新建
快速开始 DeepSPIN｜训练NiO深度势能分子动力学模型
bxu@gscaep.ac.cn
赞 16
2
39
目录
数据集
AI4SCUP-CNS-BBB(v1)
快速开始 DeepSPIN｜训练NiO深度势能分子动力学模型
作者: Ben Xu bxu@gscaep.ac.cn , Teng Yang, Zhengtao Huang, Wenlong Tang
推荐镜像：deepspin:final
推荐计算资源：CPU
内容：本教程主要介绍DeepSPIN相关操作，包括：NiO数据集的构建、模型训练及模型调用等。
使用方式: 您可在 Bohrium Notebook 上直接运行。您可以点击界面上方蓝色按钮
开始连接，选择deepspin:final镜像及任何一款节点配置，稍等片刻即可运行。如您遇到任何问题，请联系 bohrium@dp.tech 。共享协议: 本作品采用知识共享署名-非商业性使用-相同方式共享 4.0 国际许可协议进行许可。
代码
文本
DeepSPIN应用框架主要包含三大部分：
- 建立数据集：当确定磁性体系的基态构型后，生成对自旋与晶格的随机扰动构型，采用DeltaSpin进行第一性原理计算，收集结果得到数据集。
- 训练模型：将自旋与晶格信息共同映射为体系能量，通过能量、原子力、磁性力建立损失函数，进行DeepSPIN模型的训练。
- 应用模型：将训练好的DeepSPIN模型冻结后，可以在LAMMPS中进行自旋—晶格模拟，包括：能量最小化和自旋—晶格动力学计算。
在建立数据集和训练模型环节，DeepSPIN采用了类似DP-GEN的主动学习策略：
- 构型空间采样：利用现有模型对生成的随机扰动构型进行预测，根据原子力和磁性力的预测偏差筛选待学习构型。
- 第一性原理计算：对待学习构型利用DeltaSpin进行第一性原理计算，然后合并到数据集中。
- 训练模型：在合并后的数据集上对现有模型继续训练。
- 重复1-3步，直至数据集覆盖所关心的构型空间，并且模型取得良好训练效果。
代码
文本
代码
文本
[1]
! git clone https://github.com/hztttt/DeepSPIN_Tutorial.git && rm DeepSPIN_Tutorial/LICENSE
Cloning into 'DeepSPIN_Tutorial'... remote: Enumerating objects: 1081, done. remote: Counting objects: 100% (275/275), done. remote: Compressing objects: 100% (174/174), done. remote: Total 1081 (delta 122), reused 250 (delta 98), pack-reused 806 Receiving objects: 100% (1081/1081), 517.32 MiB | 4.74 MiB/s, done. Resolving deltas: 100% (515/515), done. Updating files: 100% (1411/1411), done.
代码
文本
[3]
! ls
DeepSPIN_Tutorial deepspin.ipynb
代码
文本
让我们来查看下载的 DeepSPIN_Tutorial 文件夹：
代码
文本
[4]
! tree DeepSPIN_Tutorial -L 1
DeepSPIN_Tutorial |-- 00.data |-- 01.train |-- 02.lmp `-- 03.scripts 4 directories, 0 files
代码
文本
在 DeepSPIN_Tutorial 文件夹下有 00.data，01.train， 02.lmp 和 03.scritps共 四 个子文件夹。
- 00.data 文件夹用于存放训练和测试数据，
- 01.train 包含使用 DeepSPIN 训练模型的示例脚本，
- 02.lmp 包含用于分子动力学模拟的 LAMMPS 示例脚本。
- 03.scripts 包含用于提取数据脚本和lammps结果图。
让我们首先来查看 DeepSPIN_Tutorial/00.data 文件夹。
代码
文本
[5]
! tree DeepSPIN_Tutorial/00.data -L 1
DeepSPIN_Tutorial/00.data `-- DeltaSpin 1 directory, 0 files
代码
文本
DeepSPIN 的训练数据来源于第一性原理计算数据，包含原子类型、模拟晶格、原子坐标、原子力、磁性力和系统能量。
在 00.data 文件夹下仅有 DeltaSpin 文件夹，DeltaSpin 文件夹是通过使用 VASP 进行自适应自旋约束DFT计算 (DeltaSPIN) 获得的。本教程中我们已经为您完成了NiO的计算。
DeepSPIN 需要针对磁性原子生成其对应的赝原子，并沿用DeepMD-kit中的压缩数据格式。该数据格式在 DeePMD-kit 手册中有详细解释，可以在DeePMD-kit Github中找到。
这里，我们提供了 get-raw-from-log.py 脚本，针对结构中的磁性原子进行处理，并得到 *.raw 文件。
代码
文本
接下来，我们使用 get-raw-from-log.py 脚本将 DeltaSpin 中的NiO数据转换为 *.raw 格式的数据，保存到 raw 文件夹中。
代码
文本
[6]
! python DeepSPIN_Tutorial/03.scripts/get-raw-from-log.py --dir_frames DeepSPIN_Tutorial/00.data/DeltaSpin/ \
--natoms 32 --ntypes 2 --ntypes_spin 1 --len_spin 0.4 --file_log tmp_log \
&& mkdir DeepSPIN_Tutorial/00.data/raw && mv *.raw DeepSPIN_Tutorial/00.data/raw
read frames: 200 /data/DeepSPIN_Tutorial/00.data/DeltaSpin/1 converged frame 1 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/10 converged frame 2 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/100 converged frame 3 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/101 converged frame 4 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/102 converged frame 5 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/103 converged frame 6 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/104 converged frame 7 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/105 converged frame 8 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/106 converged frame 9 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/107 converged frame 10 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/108 converged frame 11 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/109 converged frame 12 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/11 converged frame 13 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/110 converged frame 14 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/111 converged frame 15 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/112 converged frame 16 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/113 converged frame 17 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/114 converged frame 18 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/115 converged frame 19 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/116 converged frame 20 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/117 converged frame 21 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/118 converged frame 22 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/119 converged frame 23 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/12 converged frame 24 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/120 converged frame 25 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/121 converged frame 26 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/122 converged frame 27 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/123 converged frame 28 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/124 converged frame 29 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/125 converged frame 30 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/126 converged frame 31 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/127 converged frame 32 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/128 converged frame 33 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/129 converged frame 34 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/13 converged frame 35 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/130 converged frame 36 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/131 converged frame 37 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/132 converged frame 38 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/133 converged frame 39 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/134 converged frame 40 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/135 converged frame 41 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/136 converged frame 42 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/137 converged frame 43 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/138 converged frame 44 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/139 converged frame 45 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/14 converged frame 46 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/140 converged frame 47 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/141 converged frame 48 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/142 converged frame 49 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/143 converged frame 50 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/144 converged frame 51 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/145 converged frame 52 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/146 converged frame 53 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/147 converged frame 54 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/148 converged frame 55 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/149 converged frame 56 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/15 converged frame 57 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/150 converged frame 58 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/151 converged frame 59 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/152 converged frame 60 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/153 converged frame 61 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/154 converged frame 62 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/155 converged frame 63 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/156 converged frame 64 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/157 converged frame 65 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/158 converged frame 66 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/159 converged frame 67 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/16 converged frame 68 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/160 converged frame 69 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/161 converged frame 70 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/162 converged frame 71 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/163 converged frame 72 