koko/NEP-auto
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

补充了通知模块

Browse Source
This commit is contained in:
koko
2025-12-14 13:59:00 +08:00
parent f95f4011d4
commit b6a1af5264
9 changed files with 242 additions and 104 deletions
Download Patch File Download Diff File Expand all files Collapse all files

51
config/param.yaml
View File

@@ -1,13 +1,13 @@
# param.yaml
project: "LGPS"
project: "LYC_model1"
# 1. 初始文件定义 (对应 data/ 目录)
files:
poscar: "POSCAR"
poscar: "model1.vasp"
potcar: "POTCAR"
initial_pot: "nep89.txt" # 第一轮 MD 用的势函数
label: "Li Ge P S"
label: "Li Y Cl"
# 2. 迭代流程控制
iterations:
@@ -37,9 +37,33 @@ iterations:
# 逻辑：cp template/03.train/nep.in
- name: "03.train"
executor: "nep_local"
# --- 第二轮 ---
- id: 1
steps:
# Step 1: MD (预热 + 采样)
# 逻辑：会把 nep.txt (来自 initial_pot) 和 model.xyz 准备好
- name: "00.md"
sub_tasks:
# 你提到可能有预热，也可能有加工，这里支持串行执行
- template_sub: "preheat" # 使用 template/00.md/preheat/run.in
- template_sub: "production" # 使用 template/00.md/production/run.in
executor: "gpumd" # 对应 machine.yaml
# Step 2: 筛选
- name: "01.select"
method: "distance"
params: [90, 120]
# Step 3: SCF (VASP)
# 逻辑：cp template/02.scf/INCAR; check KPOINTS; cp data/POTCAR
- name: "02.scf"
executor: "vasp_std" # 对应 machine.yaml (可能调用 vasp_std.sh)
# Step 4: 训练
# 逻辑：cp template/03.train/nep.in
- name: "03.train"
executor: "nep_local"
# --- 第二轮 ---
- id: 2
steps:
- name: "00.md"
sub_tasks:
@@ -57,11 +81,20 @@ iterations:
- name: "03.train"
executor: "nep_local"
- name: "04.predict"
# 定义温度和时间列表
# [新增] 自定义模型文件 (位于 data/ 目录下)，不填则使用当前训练结果
# custom_nep: "nep_final_best.txt"
# [新增] 自定义预测结构 (位于 data/ 目录下)，不填则使用 00.md 的结果
# 注意：这里填写 .vasp 文件，程序会自动转化为 model.xyz
custom_poscar: "model1_supercell.vasp"
conditions:
- {T: 375, time: "15ns"}
- { T: 400, time: "5ns" }
- { T: 425, time: "2ns" }
- { T: 450, time: "1ns" }
- { T: 500, time: "1ns" }
- {T: 600, time: "1ns"} # 支持不同温度不同时长
- { T: 600, time: "1ns" }
- { T: 700, time: "1ns" }
- { T: 800, time: "1ns" }
- {T: 900, time: "1ns"} # 支持不同温度不同时长
- {T: 1000, time: "1ns"}
- { T: 900, time: "1ns" }

39
src/utils.py
View File

@@ -36,16 +36,43 @@ def setup_logger(work_dir, log_file="autonep.log"):
return logger
# src/utils.py 中的 Notifier 类
class Notifier:
"""(预留) 通知模块"""
def __init__(self, project_name="AutoNEP"):
self.project_name = project_name
def __init__(self, url=None):
self.url = url
def send(self, title, message, priority=3):
"""
调用本地 'post' 命令发送通知
Args:
title: 标题
message: 内容
priority: 优先级 (0-3: Low/Log, 4-7: Med/Done, 8+: High/Err)
"""
import subprocess
def send(self, title, msg, priority=5):
# 暂时只打印日志，不实际发送
logging.info(f"[[Notification]] {title}: {msg}")
# 自动加上项目前缀
full_title = f"[{self.project_name}] {title}"
# 构造命令: post -a task -t "Title" -m "Msg" -p Priority
cmd = [
"post",
"-a", "task", # 固定推送到 task 频道
"-t", full_title,
"-m", message,
"-p", str(priority)
]
try:
# 执行命令，不输出到屏幕，防止干扰主程序日志
subprocess.run(cmd, check=False, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
# 同时在本地日志留档
logging.info(f"[[Notification Sent]] P{priority} | {title}: {message}")
except FileNotFoundError:
logging.warning("Command 'post' not found. Notification skipped.")
except Exception as e:
logging.warning(f"Failed to send notification: {e}")
# src/utils.py 添加在最后

