306 lines
12 KiB
Python
306 lines
12 KiB
Python
"""
|
||
MI_headless.py
|
||
无界面版 MI 运动想象范式通讯流程模拟脚本。
|
||
复现 MI_main.py 的完整指令序列（train 0/1, rest, predict, saveData），
|
||
但不依赖 psychopy 也不打开任何窗口/音频，用 time.sleep 替代帧循环等待。
|
||
|
||
启动顺序:
|
||
1. runDecoder.py
|
||
2. datamock.py
|
||
3. MI_headless.py
|
||
"""
|
||
|
||
import sys
|
||
import os
|
||
import json
|
||
import time
|
||
import threading
|
||
import zmq
|
||
import numpy as np
|
||
import ast
|
||
from datetime import datetime
|
||
|
||
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
|
||
from PubLibrary.InifileHelper import IniRead
|
||
|
||
personname = 'demo'
|
||
session = '01'
|
||
|
||
DATAMOCK_LABEL_ADDR = 'tcp://127.0.0.1:8101' # datamock 标签命令地址
|
||
|
||
|
||
# ========== ZMQ 结果接收服务 ==========
|
||
class ZmqResultServer(threading.Thread):
|
||
def __init__(self, port=8088):
|
||
threading.Thread.__init__(self)
|
||
self.port = port
|
||
self.running = True
|
||
self.energy = 0
|
||
self.paradigm = 0 # 0=个体校准, 1=康复训练, 2=等待模型训练
|
||
self.ChoosenNum = -1
|
||
self.context = zmq.Context()
|
||
self.socket = self.context.socket(zmq.ROUTER)
|
||
self.socket.bind(f"tcp://0.0.0.0:{self.port}")
|
||
self.daemon = True
|
||
self.trial_idx = 0
|
||
|
||
def run(self):
|
||
print(f"[Server] UpperHost_Server listening on {self.port}")
|
||
while self.running:
|
||
try:
|
||
frames = self.socket.recv_multipart(zmq.NOBLOCK)
|
||
if len(frames) < 3:
|
||
continue
|
||
message = json.loads(frames[2].decode('utf-8'))
|
||
method = message.get('method')
|
||
params = message.get('params')
|
||
if method == 'energy':
|
||
self.energy = params
|
||
elif method == 'paradigm':
|
||
self.paradigm = params
|
||
print(f"[Server] paradigm -> {params}")
|
||
elif method == 'result':
|
||
self.ChoosenNum = params
|
||
self.trial_idx += 1
|
||
print(f"[Server] result={self.ChoosenNum} (trial {self.trial_idx})")
|
||
except zmq.Again:
|
||
time.sleep(0.005)
|
||
except Exception as e:
|
||
print(f"[Server] error: {e}")
|
||
|
||
def stop(self):
|
||
self.running = False
|
||
self.socket.close()
|
||
self.context.term()
|
||
|
||
|
||
# ========== ZMQ 命令发送客户端 ==========
|
||
class ZmqCmdClient:
|
||
def __init__(self, host, port):
|
||
self.host = host
|
||
self.port = port
|
||
self.context = zmq.Context()
|
||
self.socket = self.context.socket(zmq.DEALER)
|
||
# PUSH socket 用于向 datamock.py 发送标签命令
|
||
self._label_sock = self.context.socket(zmq.PUSH)
|
||
self._label_sock.connect(DATAMOCK_LABEL_ADDR)
|
||
print(f"[Client] label PUSH connected to {DATAMOCK_LABEL_ADDR}")
|
||
|
||
def connect(self):
|
||
self.socket.connect(f"tcp://{self.host}:{self.port}")
|
||
print(f"[Client] connected to {self.host}:{self.port}")
|
||
|
||
def start_recv_thread(self, result_server):
|
||
"""启动后台线程，持续接收 decoder 通过 8099 ROUTER 回发的消息，并更新 result_server 的状态"""
|
||
self._result_server = result_server
|
||
self._stop_recv = threading.Event()
|
||
|
||
def _recv_loop():
|
||
while not self._stop_recv.is_set():
|
||
try:
|
||
frames = self.socket.recv_multipart(zmq.NOBLOCK)
|
||
# DEALER 收到的格式: [b'', json_bytes]
|
||
data_bytes = frames[-1]
|
||
message = json.loads(data_bytes.decode('utf-8'))
|
||
method = message.get('method')
|
||
params = message.get('params')
|
||
ts = datetime.now().strftime('%H:%M:%S.%f')[:-3]
|
||
print(f"[{ts}] [CmdClient] recv: {method}={params}")