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/164 converged frame 73 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/165 converged frame 74 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/166 converged frame 75 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/167 converged frame 76 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/168 converged frame 77 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/169 converged frame 78 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/17 converged frame 79 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/170 converged frame 80 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/171 converged frame 81 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/172 converged frame 82 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/173 converged frame 83 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/174 converged frame 84 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/175 converged frame 85 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/176 converged frame 86 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/177 converged frame 87 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/178 converged frame 88 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/179 converged frame 89 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/18 converged frame 90 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/180 converged frame 91 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/181 converged frame 92 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/182 converged frame 93 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/183 converged frame 94 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/184 converged frame 95 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/185 converged frame 96 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/186 converged frame 97 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/187 converged frame 98 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/188 converged frame 99 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/189 converged frame 100 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/19 converged frame 101 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/190 converged frame 102 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/191 converged frame 103 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/192 converged frame 104 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/193 converged frame 105 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/194 converged frame 106 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/195 converged frame 107 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/196 converged frame 108 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/197 converged frame 109 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/198 converged frame 110 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/199 converged frame 111 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/2 converged frame 112 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/20 converged frame 113 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/200 converged frame 114 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/21 converged frame 115 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/22 converged frame 116 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/23 converged frame 117 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/24 converged frame 118 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/25 converged frame 119 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/26 converged frame 120 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/27 converged frame 121 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/28 converged frame 122 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/29 converged frame 123 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/3 converged frame 124 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/30 converged frame 125 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/31 converged frame 126 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/32 converged frame 127 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/33 converged frame 128 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/34 converged frame 129 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/35 converged frame 130 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/36 converged frame 131 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/37 converged frame 132 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/38 converged frame 133 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/39 converged frame 134 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/4 converged frame 135 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/40 converged frame 136 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/41 converged frame 137 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/42 converged frame 138 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/43 converged frame 139 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/44 converged frame 140 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/45 converged frame 141 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/46 converged frame 142 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/47 converged frame 143 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/48 converged frame 144 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/49 converged frame 145 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/5 converged frame 146 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/50 converged frame 147 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/51 converged frame 148 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/52 converged frame 149 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/53 converged frame 150 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/54 converged frame 151 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/55 converged frame 152 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/56 converged frame 153 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/57 converged frame 154 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/58 converged frame 155 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/59 converged frame 156 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/6 converged frame 157 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/60 converged frame 158 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/61 converged frame 159 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/62 converged frame 160 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/63 converged frame 161 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/64 converged frame 162 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/65 converged frame 163 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/66 converged frame 164 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/67 converged frame 165 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/68 converged frame 166 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/69 converged frame 167 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/7 converged frame 168 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/70 converged frame 169 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/71 converged frame 170 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/72 converged frame 171 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/73 converged frame 172 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/74 converged frame 173 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/75 converged frame 174 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/76 converged frame 175 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/77 converged frame 176 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/78 converged frame 177 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/79 converged frame 178 