162
src/workflow.py
View File

@@ -25,14 +25,16 @@ class Workflow:
# 初始化状态追踪
os.makedirs(self.workspace, exist_ok=True)
self.tracker = StateTracker(self.workspace)
project_name = self.param.get('project', 'AutoNEP')
self.notifier = Notifier(project_name)
self.notifier.send("Workflow Init", "AutoNEP framework initialized.", 1)
# 初始变量
self.current_nep_pot = os.path.join(self.data_dir, self.param['files']['initial_pot'])
self.current_train_set = os.path.join(self.workspace, "accumulated_train.xyz")
def run(self):
self.logger.info(f"Workflow Started: {self.param['project']}")
self.notifier.send("Workflow Start", f"Starting project: {self.param['project']}", 5)
for iteration in self.param['iterations']:
iter_id = iteration['id']
iter_name = f"iter_{iter_id:02d}"
@@ -73,6 +75,7 @@ class Workflow:
cmd = f"{kit_path} -addlabel {std_poscar_name} {atom_labels}"
if run_cmd_with_log(cmd, step_dir, "init.log"):
self.notifier.send(f"Step Done: {step_name}", f"project {self.param['project']} Finished task {task_id_init}", 2)
self.tracker.mark_done(task_id_init)
else:
self.logger.error("Custom POSCAR initialization failed.")
@@ -93,6 +96,7 @@ class Workflow:
cmd = f"{kit_path} -addlabel {poscar_name} {atom_labels}"
if run_cmd_with_log(cmd, step_dir, "init.log"):
self.notifier.send(f"Step Done: {step_name}", f"project {self.param['project']} Finished task {task_id_init}", 2)
self.tracker.mark_done(task_id_init)
else:
self.logger.error("Initialization failed.")
@@ -111,11 +115,13 @@ class Workflow:
if os.path.exists(prev_model_src):
self.logger.info(f"Copying model.xyz from {prev_iter_name}...")
shutil.copy(prev_model_src, os.path.join(step_dir, "model.xyz"))
self.notifier.send(f"Step Done: {step_name}", f"project {self.param['project']} Finished task {task_id_init}", 2)
self.tracker.mark_done(task_id_init) # 标记完成
else:
self.logger.error(f"Previous model.xyz not found: {prev_model_src}")
return
else:
self.notifier.send(f"Step Done: {step_name}", f"project {self.param['project']} Finished task {task_id_init}", 2)
self.tracker.mark_done(task_id_init)
# 确保 gpumdkit 路径可用
kit_path = self.machine.config['paths'].get('gpumdkit', 'gpumdkit.sh')
@@ -147,6 +153,7 @@ class Workflow:
if run_cmd_with_log(kit_path, preheat_dir, "step_exec.log", input_str=input_str_201):
if os.path.exists(os.path.join(preheat_dir, "sampled_structures.xyz")):
self.notifier.send(f"Step Done: {step_name}", f"project {self.param['project']} Finished task {task_id_preheat}", 2)
self.tracker.mark_done(task_id_preheat)
else:
self.logger.error("sampled_structures.xyz not generated.")
@@ -216,6 +223,7 @@ class Workflow:
run_cmd_with_log("cat sample_*/dump.xyz > dump.xyz", prod_dir, "step_exec.log")
self.last_dump_path = os.path.join(prod_dir, "dump.xyz")
self.notifier.send(f"Step Done: {step_name}", f"project {self.param['project']} Finished task {task_id_prod}", 2)
self.tracker.mark_done(task_id_prod)
else:
self.logger.info("Skipping Production (Already Done).")
@@ -395,7 +403,7 @@ class Workflow:
src_png = os.path.join(step_dir, "select.png")
if os.path.exists(src_png):
shutil.copy(src_png, os.path.join(output_dir, "select.png"))
self.notifier.send(f"Step Done: {step_name}", f"project {self.param['project']} Finished task {task_id_select}", 2)
self.tracker.mark_done(task_id_select)
# === 分支 B: 随机筛选 (保持不变) ===
@@ -409,6 +417,7 @@ class Workflow:
f.write(f"Random,{min_n},{max_n},Executed\n")
src_png = os.path.join(step_dir, "select.png")
if os.path.exists(src_png): shutil.copy(src_png, os.path.join(output_dir, "select.png"))
self.notifier.send(f"Step Done: {step_name}", f"project {self.param['project']} Finished task {task_id_select}", 2)
self.tracker.mark_done(task_id_select)
else:
self.logger.error("Random selection failed.")
@@ -562,6 +571,7 @@ class Workflow:
self.logger.info(f"VASP data collected: {res_file}")
# 保存这个路径供 Train 使用
self.new_data_chunk = res_file
self.notifier.send(f"Step Done: {step_name}", f"project {self.param['project']} Finished task {task_id_scf}", 2)
self.tracker.mark_done(task_id_scf)
else:
self.logger.error("NEP-dataset.xyz not found after collection.")
@@ -683,7 +693,7 @@ class Workflow:
dst_png = os.path.join(output_dir, file)
shutil.copy(src_png, dst_png)
self.logger.info(f"Archived plot: {file}")
self.notifier.send(f"Step Done: {step_name}", f"project {self.param['project']} Finished task {task_id_train}", 2)
self.tracker.mark_done(task_id_train)
else:
@@ -706,48 +716,108 @@ class Workflow:
self.logger.info("=== Step: 04.predict (Arrhenius) ===")
os.makedirs(step_dir, exist_ok=True)
# 准备输出目录
output_dir = os.path.join(iter_path, "05.output")
os.makedirs(output_dir, exist_ok=True)
# 1. 准备基础文件 (model.xyz, nep.txt)
# 注意：预测通常使用最新的 nep.txt 和 原始/最新 model.xyz
# 这里使用 current_nep_pot 和 初始 model.xyz (保证一致性)
model_src = os.path.join(iter_path, "00.md", "model.xyz") # 假设 00.md 里有
if not os.path.exists(model_src):
# 如果没有，尝试从 data 目录拿
model_src = os.path.join(self.data_dir, self.param['files']['poscar'])
# 这里简化处理，最好确保 00.md 产生了 model.xyz
# -------------------------------------------------
# 1. 准备 NEP 势函数 (支持自定义)
# -------------------------------------------------
custom_nep_name = step_conf.get('custom_nep')
nep_src = ""
# 2. 遍历温度点执行模拟
if custom_nep_name:
# 优先使用 data/ 下的自定义模型
nep_src = os.path.join(self.data_dir, custom_nep_name)
if not os.path.exists(nep_src):
self.logger.error(f"Custom NEP file not found: {nep_src}")
return
self.logger.info(f"Using Custom NEP model: {custom_nep_name}")
else:
# 默认使用当前流程的训练结果
nep_src = self.current_nep_pot
self.logger.info("Using current training result for prediction.")
if not os.path.exists(nep_src):
self.logger.error(f"NEP potential source missing: {nep_src}")
return
# -------------------------------------------------
# 2. 准备 model.xyz (支持自定义 VASP -> 转化)
# -------------------------------------------------
custom_poscar_name = step_conf.get('custom_poscar')
final_model_xyz = os.path.join(step_dir, "model.xyz") # 统一存放在 step_dir 根目录备用
kit_path = self.machine.config['paths'].get('gpumdkit', 'gpumdkit.sh')
