def predict_mode(self): if self.train_mode is True: self.train_mode = False self.logger_test.push(UiLogger.Item(UiLogger.LEVEL_WARNING, 'switch', '切换到预测模式')) self.t2 = 0 else: self.train_mode = True self.logger_test.push(UiLogger.Item(UiLogger.LEVEL_WARNING, 'switch', '切换到训练模式')) self.t2 = 0
def __init__(self, root=None): self.init_top = Tk() self.port_left = 'COM4' self.port_right = 'COM5' self.init_bps = StringVar() self.init_bps.set('115200') self.init_com_left = StringVar() self.init_com_left.set(self.port_left) self.init_com_right = StringVar() self.init_com_right.set(self.port_right) self.init_communication() self.bps = 115200 self.comm = None self.n = 512 self.select = 24 self.frames = [[0 for i in range(12)] for j in range(self.n)] self.raw = [[[0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]] for j in range(self.n)] # 建立网络 self.model_file = 'mc_actions.h5' # 建立网络的过程放在线程2 # 初始化手柄连接 self.comm_left = BaseComm(self.init_com_left.get(), self.bps) self.comm_right = BaseComm(self.init_com_right.get(), self.bps) # 灵敏度 self.sensitivity = 0.3 # 先读取几次,读取Init标志 print('预读取:') for i in range(10): r1 = self.comm_left.read1epoch() r2 = self.comm_right.read1epoch() print(r1, r2) # 初始化某些变化的值 # 加上Uilogger print('请保持手柄水平放置不动') # 初始化遥感的中点值(取平均) X Y X Y self.ave_left, self.ave_right = [0, 0], [0, 0] pick = 10 for i in range(pick): data = self.comm_left.read1epoch() data_ctrl = data[-4:] data_ctrl = list(map(int, data_ctrl)) self.ave_left[0] += data_ctrl[2] / pick self.ave_left[1] += data_ctrl[1] / pick data = self.comm_right.read1epoch() data_ctrl = data[-4:] data_ctrl = list(map(int, data_ctrl)) self.ave_right[0] += data_ctrl[2] / pick self.ave_right[1] += data_ctrl[1] / pick print('初始化遥感中点:', self.ave_left, self.ave_right) self.root = root if self.root is None: self.root = Tk() self.root.title("MC手柄") self.logger = UiLogger(self.root, height=10, width=32) self.logger.logger().grid(row=2, column=1, sticky=W+E) self.panel = Label(self.root) self.panel.grid(row=1, column=1, sticky=W+E) self.lock = threading.Lock() t = threading.Thread(target=self.read_thread) t.setDaemon(True) t.start() t = threading.Thread(target=self.parse_thread) t.setDaemon(True) t.start()
def parse_thread(self): # 读取神经网络模型 model = load_model(self.model_file) t1, t2 = 0, 0 click = False key1 = ctrl.ACTION_NONE key2 = ctrl.ACTION_NONE jump = ctrl.ACTION_NONE start = time.time() while True: if t1 == 5: im = self.draw() imp = ImageTk.PhotoImage(image=im) self.panel.configure(image=imp) self.panel.image = imp t1 = 0 t1 += 1 time.sleep(0.01) # data_left = self.comm_left.read1epoch() # data_right = self.comm_right.read1epoch() self.lock.acquire() data_left = self.raw[-1][0] data_right = self.raw[-1][1] self.lock.release() # 右手处理 right_ctrl = data_right[-4:] right_ctrl = list(map(int, right_ctrl)) ctrl.move((right_ctrl[2] - self.ave_left[0]) * self.sensitivity, (right_ctrl[1] - self.ave_left[1]) * self.sensitivity) if right_ctrl[0] == 0 and click is False: ctrl.left_down() click = True if right_ctrl[0] == 1 and click is True: ctrl.left_up() click = False # Jump if data_left[6] == 0 and jump == ctrl.ACTION_NONE: jump = ctrl.ACTION_UP ctrl.kbd_down(VirtualKeyCode.SPACEBAR) # ctrl.kbd_click(VirtualKeyCode.SPACEBAR) if data_left[6] == 1 and jump == ctrl.ACTION_UP: jump = ctrl.ACTION_NONE