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/8 converged frame 179 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/80 converged frame 180 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/81 converged frame 181 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/82 converged frame 182 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/83 converged frame 183 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/84 converged frame 184 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/85 converged frame 185 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/86 converged frame 186 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/87 converged frame 187 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/88 converged frame 188 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/89 converged frame 189 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/9 converged frame 190 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/90 converged frame 191 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/91 converged frame 192 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/92 converged frame 193 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/93 converged frame 194 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/94 converged frame 195 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/95 converged frame 196 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/96 converged frame 197 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/97 converged frame 198 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/98 converged frame 199 succeed /data/DeepSPIN_Tutorial/00.data/DeltaSpin/99 converged frame 200 succeed get raw files done
代码
文本
这里的参数含义为:
- natoms: 原子的总数
- ntypes: 原子类型数
- ntypes_spin: 磁性原子类型数
- len_spin: 磁性原子和其对应赝原子距离
- file_log: 计算结果文件
代码
文本
运行完上述脚本后，我们再查看一下 00.data 文件夹，其中产生了新的文件夹 raw，接下来，让我们查看下 raw 文件夹。
代码
文本
[7]
! tree DeepSPIN_Tutorial/00.data/ -L 1
DeepSPIN_Tutorial/00.data/ |-- DeltaSpin `-- raw 2 directories, 0 files
代码
文本
[8]
! tree DeepSPIN_Tutorial/00.data/raw -L 1
DeepSPIN_Tutorial/00.data/raw
|-- box.raw
|-- coord.raw
|-- energy.raw
|-- force.raw
`-- type.raw
0 directories, 5 files
代码
文本
让我们看下各个文件的作用。
box.raw: 模拟晶格大小。coord.raw: 原子的坐标。energy.raw: 构型的能量。force.raw: 原子受力（原子力和磁性力）。type.raw: 原子的类型（以整数表示）。
代码
文本
下面让我们看下这些文件
代码
文本
[9]
! head -n 2 DeepSPIN_Tutorial/00.data/raw/box.raw
10.24614096 -0.00097986 -0.00026854 8.53538475 5.66808049 0.00041704 8.53527626 2.57406695 5.05024529 10.24614096 -0.00097986 -0.00026854 8.53538475 5.66808049 0.00041704 8.53527626 2.57406695 5.05024529
代码
文本
这告诉我们模拟NiO晶格的尺寸，每一行为一个NiO构型，晶格常数顺序为: XX XY XZ YX YY YZ ZX ZY ZZ
代码
文本
[10]
! head -n 2 DeepSPIN_Tutorial/00.data/raw/coord.raw
-0.00180931 0.00050088 -0.00043977 4.26763813 1.28703347 2.52512265 4.26854324 2.82978444 0.00077193 8.53533050 4.12107372 2.52533117 5.12307048 -0.00048993 -0.00013427 9.39070861 1.28654354 2.52498838 9.39076286 2.83355031 0.00007425 13.65840098 4.12058379 2.52519690 6.82920049 2.06029189 1.26259845 11.09467432 3.35140447 3.78765628 11.09689287 4.89433214 1.26280697 15.36453100 6.18136562 3.78792962 11.94789609 2.05446187 1.26454751 16.21990910 3.34683544 3.78758683 16.21996335 4.89384221 1.26267270 20.48760148 6.18087568 3.78779535 3.41460025 1.03014595 0.63129922 7.68223837 2.31717942 3.15642187 7.68229262 3.86418619 0.63150775 11.94993075 5.15121967 3.15663039 8.53767073 1.02965602 0.63116495 12.80014594 2.30884985 3.15690006 12.80536310 3.86369626 0.63137348 17.07300123 5.15072974 3.15649612 10.24380074 3.09043784 1.89389767 14.51143887 4.37747132 4.41902032 14.51149312 5.92447809 1.89410619 18.77913124 7.21151156 4.41922884 15.36687122 3.08994791 1.89376340 19.63450935 4.37698139 4.41888605 19.63456360 5.92398816 1.89397192 23.90220172 7.21102163 4.41909457 -0.32379285 -0.23193611 0.04751487 3.94565436 1.05459658 2.57307623 3.94655961 2.59734756 0.04872651 8.21334672 3.88863706 2.57328582 4.80108686 -0.23292667 0.04782099 9.04648651 1.08368266 2.54392534 9.06877930 2.60111334 0.04802883 13.33641818 3.88814599 2.57315259 7.15118404 2.29272894 1.21464419 11.41665833 3.58384090 3.73970214 11.41887680 5.12676866 1.21485266 15.68651485 6.41380210 3.73997456 12.26987950 2.28689908 1.21659307 16.54189282 3.57927216 3.73963213 16.54194663 5.12627969 1.21471868 20.80958469 6.41331294 3.73983978 0.00000000 0.00000000 0.00000000 4.26763813 1.28703347 2.52512265 4.26769238 2.83404025 0.00020852 8.53533050 4.12107372 2.52533117 5.12307048 -0.00048993 -0.00013427 9.39070861 1.28654354 2.52498838 9.39076286 2.83355031 0.00007425 13.65840098 4.12058379 2.52519690 6.82920049 2.06029189 1.26259845 11.09683862 3.34732537 3.78772110 11.09689287 4.89433214 1.26280697 15.36919196 6.18594395 3.78867369 11.95227097 2.05980196 1.26246418 16.21990910 3.34683544 3.78758683 16.21996335 4.89384221 1.26267270 20.48760148 6.18087568 3.78779535 3.41460025 1.03014595 0.63129922 7.68223837 2.31717942 3.15642187 7.68644527 3.86234485 0.63414667 11.94790217 5.14755263 3.15610588 8.53767073 1.02965602 0.63116495 12.80530885 2.31668949 3.15628760 12.80536310 3.86369626 0.63137348 17.07300123 5.15072974 3.15649612 10.24380074 3.09043784 1.89389767 14.51289159 4.37802138 4.42129652 14.51149312 5.92447809 1.89410619 18.77913124 7.21151156 4.41922884 15.36687122 3.08994791 1.89376340 19.63450935 4.37698139 4.41888605 19.63456360 5.92398816 1.89397192 23.90220172 7.21102163 4.41909457 0.05836959 0.32871003 0.22032410 4.29062588 1.60095931 2.77194674 4.33845596 3.17105126 0.20371797 8.60609438 4.45808523 2.72883968 5.19383358 0.33652234 0.20337326 9.46147168 1.62355562 2.72849623 9.46152621 3.17056204 0.20358258 13.72916458 4.45759594 2.72870445 6.74289538 1.71603272 1.07810713 11.02607408 3.01031328 3.58421377 11.02612825 4.55732012 1.05929956 15.29842785 5.84893112 3.58516744 11.88150707 1.72279042 1.05895572 16.14914531 3.00982395 3.58407825 16.12832793 4.56484764 1.05442880 20.41683828 5.84386340 3.58428788
代码
文本
coord.raw 每一行给出一个构型中各个原子的坐标信息，在我们NiO例子中，包含16个Ni原子、16个O原子和16个Ni赝原子，每个原子坐标分为x、y、z三个方向。故每行有个数据信息。
例如，第一行表示第一个构型的原子坐标为：
代码
文本
[11]
! head -n 2 DeepSPIN_Tutorial/00.data/raw/energy.raw
-166.31078603 -166.30895240
代码
文本
energy.raw 给出了每个NiO构型的能量信息。
代码
文本
[12]
! head -n 2 DeepSPIN_Tutorial/00.data/raw/force.raw
0.01879422 -0.01310723 0.00053585 -0.00867498 -0.00351402 0.00735577 -0.02115941 0.04838151 -0.00112284 -0.00237322 -0.00019873 -0.00294130 -0.00998428 -0.02342055 -0.00552161 -0.00120521 0.00129795 0.00928352 0.00101228 0.00948972 -0.00169888 -0.01762036 -0.02519123 -0.00830600 0.00211149 0.00010605 0.00288022 0.01964222 -0.05029723 0.00177637 0.00776651 -0.00930078 -0.01162228 -0.00206736 0.00212962 -0.00028495 0.04090250 0.05745414 -0.01841154 -0.00132879 -0.00347661 -0.00093944 -0.00415174 -0.00252736 -0.00222240 -0.00762998 -0.00599896 0.02159585 -0.00362291 -0.00698837 0.00148408 -0.00249927 0.00043358 -0.00353608 -0.00291713 -0.00146677 0.00095270 0.01191511 -0.00918452 -0.00348352 -0.00357133 0.00273220 0.00090007 0.04076518 0.07750834 -0.00498085 0.00310477 -0.01813462 0.01791405 -0.00081734 -0.00120353 -0.00047618 -0.01194237 0.00100254 0.00096253 0.00083673 -0.00906895 -0.00636045 0.00242878 0.00209788 0.00135863 -0.01469314 0.00798044 0.00208763 -0.00321585 -0.00134582 0.00514050 -0.01295942 0.00005568 0.00071536 -0.01689081 -0.02113713 -0.00408999 0.00004510 -0.00510725 0.00105521 -0.00687828 -0.00508654 0.00111571 -0.00778939 -0.00567344 0.00119009 -0.00783775 -0.00574006 0.00122533 -0.00721428 -0.00536331 0.00118964 -0.00797998 -0.00578059 0.00119198 -0.00560158 -0.01209434 0.00653353 -0.00738806 -0.00547542 0.00119076 -0.00829254 -0.00597961 0.00121498 0.00783178 0.00544564 -0.00103315 0.00783223 0.00546154 -0.00105336 0.01098251 0.00343667 0.00154793 0.00783149 0.00547896 -0.00102643 0.00781645 0.00544591 -0.00101689 0.01056069 0.00311121 0.00161941 0.00790725 0.00550629 -0.00105643 0.00776418 0.00541395 -0.00100871 0.00881232 0.00498696 0.00765984 -0.00168379 -0.01036499 0.00148299 0.00356926 -0.00344972 -0.00522600 0.01116370 0.00321837 0.00764964 0.00151391 0.00420705 0.00170006 0.00947883 0.00194861 -0.00416453 0.01341142 -0.00544782 0.00294587 0.00612178 0.00432528 