if custom_poscar_name:
# === Case A: 使用自定义 VASP 结构 ===
self.logger.info(f"Using Custom POSCAR for prediction: {custom_poscar_name}")
poscar_src = os.path.join(self.data_dir, custom_poscar_name)
if os.path.exists(poscar_src):
# 复制 VASP 文件到当前目录
shutil.copy(poscar_src, os.path.join(step_dir, custom_poscar_name))
# 获取 label 并执行转化
atom_labels = self.param['files'].get('label', '')
if not atom_labels:
self.logger.error("Labels missing in config (files.label), cannot convert POSCAR.")
return
cmd = f"{kit_path} -addlabel {custom_poscar_name} {atom_labels}"
self.logger.info(f"Converting POSCAR to model.xyz: {cmd}")
if not run_cmd_with_log(cmd, step_dir, "convert_model.log"):
self.logger.error("Failed to convert custom POSCAR.")
return
if not os.path.exists(final_model_xyz):
self.logger.error("model.xyz not generated after conversion.")
return
else:
self.logger.error(f"Custom POSCAR not found: {poscar_src}")
return
else:
# === Case B: 使用流程默认结构 (00.md) ===
self.logger.info("Using default structure from 00.md")
default_src = os.path.join(iter_path, "00.md", "model.xyz")
if os.path.exists(default_src):
shutil.copy(default_src, final_model_xyz)
else:
# 尝试兜底：用 data 里的初始 POSCAR
self.logger.warning("00.md/model.xyz not found. Trying initial POSCAR...")
# 这里省略复杂的再次转化逻辑，建议保证 00.md 跑通
self.logger.error(f"Default model.xyz source missing: {default_src}")
return
# -------------------------------------------------
# 3. 遍历温度点执行模拟
# -------------------------------------------------
conditions = step_conf.get('conditions', [])
if not conditions:
self.logger.error("No conditions defined for 04.predict")
continue
# 发送开始通知
self.notifier.send("Predict Start", f"Starting {len(conditions)} tasks in {iter_name}", 5)
self.notifier.send("Predict Start", f"Tasks: {len(conditions)}", 5)
for cond in conditions:
temp = cond['T']
time_str = cond['time']
steps = parse_time_to_steps(time_str)
# [新增] 计算 MSD Window
# 公式: 10 * window * 20 = steps => window = steps / 200
# 自动计算 MSD Window (Steps / 200)
msd_window = int(steps / 200)
sub_dir_name = f"{temp}K"
sub_work_dir = os.path.join(step_dir, sub_dir_name)
os.makedirs(sub_work_dir, exist_ok=True)
self.logger.info(
f"-> Running Prediction: {temp}K, {time_str} ({steps} steps, MSD window={msd_window})")
self.logger.info(f"-> Running Prediction: {temp}K, {time_str}")
# 分发文件
if os.path.exists(model_src):
shutil.copy(model_src, os.path.join(sub_work_dir, "model.xyz"))
shutil.copy(self.current_nep_pot, os.path.join(sub_work_dir, "nep.txt"))
# 分发准备好的 model.xyz 和 nep.txt
shutil.copy(final_model_xyz, os.path.join(sub_work_dir, "model.xyz"))
shutil.copy(nep_src, os.path.join(sub_work_dir, "nep.txt"))
# 生成 run.in
template_path = os.path.join(self.template_dir, "04.predict", "run.in")
@@ -755,7 +825,6 @@ class Workflow:
with open(template_path, 'r') as f:
content = f.read()
# [修改] 替换参数，增加 MSD_WINDOW
content = content.replace("{T}", str(temp))
content = content.replace("{STEPS}", str(steps))
content = content.replace("{MSD_WINDOW}", str(msd_window))
