{'chinese': Tensor(shape=[48], dtype=UInt8, value= [228, 187, 138, 229, 164, 169, 229, 164, 169, 230, 176, 148, 229, 164, 170, 229, 165, 189, 228, 186, 134, 230, 136, 145, \n",
" 228, 187, 172, 228, 184, 128, 232, 181, 183, 229, 142, 187, 229, 164, 150, 233, 157, 162, 231, 142, 169, 229, 144, 167\n",
" ]), 'line': Tensor(shape=[20], dtype=UInt8, value= [ 84, 104, 105, 115, 32, 105, 115, 32, 97, 32, 116, 101, 120, 116, 32, 102, 105, 108, 101, 46])}\n",
"{'chinese': Tensor(shape=[57], dtype=UInt8, value= [230, 177, 159, 229, 183, 158, 229, 184, 130, 233, 149, 191, 230, 177, 159, 229, 164, 167, 230, 161, 165, 229, 143, 130, \n",
" 229, 138, 160, 228, 186, 134, 233, 149, 191, 230, 177, 159, 229, 164, 167, 230, 161, 165, 231, 154, 132, 233, 128, 154, \n",
" 232, 189, 166, 228, 187, 170, 229, 188, 143]), 'line': Tensor(shape=[22], dtype=UInt8, value= [ 71, 111, 111, 100, 32, 108, 117, 99, 107, 32, 116, 111, 32, 101, 118, 101, 114, 121, 111, 110, 101, 46])}\n"
]
}
],
"source": [
"from mindspore import dtype as mstype\n",
"schema = ds.Schema()\n",
"schema.add_column('chinese', de_type=mstype.uint8)\n",
"schema.add_column('line', de_type=mstype.uint8)\n",
"tfrecord_dataset = ds.TFRecordDataset(DATA_FILE, schema=schema)\n",
"\n",
"for tf_data in tfrecord_dataset.create_dict_iterator():\n",
" print(tf_data)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"对比上述中的步骤2和步骤3，可以看出：\n",
"\n",
"|步骤|chinese|line|words\n",
"|:---|:---|:---|:---\n",
"| 2|UInt8 |Int8|UInt8\n",
"| 3|UInt8 |UInt8|\n",
"\n",
"\n",
"示例步骤2中的`columns`中数据由`chinese`（UInt8）、`line`（Int8）和`words`（UInt8）变为了示例步骤3中的`chinese`（UInt8）、`line`（UInt8），通过Schema对象，设定数据集的数据类型和特征，使得列中的数据类型和特征相应改变了。"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### NumPy数据格式\n",
"\n",
"如果所有数据已经读入内存，可以直接使用`NumpySlicesDataset`类将其加载。\n",
"\n",
"下面的样例分别介绍了通过`NumpySlicesDataset`加载arrays数据、 list数据和dict数据的方式。\n",
"\n",
"- 加载NumPy arrays数据"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[0.89286015 0.33197981] [0.33540785]\n",
"[0.82122912 0.04169663] [0.62251943]\n",
"[0.10765668 0.59505206] [0.43814143]\n",
"[0.52981736 0.41880743] [0.73588211]\n"
]
}
],
"source": [
"import numpy as np\n",
"import mindspore.dataset as ds\n",
"\n",
"np.random.seed(6)\n",
"features, labels = np.random.sample((4, 2)), np.random.sample((4, 1))\n",
"\n",
"data = (features, labels)\n",
"dataset = ds.NumpySlicesDataset(data, column_names=[\"col1\", \"col2\"], shuffle=False)\n",
"\n",
"for np_arr_data in dataset:\n",
" print(np_arr_data[0], np_arr_data[1])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"- 加载Python list数据"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[1 2]\n",
"[3 4]\n"
]
}
],
"source": [
"import mindspore.dataset as ds\n",
"\n",
"data1 = [[1, 2], [3, 4]]\n",
"\n",
"dataset = ds.NumpySlicesDataset(data1, column_names=[\"col1\"], shuffle=False)\n",
"\n",
"for np_list_data in dataset:\n",
" print(np_list_data[0])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"- 加载Python dict数据"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'col1': Tensor(shape=[], dtype=Int64, value= 1), 'col2': Tensor(shape=[], dtype=Int64, value= 3)}\n",
"{'col1': Tensor(shape=[], dtype=Int64, value= 2), 'col2': Tensor(shape=[], dtype=Int64, value= 4)}\n"
]
}
],