0.00928090 -0.00014588 -0.00268799 -0.00103801 -0.00570170 -0.00745257 0.00094012 -0.00327739 -0.00319620 -0.00240736 -0.04257633 -0.05053956 0.00040279 -0.00174015 0.00031966 -0.00005634 -0.00674906 0.00945213 0.00219367 0.00340891 -0.00219712 0.00272354 -0.00106642 -0.00280197 -0.00057688 0.00275606 0.00119584 0.00538909 0.00419412 0.00559268 0.00366251 -0.03388881 0.02332775 -0.01902531 0.02115059 0.03441713 0.00498157 0.00268759 -0.01022120 0.00103565 0.00693075 -0.00517399 -0.00047301 0.00258486 -0.00152254 -0.00310312 -0.00001321 0.00231529 -0.00315317 0.00065551 -0.00515279 0.00010250 -0.00603027 0.00927219 -0.02379360 0.00183602 0.00104229 0.01038856 -0.00069362 0.00143452 -0.00185613 0.00053279 0.00446260 0.00249545 -0.00452544 -0.00048502 0.00060731 -0.00110776 0.00273685 -0.00026377 0.00839145 -0.00356174 -0.00050484 0.00560107 0.01079881 0.00016640 0.01227049 0.01682548 0.00217967 0.00195930 0.00864969 0.00504602 0.00189845 0.00793307 0.00462204 0.00194383 0.00833891 0.00493222 0.00183569 0.00756149 0.00438075 0.00385980 0.00669713 0.00472823 0.00354195 0.00965812 0.00344652 0.00211836 -0.00722832 -0.01076735 0.00263097 -0.00761214 -0.01021261 -0.00151233 -0.00813753 -0.00518199 -0.00146896 -0.00798242 -0.00514460 -0.00149310 -0.00804600 -0.00516192 -0.00145064 -0.00801416 -0.00512682 0.00577236 -0.01044832 -0.01006806 -0.00033856 -0.00741927 -0.00694383
代码
文本
force.raw 给出了每个NiO构型中每个原子的受力情况，每一行代表一个构型，每个原子的受力分为x、y、z三个方向，其中，受力分为Ni原子的原子力、O原子的原子力以及Ni原子的磁性力。
例如，第一行表示第一个构型的受力为：
代码
文本
[13]
! head -n 2 DeepSPIN_Tutorial/00.data/raw/type.raw
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
代码
文本
type.raw 文件给出了NiO结构的原子分布信息，这告诉我们这个例子中有48个原子(32个真实原子和16个赝原子)，其中16个Ni原子由类型"0"表示，16个O原子由类型"1"表示，16个Ni赝原子由类型"2"表示。
代码
文本
[14]
import dpdata
import numpy as np
# 读入 DEEPMD/RAW 格式的数据
data = dpdata.LabeledSystem('DeepSPIN_Tutorial/00.data/raw/', fmt = 'deepmd/raw')
print('# 数据包含%d帧' % len(data))
# 随机选择40个索引作为验证集数据
np.random.seed(1)
index_validation = np.random.choice(200, size=40, replace=False)
# 其他索引作为训练集数据
index_training = list(set(range(200))-set(index_validation))
data_training = data.sub_system(index_training)
data_validation = data.sub_system(index_validation)
# 将所有训练数据放入文件夹"training_data"中
data_training.to_deepmd_npy('DeepSPIN_Tutorial/00.data/training_data')
# 将所有验证数据放入文件夹"validation_data"中
data_validation.to_deepmd_npy('DeepSPIN_Tutorial/00.data/validation_data')
print('# 训练数据包含%d帧' % len(data_training))
print('# 验证数据包含%d帧' % len(data_validation))
! rm DeepSPIN_Tutorial/00.data/validation_data/type_map.raw DeepSPIN_Tutorial/00.data/training_data/type_map.raw
# 数据包含200帧 # 训练数据包含160帧 # 验证数据包含40帧
代码
文本
这里，我们将200个构型划分为160个训数据和40个验证数据。
代码
文本
让我们再查看一下 00.data 文件夹，其中产生了新的文件，分别是 DeepSPIN 训练所需的训练集和验证集。
代码
文本
[15]
! tree DeepSPIN_Tutorial/00.data/ -L 1
DeepSPIN_Tutorial/00.data/ |-- DeltaSpin |-- raw |-- training_data `-- validation_data 4 directories, 0 files
代码
文本
[16]
! tree DeepSPIN_Tutorial/00.data/training_data -L 1
DeepSPIN_Tutorial/00.data/training_data |-- set.000 `-- type.raw 1 directory, 1 file
代码
文本
这些文件的作用如下：
set.000：是一个目录，包含压缩格式的数据（NumPy压缩数组）。type.raw：是一个文件，包含原子的类型（以整数表示）。
让我们来看一下这些文件：
代码
文本
[17]
! cat DeepSPIN_Tutorial/00.data/training_data/type.raw
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
代码
文本
在type.raw中，每一行给出了一个原子类型，其中16个Ni原子由类型"0"表示，16个O原子由类型"1"表示，16个Ni赝原子由类型"2"表示。
代码
文本
代码
文本
[18]
! cd DeepSPIN_Tutorial/01.train/ && cat input.json
{
"_comment": " model parameters",
"model": {
"type_map": ["Ni", "O"],
"descriptor" : {
"type": "se_e2_a",
"sel": [50, 50],
"rcut_smth": 4.6,
"rcut": 5.6,
"neuron": [25, 50, 100],
"resnet_dt": false,
"axis_neuron": 16,
"seed": 22,
"_comment": " that's all"
},
"fitting_net" : {
"neuron": [240, 240, 240],
"resnet_dt": true,
"seed": 22,
"_comment": " that's all"
},
"spin" : {
"use_spin": [true, false],
"virtual_len": [0.4],
"spin_norm": [1.2737],
"_comment": " that's all"
},
"_comment": " that's all"
},
"learning_rate" : {
"type": "exp",
"decay_steps": 5000,
"start_lr": 1.00e-3,
"stop_lr": 3.51e-8,
"_comment": " that's all"
},
"loss" : {
"type": "ener_spin",
"start_pref_e": 0.02,
"limit_pref_e": 1,
"start_pref_fr": 1000,
"limit_pref_fr": 1.0,
"start_pref_fm": 1000,
"limit_pref_fm": 5.0,
"start_pref_v": 0,
"limit_pref_v": 0,
"_comment": " that's all"
},
"training" : {
"training_data": {
"systems": ["../00.data/training_data"],
"batch_size": "auto",
"_comment": "that's all"
},
"validation_data": {
"systems": ["../00.data/validation_data"],
"batch_size": "auto",
"numb_btch": 1,
"_comment": "that's all"
},
"numb_steps": 1000000,
"seed": 22,
"disp_file": "lcurve.out",
"disp_freq": 1000,
"save_freq": 100000,
"_comment": "that's all"
},
"_comment": "that's all"
}
代码
文本
在模型部分，指定了嵌入和拟合网络及磁性的参数。
"model": {
"type_map": ["Ni", "O"],
"descriptor" : {
"type": "se_e2_a",
"sel": [50, 50],
"rcut_smth": 4.6,
"rcut": 5.6,
"neuron": [25, 50, 100],
"resnet_dt": false,
"axis_neuron": 16,
"seed": 1,
"_comment": " that's all"
},
"fitting_net" : {
"neuron": [240, 240, 240],
"resnet_dt": true,
"seed": 1,
"_comment": " that's all"
},
"spin" : {
"use_spin": [true, false],
"virtual_len": [0.4],
"spin_norm": [1.2737],
"_comment": " that's all"
},
"_comment": " that's all"
},
部分参数的解释如下： 部分参数的解释如下：
| 参数 | 解释 |
|---|---|
| type_map | 每种原子的名称 |
| descriptor > type | 描述类型 |
| descriptor > rcut | 截断半径 |
| descriptor > rcut_smth | 平滑开始的位置 |
| descriptor > sel | 切割半径内第i种原子的最大数目 |
| descriptor > neuron | 嵌入神经网络的大小 |
| descriptor > axis_neuron | G矩阵的子矩阵大小(嵌入矩阵) |
| fitting_net > neuron | 拟合神经网络的大小 |
| spin > use_spin | 原子是否具有磁性 |
| spin > virtual_len | 赝原子和真实原子距离 |
| spin > spin_norm | 磁性原子磁矩模长 |
使用se_e2_a描述符来训练DP模型。neurons的参数将描述符和拟合网络的大小分别设置为[25, 50, 100]和[240, 240, 240]。本地环境中的组成部分会在从4.6到5.6 Å的范围内平滑地趋于零。
使用spin的参数控制磁性原子的信息，该例中，Ni原子具有磁性，O原子不具有磁性，因此，设置use_spin的参数为[true, false]，第一个值true表示对Ni原子打开磁性，第二个值false表示对O原子不打开磁性。后续virtual_len和spin_norm参数需要针对磁性的原子进行设置。本次NiO模型中，设置了Ni原子的距离为0.4 Å，磁矩大小为1.2737 uB。
以下是指定学习率和损失函数的参数。
"learning_rate" : {
"type": "exp",
"decay_steps": 5000,
"start_lr": 1.00e-3,
"stop_lr": 3.51e-8,
"_comment": " that's all"
},
"loss" : {
"type": "ener_spin",
"start_pref_e": 0.02,
"limit_pref_e": 200,
"start_pref_fr": 1000,
"limit_pref_fr": 1.0,
"start_pref_fm": 1000,
"limit_pref_fm": 5.0,
"start_pref_v": 0,
"limit_pref_v": 0,
"_comment": " that's all"
},
在损失函数中，pref_e从 0.02 逐渐增加到 200，pref_fr从 1000 逐渐减小到 1，pref_fm从 1000 逐渐减小到 5，这意味着原子力和磁性力项在开始时占主导地位，而能量项在最后变得重要。这种策略非常有效，并减少了总的训练时间。pref_v设置为 0，表示训练过程中不包括压力数据。起始学习率、终止学习率和衰减步数分别设置为 0.001、 和 5000。模型训练 1000000 步。
训练参数如下所示：
"training" : {
"training_data": {
"systems": ["../00.data/training_data"],
"batch_size": "auto",
"_comment": "that's all"
},
"validation_data":{
"systems": ["../00.data/validation_data/"],
"batch_size": "auto",
"numb_btch": 1,
"_comment": "that's all"
},
"numb_steps": 1000000,
"seed": 10,
"disp_file": "lcurve.out",
"disp_freq": 1000,
"save_freq": 100000,
},
代码
文本
[19]
# ########## Time Warning: 13 hours+ ##########
! cd DeepSPIN_Tutorial/01.train/ && dp train input.json