@@ -766,34 +835,28 @@ class Workflow:
self.logger.error(f"Template not found: {template_path}")
continue
# 执行 GPUMD
# 这里不使用 nohup 模式，因为是串行跑，直接记录日志即可
if not run_cmd_with_log("gpumd", sub_work_dir, "predict.log"):
self.logger.error(f"Prediction failed at {temp}K")
# 也可以选择 continue 或者 return
# 3. 后处理分析 (gpumdkit.sh -plt sigma)
# -------------------------------------------------
# 4. 后处理分析
# -------------------------------------------------
self.logger.info("Running Analysis (gpumdkit -plt sigma)...")
kit_path = self.machine.config['paths'].get('gpumdkit', 'gpumdkit.sh')
# 捕获输出用于解析
analysis_log = os.path.join(step_dir, "sigma_analysis.log")
cmd_analyze = f"{kit_path} -plt sigma"
# 我们需要捕获 stdout 内容
process = subprocess.Popen(
cmd_analyze, shell=True, cwd=step_dir,
stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True
)
stdout, _ = process.communicate()
# 将输出写入日志文件备份
with open(analysis_log, 'w') as f:
f.write(stdout)
# 4. 解析输出并生成报告
# 目标格式: CSV
# 解析输出并生成报告 (Regex Logic)
csv_data = []
ea_val = "N/A"
sigma_300k = "N/A"
@@ -803,26 +866,16 @@ class Workflow:
for line in lines:
line = line.strip()
# 解析 Ea
if "Ea:" in line:
# 格式: 04.predict, Ea: 0.305 eV
match = re.search(r"Ea:\s+([\d\.]+)\s+eV", line)
if match: ea_val = match.group(1)
# 解析 300K Sigma
if "at 300K" in line:
# 格式: at 300K, 04.predict: Sigma = 1.816e-03 S/cm
match = re.search(r"Sigma\s*=\s*([\d\.eE\+\-]+)", line)
if match: sigma_300k = match.group(1)
# 解析表格
if "----------------" in line:
in_table = not in_table # 切换状态
in_table = not in_table
continue
if in_table:
# T (K) Sigma (S/cm) Sigma·T (K·S/cm)
# 425 3.656e-02 1.554e+01
parts = line.split()
if len(parts) >= 3 and parts[0].isdigit():
csv_data.append({
@@ -831,30 +884,21 @@ class Workflow:
"Sigma*T": parts[2]
})
# 写入 CSV 到 output
report_path = os.path.join(output_dir, "conductivity_report.csv")
with open(report_path, 'w') as f:
# 写头部汇总信息
f.write(f"# Extracted from gpumdkit output\n")
f.write(f"# Ea (eV):,{ea_val}\n")
f.write(f"# Extrapolated Sigma@300K (S/cm):,{sigma_300k}\n")
f.write("\n")
# 写表格数据
f.write(f"# Sigma@300K (S/cm):,{sigma_300k}\n\n")
f.write("Temperature(K),Sigma(S/cm),Sigma*T(K*S/cm)\n")
for row in csv_data:
f.write(f"{row['T(K)']},{row['Sigma(S/cm)']},{row['Sigma*T']}\n")
self.logger.info(f"Report saved to {report_path}")
self.notifier.send("Predict Done", f"Ea: {ea_val} eV, Sigma@300K: {sigma_300k}", 5)
self.notifier.send("Predict Done", f"Ea: {ea_val} eV", 5)
# 5. 归档 Arrhenius.png
src_png = os.path.join(step_dir, "Arrhenius.png")
if os.path.exists(src_png):
shutil.copy(src_png, os.path.join(output_dir, "Arrhenius.png"))
self.logger.info("Archived Arrhenius.png")
else:
self.logger.warning("Arrhenius.png not found.")
self.notifier.send(f"Step Done: {step_name}", f"project {self.param['project']} Finished task {task_id_predict}", 2)
self.tracker.mark_done(task_id_predict)
else:
self.logger.info("Skipping Predict (Already Done).")
self.notifier.send("Workflow end", f"project {self.param['project']} success", 5)