"source": [
"import mindspore.dataset as ds\n",
"\n",
"data1 = {\"a\": [1, 2], \"b\": [3, 4]}\n",
"\n",
"dataset = ds.NumpySlicesDataset(data1, column_names=[\"col1\", \"col2\"], shuffle=False)\n",
"\n",
"for np_dic_data in dataset.create_dict_iterator():\n",
" print(np_dic_data)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### CSV数据格式\n",
"\n",
"下面的样例通过`CSVDataset`加载CSV格式数据集文件，并展示了已加载数据的`keys`。\n",
"\n",
"下载测试数据`test_csv.zip`并解压到指定位置，执行如下命令："
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"./datasets/mindspore_dataset_loading/test_csv/\n",
"├── test1.csv\n",
"└── test2.csv\n",
"\n",
"0 directories, 2 files\n"
]
}
],
"source": [
"!wget -N https://obs.dualstack.cn-north-4.myhuaweicloud.com/mindspore-website/notebook/datasets/test_csv.zip\n",
"!unzip -o ./test_csv.zip -d ./datasets/mindspore_dataset_loading/test_csv/\n",
"!tree ./datasets/mindspore_dataset_loading/test_csv/"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"传入数据集路径或CSV文件列表，Text格式数据集文件的加载方式与CSV文件类似。"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"dict_keys(['a', 'b', 'c', 'd'])\n",
"dict_keys(['a', 'b', 'c', 'd'])\n",
"dict_keys(['a', 'b', 'c', 'd'])\n",
"dict_keys(['a', 'b', 'c', 'd'])\n"
]
}
],
"source": [
"import mindspore.dataset as ds\n",
"\n",
"DATA_FILE = [\"./datasets/mindspore_dataset_loading/test_csv/test1.csv\",\"./datasets/mindspore_dataset_loading/test_csv/test2.csv\"]\n",
"csv_dataset = ds.CSVDataset(DATA_FILE)\n",
"\n",
"for csv_data in csv_dataset.create_dict_iterator(output_numpy=True):\n",
" print(csv_data.keys())"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 自定义数据集加载\n",
"\n",
"对于目前MindSpore不支持直接加载的数据集，可以通过构造`GeneratorDataset`对象实现自定义方式的加载，或者将其转换成MindRecord数据格式。下面分别展示几种不同的自定义数据集加载方法，为了便于对比，生成的随机数据保持相同。\n",
"\n",
"### 构造数据集生成函数\n",
"\n",
"构造生成函数定义数据返回方式，再使用此函数构建自定义数据集对象。此方法适用于简单场景。"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[0.36510558 0.45120592] [0.78888122]\n",
"[0.49606035 0.07562207] [0.38068183]\n",
"[0.57176158 0.28963401] [0.16271622]\n",
"[0.30880446 0.37487617] [0.54738768]\n",
"[0.81585667 0.96883469] [0.77994068]\n"
]
}
],
"source": [
"import numpy as np\n",
"import mindspore.dataset as ds\n",
"\n",
"np.random.seed(58)\n",
"data = np.random.sample((5, 2))\n",
"label = np.random.sample((5, 1))\n",
"\n",
"def GeneratorFunc():\n",
" for i in range(5):\n",
" yield (data[i], label[i])\n",
"\n",
"dataset = ds.GeneratorDataset(GeneratorFunc, [\"data\", \"label\"])\n",
"\n",
"for item in dataset.create_dict_iterator():\n",
" print(item[\"data\"], item[\"label\"])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 构造可迭代的数据集类\n",
"\n",
"构造数据集类实现`__iter__`和`__next__`方法，再使用此类的对象构建自定义数据集对象。相比于直接定义生成函数，使用数据集类能够实现更多的自定义功能。"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[0.36510558 0.45120592] [0.78888122]\n",
"[0.49606035 0.07562207] [0.38068183]\n",
"[0.57176158 0.28963401] [0.16271622]\n",
"[0.30880446 0.37487617] [0.54738768]\n",
"[0.81585667 0.96883469] [0.77994068]\n"
]
}
],
"source": [
"import numpy as np\n",
"import mindspore.dataset as ds\n",
"\n",
"class IterDatasetGenerator:\n",
" def __init__(self):\n",
" np.random.seed(58)\n",
" self.__index = 0\n",
" self.__data = np.random.sample((5, 2))\n",
" self.__label = np.random.sample((5, 1))\n",