2023-06-29 14:22:53.013178: I tensorflow/tsl/cuda/cudart_stub.cc:28] Could not find cuda drivers on your machine, GPU will not be used. 2023-06-29 14:22:53.474420: I tensorflow/tsl/cuda/cudart_stub.cc:28] Could not find cuda drivers on your machine, GPU will not be used. 2023-06-29 14:22:53.475835: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations. To enable the following instructions: AVX2 AVX512F FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags. 2023-06-29 14:22:54.874845: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/tensorflow/python/compat/v2_compat.py:107: disable_resource_variables (from tensorflow.python.ops.variable_scope) is deprecated and will be removed in a future version. Instructions for updating: non-resource variables are not supported in the long term WARNING:root:To get the best performance, it is recommended to adjust the number of threads by setting the environment variables OMP_NUM_THREADS, TF_INTRA_OP_PARALLELISM_THREADS, and TF_INTER_OP_PARALLELISM_THREADS. See https://deepmd.rtfd.io/parallelism/ for more information. DEEPMD INFO Calculate neighbor statistics... (add --skip-neighbor-stat to skip this step) 2023-06-29 14:23:00.618748: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:353] MLIR V1 optimization pass is not enabled # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: ghost idx out of upper bound (ignored if warned for more than 10 times) # warning: ghost idx out of upper bound (ignored if warned for more than 10 times) # warning: ghost idx out of upper bound (ignored if warned for more than 10 times) # warning: ghost idx out of upper bound (ignored if warned for more than 10 times) # warning: ghost idx out of upper bound (ignored if warned for more than 10 times) # warning: ghost idx out of upper bound (ignored if warned for more than 10 times) # warning: ghost idx out of upper bound (ignored if warned for more than 10 times) # warning: ghost idx out of upper bound (ignored if warned for more than 10 times) # warning: ghost idx out of upper bound (ignored if warned for more than 10 times) # warning: ghost idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) DEEPMD INFO training data with min nbor dist: 0.39999999202726677 DEEPMD INFO training data with max nbor size: [45 42 45] DEEPMD INFO _____ _____ __ __ _____ _ _ _ DEEPMD INFO | __ \ | __ \ | \/ || __ \ | | (_)| | DEEPMD INFO | | | | ___ ___ | |__) || \ / || | | | ______ | | __ _ | |_ DEEPMD INFO | | | | / _ \ / _ \| ___/ | |\/| || | | ||______|| |/ /| || __| DEEPMD INFO | |__| || __/| __/| | | | | || |__| | | < | || |_ DEEPMD INFO |_____/ \___| \___||_| |_| |_||_____/ |_|\_\|_| \__| DEEPMD INFO Please read and cite: DEEPMD INFO Wang, Zhang, Han and E, Comput.Phys.Comm. 228, 178-184 (2018) DEEPMD INFO Zeng et al, arXiv:2304.09409 DEEPMD INFO See https://deepmd.rtfd.io/credits/ for details. DEEPMD INFO installed to: /root/soft/deepspin/deepmd-kit/_skbuild/linux-x86_64-3.10/cmake-install DEEPMD INFO source : v2.2.2 DEEPMD INFO source brach: master DEEPMD INFO source commit: 92ca097b DEEPMD INFO source commit at: 2023-05-24 13:45:03 +0800 DEEPMD INFO build float prec: double DEEPMD INFO build variant: cpu DEEPMD INFO build with tf inc: /opt/mamba/lib/python3.10/site-packages/tensorflow/include;/opt/mamba/lib/python3.10/site-packages/tensorflow/include DEEPMD INFO build with tf lib: DEEPMD INFO ---Summary of the training--------------------------------------- DEEPMD INFO running on: bohrium-14073-1024925 DEEPMD INFO computing device: cpu:0 DEEPMD INFO CUDA_VISIBLE_DEVICES: unset DEEPMD INFO Count of visible GPU: 0 DEEPMD INFO num_intra_threads: 0 DEEPMD INFO num_inter_threads: 0 DEEPMD INFO ----------------------------------------------------------------- DEEPMD INFO ---Summary of DataSystem: training ----------------------------------------------- DEEPMD INFO found 1 system(s): DEEPMD INFO system natoms bch_sz n_bch prob pbc DEEPMD INFO ../00.data/training_data 48 1 160 1.000 T DEEPMD INFO -------------------------------------------------------------------------------------- DEEPMD INFO ---Summary of DataSystem: validation ----------------------------------------------- DEEPMD INFO found 1 system(s): DEEPMD INFO system natoms bch_sz n_bch prob pbc DEEPMD INFO ../00.data/validation_data 48 1 40 1.000 T DEEPMD INFO -------------------------------------------------------------------------------------- DEEPMD INFO training without frame parameter DEEPMD INFO data stating... (this step may take long time) DEEPMD INFO built lr WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/tensorflow/python/util/dispatch.py:1176: accumulate_n (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version. Instructions for updating: Use `tf.math.add_n` Instead WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/tensorflow/python/util/dispatch.py:1176: accumulate_n (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version. Instructions for updating: Use `tf.math.add_n` Instead DEEPMD INFO built network DEEPMD INFO built training WARNING:root:To get the best performance, it is recommended to adjust the number of threads by setting the environment variables OMP_NUM_THREADS, TF_INTRA_OP_PARALLELISM_THREADS, and TF_INTER_OP_PARALLELISM_THREADS. See https://deepmd.rtfd.io/parallelism/ for more information. DEEPMD INFO initialize model from scratch DEEPMD INFO start training at lr 1.00e-03 (== 1.00e-03), decay_step 5000, decay_rate 0.950006, final lr will be 3.51e-08 WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/deepmd/train/trainer.py:1477: py_func (from tensorflow.python.ops.script_ops) is deprecated and will be removed in a future version. Instructions for updating: tf.py_func is deprecated in TF V2. Instead, there are two options available in V2. - tf.py_function takes a python function which manipulates tf eager tensors instead of numpy arrays. It's easy to convert a tf eager tensor to an ndarray (just call tensor.numpy()) but having access to eager tensors means `tf.py_function`s can use accelerators such as GPUs as well as being differentiable using a gradient tape. - tf.numpy_function maintains the semantics of the deprecated tf.py_func (it is not differentiable, and manipulates numpy arrays). It drops the stateful argument making all functions stateful. WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/deepmd/train/trainer.py:1477: py_func (from tensorflow.python.ops.script_ops) is deprecated and will be removed in a future version. Instructions for updating: tf.py_func is deprecated in TF V2. Instead, there are two options available in V2. - tf.py_function takes a python function which manipulates tf eager tensors instead of numpy arrays. It's easy to convert a tf eager tensor to an ndarray (just call tensor.numpy()) but having access to eager tensors means `tf.py_function`s can use accelerators such as GPUs as well as being differentiable using a gradient tape. - tf.numpy_function maintains the semantics of the deprecated tf.py_func (it is not differentiable, and manipulates numpy arrays). It drops the stateful argument making all functions stateful. DEEPMD INFO batch 1000 training time 47.98 s, testing time 0.04 s, total wall time 49.23 s DEEPMD INFO batch 2000 training time 45.86 s, testing time 0.04 s, total wall time 46.27 s ^C Traceback (most recent call last): File "/opt/mamba/bin/dp", line 8, in <module> sys.exit(main()) File "/opt/mamba/lib/python3.10/site-packages/deepmd/entrypoints/main.py", line 623, in main train_dp(**dict_args) File "/opt/mamba/lib/python3.10/site-packages/deepmd/entrypoints/train.py", line 164, in train _do_work(jdata, run_opt, is_compress) File "/opt/mamba/lib/python3.10/site-packages/deepmd/entrypoints/train.py", line 278, in _do_work model.train(train_data, valid_data) File "/opt/mamba/lib/python3.10/site-packages/deepmd/train/trainer.py", line 998, in train _, next_train_batch_list = run_sess( File "/opt/mamba/lib/python3.10/site-packages/deepmd/utils/sess.py", line 30, in run_sess return sess.run(*args, **kwargs) File "/opt/mamba/lib/python3.10/site-packages/tensorflow/python/client/session.py", line 968, in run result = self._run(None, fetches, feed_dict, options_ptr, File "/opt/mamba/lib/python3.10/site-packages/tensorflow/python/client/session.py", line 1191, in _run results = self._do_run(handle, final_targets, final_fetches, File "/opt/mamba/lib/python3.10/site-packages/tensorflow/python/client/session.py", line 1371, in _do_run return self._do_call(_run_fn, feeds, fetches, targets, options, File "/opt/mamba/lib/python3.10/site-packages/tensorflow/python/client/session.py", line 1378, in _do_call return fn(*args) File "/opt/mamba/lib/python3.10/site-packages/tensorflow/python/client/session.py", line 1361, in _run_fn return self._call_tf_sessionrun(options, feed_dict, fetch_list, File "/opt/mamba/lib/python3.10/site-packages/tensorflow/python/client/session.py", line 1454, in _call_tf_sessionrun return tf_session.TF_SessionRun_wrapper(self._session, options, feed_dict, KeyboardInterrupt
代码
文本
屏幕上会显示数据系统的信息，例如：
DEEPMD INFO ---Summary of DataSystem: training -----------------------------------------------
DEEPMD INFO found 1 system(s):
DEEPMD INFO system natoms bch_sz n_bch prob pbc
DEEPMD INFO ../00.data/training_data 48 1 50 1.000 T
DEEPMD INFO --------------------------------------------------------------------------------------