|
||
if method == 'paradigm':
|
||
self._result_server.paradigm = params
|
||
print(f"[{ts}] [CmdClient] paradigm updated -> {params}")
|
||
elif method == 'result':
|
||
self._result_server.ChoosenNum = params
|
||
self._result_server.trial_idx += 1
|
||
print(f"[{ts}] [CmdClient] result={params} (trial {self._result_server.trial_idx})")
|
||
elif method == 'energy':
|
||
self._result_server.energy = params
|
||
except zmq.Again:
|
||
time.sleep(0.005)
|
||
except Exception as e:
|
||
print(f"[CmdClient recv] error: {e}")
|
||
time.sleep(0.01)
|
||
|
||
self._recv_thread = threading.Thread(target=_recv_loop, daemon=True)
|
||
self._recv_thread.start()
|
||
print(f"[Client] 后台接收线程已启动（监听 decoder 8099 回发消息）")
|
||
|
||
def stop_recv_thread(self):
|
||
if hasattr(self, '_stop_recv'):
|
||
self._stop_recv.set()
|
||
|
||
def _send_label(self, label_value):
|
||
"""向 datamock.py 发送标签命令"""
|
||
try:
|
||
self._label_sock.send_string(str(label_value), zmq.NOBLOCK)
|
||
except Exception as e:
|
||
print(f"[Client] label send error: {e}")
|
||
|
||
def send_data(self, method, params):
|
||
msg = {'method': method, 'params': params}
|
||
try:
|
||
self.socket.send_multipart([b'', json.dumps(msg).encode('utf-8')])
|
||
ts = datetime.now().strftime('%H:%M:%S.%f')[:-3]
|
||
print(f"[{ts}] send_data: {method}={params}")
|
||
# 根据 train/predict 命令向 datamock 发送标签
|
||
if method == 'train':
|
||
if params == 0:
|
||
self._send_label(1)
|
||
print(f"[Label] train 0 -> datamock label=1")
|
||
elif params == 1:
|
||
self._send_label(2)
|
||
print(f"[Label] train 1 -> datamock label=2")
|
||
elif method == 'predict':
|
||
self._send_label(99)
|
||
print(f"[Label] predict -> datamock label=99")
|
||
except Exception as e:
|
||
print(f"[Client] send error: {e}")
|
||
|
||
|
||
# ========== 主流程 ==========
|
||
def run_headless():
|
||
server = ZmqResultServer(port=8088)
|
||
server.start()
|
||
|
||
_dh = str(IniRead('system', 'Decoder_Host'))
|
||
_dp = int(IniRead('system', 'Decoder_Port'))
|
||
client = ZmqCmdClient(_dh, _dp)
|
||
client.connect()
|
||
client.start_recv_thread(server) # 启动后台接收线程，监听 decoder 8099 回发的 paradigm/result 消息
|
||
|
||
time.sleep(1) # 等待连接建立
|
||
client.send_data('decoderClass', 'mi')
|
||
|
||
# MI_IntervalEpoch = [0.5, 4.5]，trial时长 = 4.5-0.5 = 4.0s
|
||
_mi_iv = ast.literal_eval(IniRead('system', 'MI_IntervalEpoch')) # [0.5, 4.5]
|
||
_trial_sec = float(_mi_iv[1] - _mi_iv[0]) # 4.0s
|
||
_margin = 1.0
|
||
train_time = max(5.0, _trial_sec + _margin) # 训练刺激时长（与 MI_main.py 保持一致）
|
||
|
||
# MI epoch latency = interval_epoch[1] // 5 = (4.5*250)//5 = 225包 × 20ms = 4.5s
|
||
# train_latency = 225包（MI中 train_latency == latency）
|
||
# 在 train_time 后需再等 epoch_wait 秒，decoder 才能完成 epoch 采集
|
||
epoch_wait = _mi_iv[1] / _mi_iv[1] * (_mi_iv[1] * 250 // 5) * 0.02 # = latency * 20ms
|
||
# 更直接的计算：latency = interval_epoch[1] // 5 = int(4.5*250)//5 = 225，225*0.02 = 4.5s
|
||
epoch_wait = (int(_mi_iv[1] * 250) // 5) * 0.02 # 4.5s
|
||
|
||
# predict epoch wait（与 train 相同，MI中 latency == train_latency）
|
||
predict_epoch_wait = epoch_wait # 4.5s
|
||
|
||
test_time = 7.0 # 预测窗口时长（与 MI_main.py 保持一致）
|
||
right_rehabilitation = float(IniRead('system', 'Right_rehabilitation'))
|
||
fault_rehabilitation = float(IniRead('system', 'Fault_rehabilitation'))
|
||