"\n",
" def __next__(self):\n",
" if self.__index >= len(self.__data):\n",
" raise StopIteration\n",
" else:\n",
" item = (self.__data[self.__index], self.__label[self.__index])\n",
" self.__index += 1\n",
" return item\n",
"\n",
" def __iter__(self):\n",
" self.__index = 0\n",
" return self\n",
"\n",
" def __len__(self):\n",
" return len(self.__data)\n",
"\n",
"dataset_generator = IterDatasetGenerator()\n",
"dataset = ds.GeneratorDataset(dataset_generator, [\"data\", \"label\"], shuffle=False)\n",
"\n",
"for data in dataset.create_dict_iterator():\n",
" print(data[\"data\"], data[\"label\"])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 构造可随机访问的数据集类\n",
"\n",
"构造数据集类实现`__getitem__`方法，再使用此类的对象构建自定义数据集对象。此方法可以用于实现分布式训练。"
]
},
{
"cell_type": "code",
"execution_count": 18,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[0.36510558 0.45120592] [0.78888122]\n",
"[0.49606035 0.07562207] [0.38068183]\n",
"[0.57176158 0.28963401] [0.16271622]\n",
"[0.30880446 0.37487617] [0.54738768]\n",
"[0.81585667 0.96883469] [0.77994068]\n"
]
}
],
"source": [
"import numpy as np\n",
"import mindspore.dataset as ds\n",
"\n",
"class GetDatasetGenerator:\n",
" def __init__(self):\n",
" np.random.seed(58)\n",
" self.__data = np.random.sample((5, 2))\n",
" self.__label = np.random.sample((5, 1))\n",
"\n",
" def __getitem__(self, index):\n",
" return (self.__data[index], self.__label[index])\n",
"\n",
" def __len__(self):\n",
" return len(self.__data)\n",
"\n",
"dataset_generator = GetDatasetGenerator()\n",
"dataset = ds.GeneratorDataset(dataset_generator, [\"data\", \"label\"], shuffle=False)\n",
"\n",
"for data in dataset.create_dict_iterator():\n",
" print(data[\"data\"], data[\"label\"])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"如果用户希望实现分布式训练，则需要在此方式的基础上，在采样器类中实现`__iter__`方法，每次返回采样数据的索引。需要补充的代码如下："
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[0.36510558 0.45120592] [0.78888122]\n",
"[0.57176158 0.28963401] [0.16271622]\n",
"[0.81585667 0.96883469] [0.77994068]\n"
]
}
],
"source": [
"import math\n",
"\n",
"class MySampler():\n",
" def __init__(self, dataset, local_rank, world_size):\n",
" self.__num_data = len(dataset)\n",
" self.__local_rank = local_rank\n",
" self.__world_size = world_size\n",
" self.samples_per_rank = int(math.ceil(self.__num_data / float(self.__world_size)))\n",
" self.total_num_samples = self.samples_per_rank * self.__world_size\n",
"\n",
" def __iter__(self):\n",
" indices = list(range(self.__num_data))\n",
" indices.extend(indices[:self.total_num_samples-len(indices)])\n",
" indices = indices[self.__local_rank:self.total_num_samples:self.__world_size]\n",
" return iter(indices)\n",
"\n",
" def __len__(self):\n",
" return self.samples_per_rank\n",
"\n",
"dataset_generator = GetDatasetGenerator()\n",
"sampler = MySampler(dataset_generator, local_rank=0, world_size=2)\n",
"dataset = ds.GeneratorDataset(dataset_generator, [\"data\", \"label\"], shuffle=False, sampler=sampler)\n",
"\n",
"for data in dataset.create_dict_iterator():\n",
" print(data[\"data\"], data[\"label\"])"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "MindSpore-1.1.1",
"language": "python",
"name": "mindspore-1.1.1"
},
"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
}