DEEPMD INFO ---Summary of DataSystem: validation -----------------------------------------------
DEEPMD INFO found 1 system(s):
DEEPMD INFO system natoms bch_sz n_bch prob pbc
DEEPMD INFO ../00.data/validation_data 48 1 9 1.000 T
DEEPMD INFO --------------------------------------------------------------------------------------
以及该训练的起始和最终学习率：
DEEPMD INFO start training at lr 1.00e-03 (== 1.00e-03), decay_step 50, decay_rate 0.950006, final lr will be 3.51e-08
如果一切正常，您将看到每 1000 batch 打印的信息，例如：
DEEPMD INFO batch 1000 training time 49.18 s, testing time 0.21 s, total wall time 50.79 s
DEEPMD INFO batch 2000 training time 47.20 s, testing time 0.21 s, total wall time 47.78 s
DEEPMD INFO batch 3000 training time 47.07 s, testing time 0.21 s, total wall time 47.64 s
DEEPMD INFO batch 4000 training time 46.87 s, testing time 0.23 s, total wall time 47.47 s
DEEPMD INFO batch 5000 training time 47.07 s, testing time 0.21 s, total wall time 47.65 s
DEEPMD INFO batch 6000 training time 47.15 s, testing time 0.20 s, total wall time 47.72 s
DEEPMD INFO batch 7000 training time 47.28 s, testing time 0.21 s, total wall time 47.86 s
DEEPMD INFO batch 8000 training time 47.02 s, testing time 0.23 s, total wall time 47.61 s
它们显示了训练和测试时间计数。在每 100000 batch 结束时，模型将保存在 Tensorflow 的 checkpoint 文件 model.ckpt 中。
同时，训练和测试误差将在文件lcurve.out中呈现。
# step rmse_val rmse_trn rmse_e_val rmse_e_trn rmse_fr_val rmse_fr_trn rmse_fm_val rmse_fm_trn lr
0 8.42e-01 1.50e+00 7.93e-02 7.93e-02 2.43e-02 2.27e-02 1.02e-02 4.15e-02 3.0e-03
1000 4.42e-01 5.46e-01 1.67e-03 1.49e-03 9.05e-03 1.25e-02 1.03e-02 1.19e-02 3.0e-03
2000 4.17e-01 3.92e-01 2.03e-02 2.07e-02 7.75e-03 8.39e-03 1.05e-02 9.09e-03 3.0e-03
3000 4.16e-01 4.61e-01 2.66e-02 2.62e-02 6.81e-03 8.14e-03 1.10e-02 1.21e-02 3.0e-03
4000 3.88e-01 3.69e-01 4.38e-02 4.39e-02 5.17e-03 5.40e-03 1.09e-02 1.03e-02 3.0e-03
5000 3.35e-01 2.34e-01 4.87e-03 4.86e-03 4.83e-03 2.17e-03 9.59e-03 7.31e-03 2.8e-03
6000 3.66e-01 3.88e-01 2.82e-03 2.93e-03 4.09e-03 5.53e-03 1.12e-02 1.14e-02 2.8e-03
7000 3.70e-01 3.00e-01 5.74e-03 5.98e-03 5.85e-03 4.49e-03 1.05e-02 8.68e-03 2.8e-03
8000 3.35e-01 1.30e+00 2.18e-02 2.19e-02 4.38e-03 1.24e-02 9.83e-03 4.05e-02 2.8e-03
该文件包含 10 列，从左到右依次是：
- 训练步数
- 验证损失
- 训练损失
- 能量的均方根（RMS）验证误差
- 能量的 RMS 训练误差
- 原子力的 RMS 验证误差
- 原子力的 RMS 训练误差
- 磁性力的 RMS 验证误差
- 磁性力的 RMS 训练误差
- 学习率
学习率是机器学习中的一个重要概念。在 DP 模型中，学习率会经历一个 从大到小指数衰减的过程。这样既能保证模型收敛的效率，也能保证模型的精度。因此在学习率的参数中，有起始学习率(start_lr）和结束学习率（end_rate） 两种。在上面的例子中，我们将起始学习率、结束学习率和学习率的衰减步长分别设置为 0.001，3.51e-8，和 50，那么模型学习率会从 0.001 开始，每 50 步降低一点，直到降低到 3.51e-8（或者训练结束）为止。
代码
文本
我们来看一下 lcurve.out 文件的初始与结束两行。
代码
文本
[20]
! cd DeepSPIN_Tutorial/01.train/ && head -n 2 lcurve_bak.out && tail -n 2 lcurve_bak.out
# step rmse_val rmse_trn rmse_e_val rmse_e_trn rmse_fr_val rmse_fr_trn rmse_fm_val rmse_fm_trn lr
0 6.60e-01 8.06e-01 7.95e-02 7.93e-02 2.01e-02 2.49e-02 4.88e-03 4.80e-03 1.0e-03
1000 9.67e-02 2.93e-01 1.07e-02 1.10e-02 2.30e-03 8.56e-03 1.98e-03 3.52e-03 1.0e-03
2000 1.49e-01 2.13e-01 3.11e-03 3.14e-03 4.69e-03 6.64e-03 2.50e-04 1.07e-03 1.0e-03
代码
文本
下面我们对损失函数进行可视化展示
代码
文本
[42]
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
with open("./DeepSPIN_Tutorial/01.train/lcurve_bak.out") as f:
headers = f.readline().split()[1:]
lcurve = pd.DataFrame(np.loadtxt("./DeepSPIN_Tutorial/01.train/lcurve_bak.out"), columns=headers)
legends = ["rmse_e_val", "rmse_fr_val" , "rmse_fm_val" ]
legends_label = ["rmse_e_val (eV)", "rmse_fr_val (eV/A)" , "rmse_fm_val (eV/uB)" ]
fig, ax = plt.subplots()
for legend in legends:
ax.plot(lcurve["step"], lcurve[legend], label = legends_label[legends.index(legend)], linewidth=1)
ax.set_yscale('log')
ax.legend()
ax.set_xlabel("Training steps")
ax.set_ylabel("Loss")
plt.show()
代码
文本
[3]
# # 进入 DeepSPIN_Tutorial/01.train/ 训练文件夹并冻结模型
! cd DeepSPIN_Tutorial/01.train/ && dp freeze -o graph.pb
2023-06-29 18:21:28.951520: I tensorflow/tsl/cuda/cudart_stub.cc:28] Could not find cuda drivers on your machine, GPU will not be used. 2023-06-29 18:21:29.007263: I tensorflow/tsl/cuda/cudart_stub.cc:28] Could not find cuda drivers on your machine, GPU will not be used. 2023-06-29 18:21:29.007936: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations. To enable the following instructions: AVX2 AVX512F FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags. 2023-06-29 18:21:30.447795: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/tensorflow/python/compat/v2_compat.py:107: disable_resource_variables (from tensorflow.python.ops.variable_scope) is deprecated and will be removed in a future version. Instructions for updating: non-resource variables are not supported in the long term WARNING:root:To get the best performance, it is recommended to adjust the number of threads by setting the environment variables OMP_NUM_THREADS, TF_INTRA_OP_PARALLELISM_THREADS, and TF_INTER_OP_PARALLELISM_THREADS. See https://deepmd.rtfd.io/parallelism/ for more information. 2023-06-29 18:21:35.007303: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:353] MLIR V1 optimization pass is not enabled DEEPMD INFO The following nodes will be frozen: ['model_type', 'descrpt_attr/rcut', 'descrpt_attr/ntypes', 'model_attr/tmap', 'model_attr/model_type', 'model_attr/model_version', 'train_attr/min_nbor_dist', 'train_attr/training_script', 'o_energy', 'o_force', 'o_virial', 'o_atom_energy', 'o_atom_virial', 'spin_attr/ntypes_spin', 'fitting_attr/dfparam', 'fitting_attr/daparam'] WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/deepmd/entrypoints/freeze.py:367: convert_variables_to_constants (from tensorflow.python.framework.convert_to_constants) is deprecated and will be removed in a future version. Instructions for updating: This API was designed for TensorFlow v1. See https://www.tensorflow.org/guide/migrate for instructions on how to migrate your code to TensorFlow v2. WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/deepmd/entrypoints/freeze.py:367: convert_variables_to_constants (from tensorflow.python.framework.convert_to_constants) is deprecated and will be removed in a future version. Instructions for updating: This API was designed for TensorFlow v1. See https://www.tensorflow.org/guide/migrate for instructions on how to migrate your code to TensorFlow v2. WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/tensorflow/python/framework/convert_to_constants.py:952: extract_sub_graph (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version. Instructions for updating: This API was designed for TensorFlow v1. See https://www.tensorflow.org/guide/migrate for instructions on how to migrate your code to TensorFlow v2. WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/tensorflow/python/framework/convert_to_constants.py:952: extract_sub_graph (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version. Instructions for updating: This API was designed for TensorFlow v1. See https://www.tensorflow.org/guide/migrate for instructions on how to migrate your code to TensorFlow v2. DEEPMD INFO 2335 ops in the final graph.
代码
文本
它将在当前目录中输出一个名为 graph.pb 的模型文件。
代码
文本
到目前为止，我们就获得了一个使用 DeepSPIN 通过高精度的NiO数据获得的深度势能模型：DeepSPINt_Tutorial/01.train/graph.pb
代码
文本
[4]
! cd DeepSPIN_Tutorial/01.train/ && dp test -m graph.pb -s ../00.data/validation_data -d detail