rest_time = float(IniRead('system', 'Rest_time'))
|
||
|
||
num_blocks = int(IniRead('system', 'Num_blocks'))
|
||
num_trials = int(IniRead('system', 'Num_trials'))
|
||
|
||
trained = 0
|
||
Num_Total = 0
|
||
Num_Success = 0
|
||
user_choice = []
|
||
|
||
print("=" * 50)
|
||
print("[Headless] 开始运行 MI 通讯流程（无界面）")
|
||
print(f" MI_IntervalEpoch={_mi_iv}, trial_sec={_trial_sec:.2f}s")
|
||
print(f" train_time={train_time:.2f}s, epoch_wait={epoch_wait:.2f}s")
|
||
print(f" test_time={test_time:.2f}s, predict_epoch_wait={predict_epoch_wait:.2f}s")
|
||
print(f" num_blocks={num_blocks}, num_trials={num_trials}")
|
||
print("=" * 50)
|
||
|
||
try:
|
||
while True:
|
||
# -------- 个体校准阶段 --------
|
||
print("\n[Phase] 个体校准阶段 (paradigm=0)")
|
||
client.send_data('rest', 0)
|
||
time.sleep(1)
|
||
|
||
while server.paradigm == 0:
|
||
# 左侧 MI 刺激（train 0，label=1）
|
||
print(f"\n[Train] 左侧 MI 刺激 (train 0) trained={trained}")
|
||
client.send_data('rest', 0)
|
||
time.sleep(0.5) # ding 提示后等待
|
||
|
||
client.send_data('train', 0)
|
||
time.sleep(train_time + epoch_wait) # 等待刺激时间 + epoch 完成时间
|
||
|
||
trained += 1
|
||
client.send_data('rest', 0)
|
||
time.sleep(1.0) # 类间休息
|
||
|
||
# 空闲态样本采集（train 1，label=2）
|
||
print(f"\n[Train] 空闲态采集 (train 1) trained={trained}")
|
||
client.send_data('train', 1)
|
||
time.sleep(train_time + epoch_wait) # 等待刺激时间 + epoch 完成时间
|
||
|
||
trained += 1
|
||
client.send_data('rest', 0)
|
||
time.sleep(1.0) # 类间休息
|
||
|
||
# 个体校准阶段结束
|
||
print("\n[Phase] 个体校准结束，等待模型训练 (paradigm=2) ...")
|
||
trained = 0
|
||
time.sleep(1)
|
||
|
||
# 等待模型训练完成 (paradigm=2 -> paradigm=1)
|
||
while server.paradigm == 2:
|
||
print("[Phase] 等待模型训练完成 ...")
|
||
time.sleep(0.5)
|
||
|
||
# -------- 康复训练阶段 --------
|
||
while server.paradigm == 1:
|
||
print("\n[Phase] 康复训练阶段 (paradigm=1)")
|
||
for block_idx in range(num_blocks):
|
||
print(f"\n [Block {block_idx+1}/{num_blocks}]")
|
||
time.sleep(10) # 每轮开始前等待
|
||
|
||
for trial_idx in range(num_trials):
|
||
print(f" [Trial {trial_idx+1}/{num_trials}]")
|
||
|
||
time.sleep(0.5) # ding 提示
|
||
server.ChoosenNum = -1
|
||
|
||
# 开始预测
|
||
# MI predict epoch latency = 225包 × 20ms = 4.5s，需额外等待 epoch 完成
|
||
client.send_data('predict', 1)
|
||
t_start = time.perf_counter()
|
||
while time.perf_counter() - t_start < test_time + predict_epoch_wait:
|
||
if server.ChoosenNum >= 0:
|
||
Num_Total += 1
|
||
user_choice.append(server.ChoosenNum)
|
||
if server.ChoosenNum == 0:
|
||
Num_Success += 1
|
||
rest_time = right_rehabilitation
|
||
elif server.ChoosenNum == 1:
|
||
rest_time = fault_rehabilitation
|
||
break
|
||
time.sleep(0.02)
|
||
|
||
trained += 1
|
||
client.send_data('rest', 0)
|
||
time.sleep(0.5)
|
||
time.sleep(rest_time)
|
||
server.ChoosenNum = -1
|
||
|
||
# 训练结束
|
||
print("\n[Phase] 康复训练结束")
|
||
break # 退出康复训练循环
|
||
|
||
# 统计结果
|
||
overall_accuracy = Num_Success / Num_Total if Num_Total > 0 else 0
|
||
print(f"\n[Result] Overall={overall_accuracy:.3f} ({Num_Success}/{Num_Total})")
|
||
print(f"[Result] user_choice={user_choice}")
|
||
break # 完成一个完整流程后退出
|
||
|
||
except KeyboardInterrupt:
|
||
print("\n[Headless] 用户中断")
|
||
finally:
|
||
client.send_data('predict', 2) # 关闭系统
|
||
client.send_data('saveData', 0)
|
||
server.stop()
|
||
print("[Headless] 已发送关闭指令，退出。")
|
||
|
||
|
||
if __name__ == '__main__':
|
||
run_headless()
|