4
template/00.md/production/run.in
View File

@@ -1,7 +1,7 @@
potential ./nep.txt
velocity 400
velocity 350
ensemble npt_mttk temp 400 800 aniso 0 0
ensemble npt_mttk temp 350 900 aniso 0 0
dump_thermo 10
dump_exyz 100
run 500000

51
template/02.scf/INCAR
View File

@@ -1,24 +1,29 @@
SYSTEM = LiGePS static SCF
ISTART = 0
ICHARG = 2
ENCUT = 520
NPAR = 8
PREC = Normal
GGA = PE
EDIFF = 1E-6
ALGO = Normal
NELM = 120
NSW = 0
IBRION = -1
ISPIN = 1
ISMEAR = 0
SIGMA = 0.05
LASPH = .TRUE.
LREAL = .FALSE.
ADDGRID = .TRUE.
ISYM = 2
LCHARG = .FALSE.
LWAVE = .FALSE.
SYSTEM = Li3YCl6 Training Data Generation
KSPACING= 0.25
KGAMMA = .TRUE.
! --- 电子步基础设置 ---
ENCUT = 520 ! [anie202215544-sup-0001-misc_information (1).pdf] 明确指定
PREC = Accurate ! 保证力计算的精度
ALGO = Normal ! 或者 Normal
LREAL = .FALSE. ! 对于生成 Force training data，Projection 最好关掉(False)
! --- 电子收敛标准 ---
! 文献提到能量收敛 1e-4，但为了训练势函数，建议稍微严一点
EDIFF = 1E-6
ISYM = 0 ! 关闭对称性，防止 MD/采样 过程中因为对称性导致力计算的人为约束
! --- 展宽设置 (绝缘体/半导体) ---
ISMEAR = 0 ! Gaussian Smearing
SIGMA = 0.05 ! 配合 ISMEAR=0 使用，文献常用值
! --- 关键：optB88-vdW 泛函设置 ---
! [anie202215544-sup-0001-misc_information (1).pdf] 明确指出使用 optB88-vdW
! 以下参数必须完全照抄，不能改动
LUSE_VDW = .TRUE. ! 开启 vdW 修正
AGGAC = 0.0000 ! 必须为 0
GGA = BO ! optB88 基于 MK 交换泛函
PARAM1 = 0.18333333 ! optB88 特有参数
PARAM2 = 0.22000000 ! optB88 特有参数
! --- 输出控制 ---
LWAVE = .FALSE. ! 训练数据不需要波函数，节省空间
LCHARG = .FALSE. ! 不需要电荷密度

24
template/02.scf/INCAR_unifrom Normal file
View File

@@ -0,0 +1,24 @@
SYSTEM = static SCF
ISTART = 0
ICHARG = 2
ENCUT = 520
NPAR = 8
PREC = Normal
GGA = PE
EDIFF = 1E-6
ALGO = Normal
NELM = 120
NSW = 0
IBRION = -1
ISPIN = 1
ISMEAR = 0
SIGMA = 0.05
LASPH = .TRUE.
LREAL = .FALSE.
ADDGRID = .TRUE.
ISYM = 2
LCHARG = .FALSE.
LWAVE = .FALSE.
KSPACING= 0.25
KGAMMA = .TRUE.

5
template/02.scf/KPOINTS Normal file
View File

@@ -0,0 +1,5 @@
Automatic Mesh for Training Data
0 ! 0 表示自动生成
Gamma ! 必须用 Gamma centered（六角/三角晶系推荐）
3 3 3 ! 或者 3 3 3，取决于你的计算资源，2x2x2 通常够用
0 0 0 ! shift (通常为0)

BIN
template/02.scf/vdw_kernel.bindat Normal file
View File

Binary file not shown.

2
template/03.train/nep.in
View File

@@ -1,4 +1,4 @@
type 4 Li Ge P S
type 4 Li Y Cl
zbl 2
cutoff 6 5
generation 100000