2023-06-29 18:21:40.173359: I tensorflow/tsl/cuda/cudart_stub.cc:28] Could not find cuda drivers on your machine, GPU will not be used. 2023-06-29 18:21:40.228926: I tensorflow/tsl/cuda/cudart_stub.cc:28] Could not find cuda drivers on your machine, GPU will not be used. 2023-06-29 18:21:40.229567: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations. To enable the following instructions: AVX2 AVX512F FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags. 2023-06-29 18:21:41.112876: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/tensorflow/python/compat/v2_compat.py:107: disable_resource_variables (from tensorflow.python.ops.variable_scope) is deprecated and will be removed in a future version. Instructions for updating: non-resource variables are not supported in the long term WARNING:root:To get the best performance, it is recommended to adjust the number of threads by setting the environment variables OMP_NUM_THREADS, TF_INTRA_OP_PARALLELISM_THREADS, and TF_INTER_OP_PARALLELISM_THREADS. See https://deepmd.rtfd.io/parallelism/ for more information. 2023-06-29 18:21:42.968629: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:353] MLIR V1 optimization pass is not enabled WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/deepmd/utils/batch_size.py:61: is_gpu_available (from tensorflow.python.framework.test_util) is deprecated and will be removed in a future version. Instructions for updating: Use `tf.config.list_physical_devices('GPU')` instead. WARNING:tensorflow:From /opt/mamba/lib/python3.10/site-packages/deepmd/utils/batch_size.py:61: is_gpu_available (from tensorflow.python.framework.test_util) is deprecated and will be removed in a future version. Instructions for updating: Use `tf.config.list_physical_devices('GPU')` instead. DEEPMD WARNING You can use the environment variable DP_INFER_BATCH_SIZE tocontrol the inference batch size (nframes * natoms). The default value is 1024. DEEPMD INFO # ---------------output of dp test--------------- DEEPMD INFO # testing system : ../00.data/validation_data # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) # warning: loc idx out of upper bound (ignored if warned for more than 10 times) # warning: loc idx out of lower bound (ignored if warned for more than 10 times) DEEPMD INFO # number of test data : 40 DEEPMD INFO Energy MAE : 4.218759e-04 eV DEEPMD INFO Energy RMSE : 5.139633e-04 eV DEEPMD INFO Energy MAE/Natoms : 8.789082e-06 eV DEEPMD INFO Energy RMSE/Natoms : 1.070757e-05 eV DEEPMD INFO Force atom MAE : 4.454114e-04 eV/A DEEPMD INFO Force spin MAE : 5.101605e-04 eV/uB DEEPMD INFO Force atom RMSE : 7.330575e-04 eV/A DEEPMD INFO Force spin RMSE : 8.777647e-04 eV/uB DEEPMD INFO Virial MAE : 6.327634e+00 eV DEEPMD INFO Virial RMSE : 1.061004e+01 eV DEEPMD INFO Virial MAE/Natoms : 1.318257e-01 eV DEEPMD INFO Virial RMSE/Natoms : 2.210425e-01 eV DEEPMD INFO # -----------------------------------------------
代码
文本
DeepSPIN模型的能量、原子力、磁性力的预测结果保存在detail.e.out、detail.fr.out和detail.fm.out中。
下面让我们看下分别看下能量的预测结果，并进行可视化。
代码
文本
[5]
! head -n 5 DeepSPIN_Tutorial/01.train/detail.e.out
# ../00.data/validation_data: data_e pred_e -1.663128020599999957e+02 -1.663124941474969773e+02 -1.663122044899999992e+02 -1.663111771631338911e+02 -1.663113797100000113e+02 -1.663108779744183607e+02 -1.663116977100000042e+02 -1.663117939932237164e+02
代码
文本
其中，第一列为构型DFT能量，第二列为模型预测的能量。
代码
文本
[6]
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
e_out = pd.DataFrame(np.loadtxt("./DeepSPIN_Tutorial/01.train/detail.e.out"), columns=['data_e', 'pred_e'])
plt.scatter(e_out["data_e"]/32, e_out["pred_e"]/32, s=10)
x_range = np.linspace(plt.xlim()[0], plt.xlim()[1])
plt.plot(x_range, x_range, "r--", linewidth = 0.25)
plt.xlabel("Atomic energy of DFT (eV/atom)") # 设置 x 轴标题
plt.ylabel("Atomic energy predicted by DeepSPIN (eV/atom)") # 设置 y 轴标题
plt.show()
代码
文本
接下来我们看一下NiO模型的原子力预测结果。
代码
文本
[7]
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
fr_out = pd.DataFrame(np.loadtxt("./DeepSPIN_Tutorial/01.train/detail.fr.out"), columns=['data_frx', 'data_fry', 'data_frz', 'pred_frx', 'pred_fry', 'pred_frz'])
plt.scatter(fr_out["data_frx"], fr_out["pred_frx"], label='$F_{r_x}$', s=10)
plt.scatter(fr_out["data_fry"], fr_out["pred_fry"], label='$F_{r_y}$', s=10)
plt.scatter(fr_out["data_frz"], fr_out["pred_frz"], label='$F_{r_z}$', s=10)
x_range = np.linspace(plt.xlim()[0], plt.xlim()[1])
plt.plot(x_range, x_range, "r--", linewidth = 0.25)
plt.xlabel("Atomic force of DFT (eV/A)") # 设置 x 轴标题
plt.ylabel("Atomic force predicted by deep potential (eV/A)") # 设置 y 轴标题
plt.legend()
plt.show()
代码
文本
NiO模型磁性力的预测结果如下。
代码
文本
[8]
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
fm_out = pd.DataFrame(np.loadtxt("./DeepSPIN_Tutorial/01.train/detail.fm.out"), columns=['data_fmx', 'data_fmy', 'data_fmz', 'pred_fmx', 'pred_fmy', 'pred_fmz'])
plt.scatter(fm_out["data_fmx"], fm_out["pred_fmx"], label='$F_{m_x}$', s=10)
plt.scatter(fm_out["data_fmy"], fm_out["pred_fmy"], label='$F_{m_y}$', s=10)
plt.scatter(fm_out["data_fmz"], fm_out["pred_fmz"], label='$F_{m_z}$', s=10)
x_range = np.linspace(plt.xlim()[0], plt.xlim()[1])
plt.plot(x_range, x_range, "r--", linewidth = 0.25)
plt.xlabel("Magnetic force of DFT (eV/uB)") # 设置 x 轴标题
plt.ylabel("Magnetic force predicted by deep potential (eV/uB)") # 设置 y 轴标题
plt.legend()
plt.show()
代码
文本
代码
文本
进行能量最小化计算的主要输入文件有: in.lammps data.lmp model.pb。其中，model.pb 就是训练好的DeepSPIN模型。
代码
文本
[40]
! cp ./DeepSPIN_Tutorial/01.train/graph.pb ./DeepSPIN_Tutorial/02.lmp/model.pb && tree ./DeepSPIN_Tutorial/02.lmp -L 1
./DeepSPIN_Tutorial/02.lmp
|-- 0.dump
|-- 67.dump
|-- data.lmp
|-- in.lammps
|-- log.lammps
`-- model.pb
0 directories, 6 files
代码
文本
[33]
! cd ./DeepSPIN_Tutorial/02.lmp && head -n 20 data.lmp
2880 atoms
2 atom types
0.0000000000 80.0046997070 xlo xhi
0.0000000000 8.4331998825 ylo yhi
0.0000000000 38.7482986450 zlo zhi
0.0000000000 0.0000000000 0.0000000000 xy xz yz
Masses
1 58.690000
2 16.000000
Atoms # spin
1 1 11.3430700000 2.0963200000 0.0073600000 -0.51638874 -0.81647984 -0.25826989 1.2737
2 1 24.6774500000 2.0963200000 0.0073600000 -0.51638874 -0.81647984 -0.25826989 1.2737
3 1 4.6762700000 2.0963200000 0.0073600000 -0.51638874 -0.81647984 -0.25826989 1.2737
4 1 18.0106600000 2.0963200000 0.0073600000 -0.51638874 -0.81647984 -0.25826989 1.2737
5 1 38.0118300000 2.0963200000 0.0073600000 -0.51638874 -0.81647984 -0.25826989 1.2737
代码
文本
这里的data.lmp给出了NiO晶界模拟的初始构型。其中大部分内容和一般的LAMMSPS结构文件相同，但对于体系中的每个原子，除了需要给出原子位置坐标外，还需要给出原子磁矩在笛卡尔坐标系下3个方向的分量和磁矩模长，即Atoms # spin后6~9列的内容。
代码
文本
[34]
! cd ./DeepSPIN_Tutorial/02.lmp && cat in.lammps
#!/bin/sh units metal dimension 3 boundary p p p atom_style spin atom_modify map array box tilt large read_data data.lmp pair_style deepmd model.pb virtual_len 0.4 spin_norm 1.2737 pair_coeff * * neighbor 0.1 bin neigh_modify every 10 check yes delay 20 delete_atoms overlap 0.05 all all compute s all property/atom sp spx spy spz fmx fmy fmz fx fy fz dump 1 all custom 100 *.dump id x y z c_s[1] c_s[2] c_s[3] c_s[4] c_s[5] c_s[6] c_s[7] c_s[8] c_s[9] c_s[10] thermo_style custom vol pe etotal thermo 100 thermo_modify line one format float %10.8f min_style spin/lbfgs min_modify line spin_cubic discrete_factor 10.0 minimize 1.0e-12 1.0e-12 10000 20000
代码
文本
in.lammps是LAMMPS的输入文件，用来优化NiO晶界的磁矩方向以实现能量最小化，其中大部分命令和通常的 LAMMPS 分子动力学模拟输入文件类似（关于 LAMMPS 分子动力学模拟输入文件的更多信息，可参阅LAMMPS手册)
其中，调用势函数的命令与标准的DeepMD类似，不过，在DeepSPIN势函数中，需要额外指定 virtual_len 和 spin_norm 两个参数，设置与DeepSPIN模型训练脚本中的spin参数保持一致。
pair_style deepmd model.pb virtual_len 0.4 spin_norm 1.2737
pair_coeff * *
与能量最小化相关的命令为最后三行，其中，min_style 和 min_modify 共同指定了自旋优化的算法和参数，可参阅LAMMPS SPIN手册；minimize 指定了能量和原子力的收敛判据。
min_style spin/lbfgs
min_modify line spin_cubic discrete_factor 10.0
minimize 1.0e-12 1.0e-12 10000 20000
代码
文本
在准备好输入文件后，执行以下命令即可提交 LAMMPS 任务进行NiO晶界能量最小化计算。
代码
文本
[1]
# ########## Time Warning: 3 min ##########
! cd /data/DeepSPIN_Tutorial/02.lmp && ~/soft/lammps_23Jun2022/src/lmp_mpi -i in.lammps
2023-06-29 18:52:31.346957: I tensorflow/tsl/cuda/cudart_stub.cc:28] Could not find cuda drivers on your machine, GPU will not be used. LAMMPS (23 Jun 2022 - Update 1) Reading data file ... triclinic box = (0 0 0) to (80.0047 8.4331999 38.748299) with tilt (0 0 0) 1 by 1 by 1 MPI processor grid reading atoms ... 2880 atoms read_data CPU = 0.111 seconds DeePMD-kit WARNING: Environmental variable TF_INTRA_OP_PARALLELISM_THREADS is not set. Tune TF_INTRA_OP_PARALLELISM_THREADS for the best performance. See https://deepmd.rtfd.io/parallelism/ for more information. DeePMD-kit WARNING: Environmental variable TF_INTER_OP_PARALLELISM_THREADS is not set. Tune TF_INTER_OP_PARALLELISM_THREADS for the best performance. See https://deepmd.rtfd.io/parallelism/ for more information. DeePMD-kit WARNING: Environmental variable OMP_NUM_THREADS is not set. Tune OMP_NUM_THREADS for the best performance. See https://deepmd.rtfd.io/parallelism/ for more information. Summary of lammps deepmd module ... >>> Info of deepmd-kit: installed to: /root/soft/deepspin/deepmd_root source: v2.2.2 source branch: master source commit: 92ca097b source commit at: 2023-05-24 13:45:03 +0800 surpport model ver.:1.1 build variant: cpu build with tf inc: /opt/mamba/lib/python3.10/site-packages/tensorflow/include;/opt/mamba/lib/python3.10/site-packages/tensorflow/include build with tf lib: /opt/mamba/lib/python3.10/site-packages/tensorflow/libtensorflow_cc.so.2 set tf intra_op_parallelism_threads: 0 set tf inter_op_parallelism_threads: 0 >>> Info of lammps module: use deepmd-kit at: /root/soft/deepspin/deepmd_root source: v2.2.2 source branch: master source commit: 92ca097b source commit at: 2023-05-24 13:45:03 +0800 build float prec: double build with tf inc: /opt/mamba/lib/python3.10/site-packages/tensorflow/include;/opt/mamba/lib/python3.10/site-packages/tensorflow/include build with tf lib: /opt/mamba/lib/python3.10/site-packages/tensorflow/libtensorflow_cc.so.2 DeePMD-kit WARNING: Environmental variable TF_INTRA_OP_PARALLELISM_THREADS is not set. Tune TF_INTRA_OP_PARALLELISM_THREADS for the best performance. See https://deepmd.rtfd.io/parallelism/ for more information. DeePMD-kit WARNING: Environmental variable TF_INTER_OP_PARALLELISM_THREADS is not set. Tune TF_INTER_OP_PARALLELISM_THREADS for the best performance. See https://deepmd.rtfd.io/parallelism/ for more information. DeePMD-kit WARNING: Environmental variable OMP_NUM_THREADS is not set. Tune OMP_NUM_THREADS for the best performance. See https://deepmd.rtfd.io/parallelism/ for more information. 2023-06-29 18:52:32.427079: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations. To enable the following instructions: AVX2 AVX512F FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags. 2023-06-29 18:52:32.495243: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:353] MLIR V1 optimization pass is not enabled >>> Info of model(s): using 1 model(s): model.pb rcut in model: 5.6 ntypes in model: 3 using compute id: System init for delete_atoms ... Generated 0 of 1 mixed pair_coeff terms from geometric mixing rule Neighbor list info ... update every 10 steps, delay 20 steps, check yes max neighbors/atom: 2000, page size: 100000 master list distance cutoff = 5.7 ghost atom cutoff = 5.7 binsize = 2.85, bins = 29 3 14 2 neighbor lists, perpetual/occasional/extra = 1 1 0 (1) command delete_atoms, occasional, copy from (2) attributes: full, newton on pair build: copy stencil: none bin: none (2) pair deepmd, perpetual attributes: full, newton on pair build: full/bin/atomonly stencil: full/bin/3d bin: standard Deleted 48 atoms, new total = 2832 CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE Your simulation uses code contributions which should be cited: - USER-DEEPMD package: - min_style spin/lbfgs command: The log file lists these citations in BibTeX format. CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE-CITE WARNING: Using 'neigh_modify every 1 delay 0 check yes' setting during minimization (../min.cpp:187) WARNING: Line search incompatible gneb (../min_spin_lbfgs.cpp:108) Generated 0 of 1 mixed pair_coeff terms from geometric mixing rule Neighbor list info ... update every 1 steps, delay 0 steps, check yes max neighbors/atom: 2000, page size: 100000 master list distance cutoff = 5.7 ghost atom cutoff = 5.7 binsize = 2.85, bins = 29 3 14 1 neighbor lists, perpetual/occasional/extra = 1 0 0 (1) pair deepmd, perpetual attributes: full, newton on pair build: full/bin/atomonly stencil: full/bin/3d bin: standard Setting up spin/lbfgs style minimization ... Unit style : metal Current step : 0 Per MPI rank memory allocation (min/avg/max) = 8.86 | 8.86 | 8.86 Mbytes Volume PotEng TotEng 26143.30753976 -14656.12464332 -14656.12464332 26143.30753976 -14673.15632414 -14673.15632414 Loop time of 132.703 on 1 procs for 67 steps with 2832 atoms 1234.9% CPU use with 1 MPI tasks x no OpenMP threads Minimization stats: Stopping criterion = energy tolerance Energy initial, next-to-last, final = -14656.1246433169 -14673.1563241283 -14673.1563241416 Force two-norm initial, final = 59.508486 58.803457 Force max component initial, final = 2.9980712 3.0199425 Final line search alpha, max atom move = 0 0 Iterations, force evaluations = 67 68 MPI task timing breakdown: Section | min time | avg time | max time |%varavg| %total --------------------------------------------------------------- Pair | 132.62 | 132.62 | 132.62 | 0.0 | 99.93 Neigh | 0 | 0 | 0 | 0.0 | 0.00 Comm | 0.021493 | 0.021493 | 0.021493 | 0.0 | 0.02 Output | 0 | 0 | 0 | 0.0 | 0.00 Modify | 0 | 0 | 0 | 0.0 | 0.00 Other | | 0.06625 | | | 0.05 Nlocal: 2832 ave 2832 max 2832 min Histogram: 1 0 0 0 0 0 0 0 0 0 Nghost: 6546 ave 6546 max 6546 min Histogram: 1 0 0 0 0 0 0 0 0 0 Neighs: 0 ave 0 max 0 min Histogram: 1 0 0 0 0 0 0 0 0 0 FullNghs: 230880 ave 230880 max 230880 min Histogram: 1 0 0 0 0 0 0 0 0 0 Total # of neighbors = 230880 Ave neighs/atom = 81.525424 Neighbor list builds = 0 Dangerous builds = 0 Total wall time: 0:02:15
代码
文本
当打印出下述信息，表示能量最小化已经完成，可以看到计算完成后，已经达到了能量收敛判据。
Minimization stats:
Stopping criterion = energy tolerance
Energy initial, next-to-last, final =
-14656.1246433169 -14673.1563241283 -14673.1563241416
代码
文本
主要的输出文件为log.lammps 0.dump 67.dump，其中，log.lammps文件给出了LAMMPS运行时的输出信息；0.dump 67.dump文件就是上面提到的用dump命令输出的文件，里面包含每个时间步体系中每个原子的位置，磁矩，磁性力和原子力，数字代表minimize的迭代步数
代码
文本
[3]
! head -n 30 DeepSPIN_Tutorial/02.lmp/0.dump
ITEM: TIMESTEP 0 ITEM: NUMBER OF ATOMS 2832 ITEM: BOX BOUNDS xy xz yz pp pp pp 0.0000000000000000e+00 8.0004699707000000e+01 0.0000000000000000e+00 0.0000000000000000e+00 8.4331998825000003e+00 0.0000000000000000e+00 0.0000000000000000e+00 3.8748298644999998e+01 0.0000000000000000e+00 ITEM: ATOMS id x y z c_s[1] c_s[2] c_s[3] c_s[4] c_s[5] c_s[6] c_s[7] c_s[8] c_s[9] c_s[10] 1 74.0139 8.42122 24.0879 0 0 0 1 0 0 0 -0.040416 0.00526837 -0.870171 2 63.3461 8.42122 29.4216 0 0 0 1 0 0 0 0.00186022 0.000116457 0.000917555 3 79.3471 8.42122 34.7553 0 0 0 1 0 0 0 -0.300093 0.00154569 -1.37227 4 59.3459 8.42122 21.4212 0 0 0 1 0 0 0 1.83373 0.0172092 -0.530049 5 75.3468 8.42122 26.7549 0 0 0 1 0 0 0 0.0205602 -0.00230077 0.112241 6 64.6798 8.42122 32.0886 0 0 0 1 0 0 0 0.0348865 0.00271567 0.13419 7 54.0128 8.42122 37.4223 0 0 0 1 0 0 0 0.951196 0.00522673 -2.99553 8 60.6796 8.42122 24.0879 0 0 0 1 0 0 0 -0.0393569 0.00531156 -0.870092 9 76.6805 8.42122 29.4216 0 0 0 1 0 0 0 -2.63076e-06 8.50603e-05 0.0013302 10 66.0135 8.42122 34.7553 0 0 0 1 0 0 0 -0.305564 0.00149207 -1.37128 11 72.6803 8.42122 21.4212 0 0 0 1 0 0 0 1.83279 0.0171116 -0.531242 12 62.0132 8.42122 26.7549 0 0 0 1 0 0 0 0.0168948 -0.00235642 0.112991 13 11.3431 2.09632 0.00736 1.2737 -0.516389 -0.81648 -0.25827 25.5509 40.6885 22.9967 -0.726702 0.0831648 0.00153699 14 24.6775 2.09632 0.00736 1.2737 -0.516389 -0.81648 -0.25827 25.3538 40.6862 22.9961 -0.728981 0.083172 0.00149955 15 4.67627 2.09632 0.00736 1.2737 -0.516389 -0.81648 -0.25827 25.3553 40.6866 22.9965 -0.72894 0.0831724 0.00150025 16 18.0107 2.09632 0.00736 1.2737 -0.516389 -0.81648 -0.25827 25.3444 40.6687 22.9869 -0.72812 0.0831795 0.00147424 17 38.0118 2.09632 0.00736 1.2737 -0.516389 -0.81648 -0.25827 25.3463 40.6689 22.9873 -0.728065 0.0831792 0.00147494 18 51.3454 2.09632 0.00736 1.2737 -0.516389 -0.81648 -0.25827 25.5509 40.6885 22.9967 -0.72669 0.0831651 0.00153707 19 31.3442 2.09632 0.00736 1.2737 -0.516389 -0.81648 -0.25827 25.5503 40.6895 22.9971 -0.726666 0.0831641 0.00153771 20 44.6786 2.09632 0.00736 1.2737 -0.516389 -0.81648 -0.25827 25.355 40.6865 22.9965 -0.728951 0.0831722 0.00150005 21 64.6798 2.09632 0.00736 1.2737 -0.516389 -0.81648 -0.25827 25.3544 40.6864 22.996 -0.728987 0.0831718 0.00149921
代码
文本
接下来，我们介绍下与自旋——晶格动力学计算相关的命令。
fix nve/spin设置了时间步积分方式，moving 表示自旋演化与晶格演化是相互耦合的，即进行自旋——晶格动力学计算。若改成 frozen，则表示固定晶格，只进行自旋动力学计算。
timestep表示时间积分的步长，单位为ps，这里对应的是0.1 fs。
最后，通过 run 命令设置动力学演化的步数。
fix 1 all nve/spin lattice moving
timestep 0.0001
run 1000
此外，对 NVT 系综的模拟可以通过 fix 命令加入 Langevin 热浴实现，可参阅fix_langevin_spin。
代码
文本
[ ]
代码
文本
已赞16
本文被以下合集收录
机器学习与DFT精华帖
gtang
更新于 2024-09-10
38 篇21 人关注
测试合集文章列表100篇
xingyanshi@dp.tech很长的名字xingyanshi@dp.tech很长的名字很长的
更新于 2024-08-04
104 篇2 人关注
推荐阅读
公开
Pytorch基础教程（一）：Tensor
爱学习的王一博
发布于 2024-03-11
4 赞5 转存文件
公开
快速开始 Scikit-Learn｜使用 Scikit-Learn 建立机器学习模型
MileAway
发布于 2023-06-09
7 赞20 转存文件
评论
Hui_Zhou