ctrl.kbd_up(VirtualKeyCode.SPACEBAR) # ctrl.kbd_click(VirtualKeyCode.SPACEBAR) # 左手处理 pos = data_left[-4:][1:3] # Right if pos[1] > 800 and key1 == ctrl.ACTION_NONE: key1 = ctrl.ACTION_D ctrl.kbd_down(VirtualKeyCode.D_key) if pos[1] <= 800 and key1 == ctrl.ACTION_D: key1 = ctrl.ACTION_NONE ctrl.kbd_up(VirtualKeyCode.D_key) # Left if pos[1] < 200 and key1 == ctrl.ACTION_NONE: key1 = ctrl.ACTION_A ctrl.kbd_down(VirtualKeyCode.A_key) if pos[1] >= 200 and key1 == ctrl.ACTION_A: key1 = ctrl.ACTION_NONE ctrl.kbd_up(VirtualKeyCode.A_key) # Backward if pos[0] < 200 and key2 == ctrl.ACTION_NONE: key2 = ctrl.ACTION_W ctrl.kbd_down(VirtualKeyCode.W_key) if pos[0] >= 200 and key2 == ctrl.ACTION_W: key2 = ctrl.ACTION_NONE ctrl.kbd_up(VirtualKeyCode.W_key) # Forward if pos[0] > 800 and key2 == ctrl.ACTION_NONE: key2 = ctrl.ACTION_S ctrl.kbd_down(VirtualKeyCode.S_key) if pos[0] <= 800 and key2 == ctrl.ACTION_S: key2 = ctrl.ACTION_NONE ctrl.kbd_up(VirtualKeyCode.S_key) # 处理神经网络判断 t2 += 1 # 隔一段时间再判断 if t2 == 15: t2 = 0 self.lock.acquire() x = np.array(self.frames[len(self.frames) - self.select:]) self.lock.release() x = x.reshape((1, x.size)) # print('X shape:', x.shape) # res = model.train_on_batch(x=x, y=y) predict = model.predict(x=x)[0] predict = predict.tolist() res = predict.index(max(predict)) res = self.ACTIONS[res] # print('predict:', res) self.logger.push(UiLogger.Item(UiLogger.LEVEL_INFO, 'predict %.2f' % (time.time() - start), '%s' % res))
class MCHandle: ACTION_NONE = '无动作' ACTION_FORWARD = '前进' ACTION_JUMP = '起跳' ACTION_DOWN = '下降' ACTION_HIT = '打击' ACTION_PUT = '放置' ACTIONS = [ACTION_NONE, ACTION_FORWARD, ACTION_JUMP, ACTION_DOWN, ACTION_HIT, ACTION_PUT] def __init__(self, root=None): self.init_top = Tk() self.port_left = 'COM4' self.port_right = 'COM5' self.init_bps = StringVar() self.init_bps.set('115200') self.init_com_left = StringVar() self.init_com_left.set(self.port_left) self.init_com_right = StringVar() self.init_com_right.set(self.port_right) self.init_communication() self.bps = 115200 self.comm = None self.n = 512 self.select = 24 self.frames = [[0 for i in range(12)] for j in range(self.n)] self.raw = [[[0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]] for j in range(self.n)] # 建立网络 self.model_file = 'mc_actions.h5' # 建立网络的过程放在线程2 # 初始化手柄连接 self.comm_left = BaseComm(self.init_com_left.get(), self.bps) self.comm_right = BaseComm(self.init_com_right.get(), self.bps) # 灵敏度 self.sensitivity = 0.3 # 先读取几次,读取Init标志 print('预读取:') for i in range(10): r1 = self.comm_left.read1epoch() r2 = self.comm_right.read1epoch() print(r1, r2) # 初始化某些变化的值 # 加上Uilogger print('请保持手柄水平放置不动') # 初始化遥感的中点值(取平均) X Y X Y self.ave_left, self.ave_right = [0, 0], [0, 0] pick = 10 for i in range(pick): data = self.comm_left.read1epoch() data_ctrl = data[-4:] data_ctrl = list(map(int, data_ctrl)) self.ave_left[0] += data_ctrl[2] / pick self.ave_left[1] += data_ctrl[1] / pick data = self.comm_right.read1epoch() data_ctrl = data[-4:] data_ctrl = list(map(int, data_ctrl)) self.ave_right[0] += data_ctrl[2] / pick self.ave_right[1] += data_ctrl[1] / pick print('初始化遥感中点:', self.ave_left, self.ave_right) self.root = root if self.root is None: self.root = Tk() self.root.title("MC手柄") self.logger = UiLogger(self.root, height=10, width=32) self.logger.logger().grid(row=2, column=1, sticky=W+E) self.panel = Label(self.root) self.panel.grid(row=1, column=1, sticky=W+E) self.lock = threading.Lock() t = threading.Thread(target=self.read_thread) t.setDaemon(True) t.start() t = threading.Thread(target=self.parse_thread) t.setDaemon(True) t.start() def init_communication(self): top = self.init_top frame = LabelFrame(top, text="连接设置") Label(frame, text="左手柄").grid(row=1, column=1) Entry(frame, textvariable=self.init_com_left).grid(row=1, column=2) Label(frame, text="右手柄").grid(row=2, column=1) Entry(frame, textvariable=self.init_com_right).grid(row=2, column=2) Label(frame, text="波特率").grid(row=3, column=1) Entry(frame, textvariable=self.init_bps).grid(row=3, column=2) frame.grid(row=1, columnspan=3, column=1) Button(top, text="测试", command=self.init_communication_test).grid(row=2, column=1, sticky=W + E) Button(top, text="刷新", command=self.init_communication_refresh).grid(row=2, column=2, sticky=W + E) Button(top, text="确定", command=self.init_communication_ok).grid(row=2, column=3, sticky=W + E) top.mainloop() def init_communication_ok(self): try: bps = int(self.init_bps.get()) except ValueError: messagebox.showerror("错误", '数值错误!') return self.bps = bps self.port_left = self.init_com_left.get() self.port_right = self.init_com_right.get() if self.init_communication_test(show=False) is False: messagebox.showerror("错误", '手柄测试不通过!') return self.init_top.destroy() def mainloop(self): self.root.mainloop() def init_communication_test(self, show=True): try: bps = int(self.init_bps.get()) except ValueError: messagebox.showerror("错误", '数值错误!') return res = True print('测试左手柄') comm = BaseComm(self.init_com_left.get(), bps) if not comm.test(): if show is True: messagebox.showerror("错误", '测试左手柄失败') res = False comm.close() print('测试右手柄') comm = BaseComm(self.init_com_right.get(), bps) if not comm.test(): if show is True: messagebox.showerror("错误", '测试右手柄失败') res = False comm.close() return res def init_communication_refresh(self): pass def draw(self): width = 1 height = 32 colors = [ 'red', 'orange', 'yellow', 'green', 'cyan', 'blue', 'purple', 'red', 'orange', 'yellow', 'green', 'cyan', 'blue', 'purple', ] size = (width * self.n, height * 6) im = Image.new("RGB", size, color='white') draw = ImageDraw.Draw(im) for i in range(self.n - 2): for j in range(12): draw.line((width * i, self.frames[i][j] + size[1] / 2, width * (i + 1), self.frames[i + 1][j] + size[1] / 2), fill=colors[j]) sx = size[0] - width * self.select draw.line((sx, 0, sx, size[1]), fill='red') return im # 第二个线程,负责读取 def read_thread(self): while True: time.sleep(0.01) data_left = self.comm_left.read1epoch() data_right = self.comm_right.read1epoch() self.lock.acquire() self.raw.append([data_left, data_right]) if len(self.raw) > self.n: self.raw = self.raw[1:-1] self.lock.release() # frames添加数据 ann = data_left[0:6] ann.extend(data_right[0:6]) self.lock.acquire() self.frames.append(ann) if len(self.frames) > self.n: self.frames = self.frames[1:-1] self.lock.release() # print('ANN DATA:', ann) # 第三个线程,负责解析数据 def parse_thread(self): # 读取神经网络模型 model = load_model(self.model_file) t1, t2 = 0, 0 click = False key1 = ctrl.ACTION_NONE key2 = ctrl.ACTION_NONE jump = ctrl.ACTION_NONE start = time.time() while True: if t1 == 5: im = self.draw() imp = ImageTk.PhotoImage(image=im) self.panel.configure(image=imp) self.panel.image = imp t1 = 0 t1 += 1 time.sleep(0.01) # data_left = self.comm_left.read1epoch() # data_right = self.comm_right.read1epoch() self.lock.acquire() data_left = self.raw[-1][0] data_right = self.raw[-1][1] self.lock.release() # 右手处理 right_ctrl = data_right[-4:] right_ctrl = list(map(int, right_ctrl)) ctrl.move((right_ctrl[2] - self.ave_left[0]) * self.sensitivity, (right_ctrl[1] - self.ave_left[1]) * self.sensitivity) if right_ctrl[0] == 0 and click is False: ctrl.left_down() click = True if right_ctrl[0] == 1 and click is True: ctrl.left_up() click = False # Jump if data_left[6] == 0 and jump == ctrl.ACTION_NONE: jump = ctrl.ACTION_UP ctrl.kbd_down(VirtualKeyCode.SPACEBAR) # ctrl.kbd_click(VirtualKeyCode.SPACEBAR) if data_left[6] == 1 and jump == ctrl.ACTION_UP: jump = ctrl.ACTION_NONE ctrl.kbd_up(VirtualKeyCode.SPACEBAR) # ctrl.kbd_click(VirtualKeyCode.SPACEBAR) # 左手处理 pos = data_left[-4:][1:3] # Right if pos[1] > 800 and key1 == ctrl.ACTION_NONE: key1 = ctrl.ACTION_D ctrl.kbd_down(VirtualKeyCode.D_key) if pos[1] <= 800 and key1 == ctrl.ACTION_D: key1 = ctrl.ACTION_NONE ctrl.kbd_up(VirtualKeyCode.D_key) # Left if pos[1] < 200 and key1 == ctrl.ACTION_NONE: key1 = ctrl.ACTION_A ctrl.kbd_down(VirtualKeyCode.A_key) if pos[1] >= 200 and key1 == ctrl.ACTION_A: key1 = ctrl.ACTION_NONE ctrl.kbd_up(VirtualKeyCode.A_key) # Backward if pos[0] < 200 and key2 == ctrl.ACTION_NONE: key2 = ctrl.ACTION_W ctrl.kbd_down(VirtualKeyCode.W_key) if pos[0] >= 200 and key2 == ctrl.ACTION_W: key2 = ctrl.ACTION_NONE ctrl.kbd_up(VirtualKeyCode.W_key) # Forward if pos[0] > 800 and key2 == ctrl.ACTION_NONE: key2 = ctrl.ACTION_S ctrl.kbd_down(VirtualKeyCode.S_key) if pos[0] <= 800 and key2 == ctrl.ACTION_S: key2 = ctrl.ACTION_NONE ctrl.kbd_up(VirtualKeyCode.S_key) # 处理神经网络判断 t2 += 1 # 隔一段时间再判断 if t2 == 15: t2 = 0 self.lock.acquire() x = np.array(self.frames[len(self.frames) - self.select:]) self.lock.release() x = x.reshape((1, x.size)) # print('X shape:', x.shape) # res = model.train_on_batch(x=x, y=y) predict = model.predict(x=x)[0] predict = predict.tolist() res = predict.index(max(predict)) res = self.ACTIONS[res] # print('predict:', res) self.logger.push(UiLogger.Item(UiLogger.LEVEL_INFO, 'predict %.2f' % (time.time() - start), '%s' % res))
def __init__(self, root=None): self.init_top = Tk() self.port_left = 'COM4' self.port_right = 'COM5' self.init_bps = StringVar() self.init_bps.set('115200') self.init_com_left = StringVar() self.init_com_left.set(self.port_left) self.init_com_right = StringVar() self.init_com_right.set(self.port_right) self.init_communication() self.bps = 115200 self.comm = None self.n = 512 self.select = 24 self.frames = [[0 for i in range(12)] for j in range(self.n)] self.raw = [[[0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]] for j in range(self.n)] # 建立网络 self.model_file = 'mc_actions.h5' # 建立网络的过程放在线程2 # self.model = model # print(self.model.get_config()) self.comm_left = BaseComm(self.init_com_left.get(), self.bps) self.comm_right = BaseComm(self.init_com_right.get(), self.bps) self.root = root if self.root is None: self.root = Tk() self.root.title("MC手柄训练器") self.panel = Label(self.root) self.panel.pack(side=TOP, expand=1, fill=X) frame = Frame(self.root) Button(frame, text='切换模式', command=self.predict_mode).grid(row=1, column=1, sticky=W + E) Button(frame, text='前进', command=self.action_forward).grid(row=1, column=2, sticky=W + E) Button(frame, text='上跳', command=self.action_jump).grid(row=1, column=3, sticky=W + E) Button(frame, text='下降', command=self.action_down).grid(row=1, column=4, sticky=W + E) Button(frame, text='打击', command=self.action_hit).grid(row=1, column=5, sticky=W + E) Button(frame, text='放置', command=self.action_put).grid(row=1, column=6, sticky=W + E) Button(frame, text='无动作', command=self.action_none).grid(row=1, column=7, sticky=W + E) Button(frame, text='保存模型', command=self.save_model).grid(row=1, column=8, sticky=W + E) Label(frame, text='正在训练:').grid(row=1, column=9, sticky=W + E) self.var_training = StringVar() self.var_training.set('...') Label(frame, textvariable=self.var_training).grid(row=1, column=10, sticky=W + E) frame.pack(side=BOTTOM, expand=1, fill=X) self.logger_test = UiLogger(self.root, title='程序日志', simplify=False, height=10) self.logger_test.logger().pack(side=BOTTOM, expand=1, fill=X) self.lock = threading.Lock() self.training = self.ACTION_NONE self.will_save_model = False self.train_mode = True self.t1 = 0 self.t2 = 0 t = threading.Thread(target=self.read_thread) t.setDaemon(True) t.start() t = threading.Thread(target=self.parse_thread) t.setDaemon(True) t.start()
def parse_thread(self): # 建模 try: model = load_model(self.model_file) except OSError: print("Can't find", self.model_file) model = Sequential() model.add(Dense(self.select * 12, activation='tanh', input_dim=self.select * 12)) model.add(Dense(self.select * 24, activation='tanh')) model.add(Dense(self.select * 32, activation='tanh')) model.add(Dense(self.select * 48, activation='tanh')) model.add(Dense(self.select * 32, activation='tanh')) model.add(Dense(self.select * 24, activation='tanh')) model.add(Dense(self.select * 12, activation='tanh')) model.add(Dense(self.select, activation='tanh')) model.add(Dense(6, activation='softmax')) model.compile(loss='binary_crossentropy', optimizer='adam') start = time.time() while True: self.var_training.set(self.training) # 只需要MPU数据 self.lock.acquire() data_left = self.raw[-1][0][:6] data_right = self.raw[-1][1][:6] self.lock.release() # data_left = self.comm_left.read1epoch()[:6] # data_right = self.comm_right.read1epoch()[:6] # print(data) data = data_left data.extend(data_right) # print(data) # self.lock.acquire() # self.frames.append(data) # if len(self.frames) > self.n: # self.frames = self.frames[1:-1] # self.lock.release() if self.t1 == 5: im = self.draw() imp = ImageTk.PhotoImage(image=im) self.panel.configure(image=imp) self.panel.image = imp self.t1 = 0 self.t1 += 1 # 开始训练 if self.t2 == 5 and self.train_mode is True: self.lock.acquire() x = np.array(self.frames[len(self.frames) - self.select:]) self.lock.release() x = x.reshape((1, x.size)) # print('X shape:', x.shape) one = [0 for i in range(6)] one[self.ACTIONS.index(self.training)] = 1 y = np.array(one) y = y.reshape((1, 6)) # print('Y shape:', y.shape) self.t2 = 0 res = model.train_on_batch(x=x, y=y) # print('train:', res) self.logger_test.push(UiLogger.Item(UiLogger.LEVEL_INFO, 'training', '%s' % res)) self.t2 += 1 if self.will_save_model is True: print('保存模型...') self.lock.acquire() model.save(self.model_file) self.will_save_model = False self.lock.release() # 预测模式 if self.t2 == 5 and self.train_mode is False: self.t2 = 0 self.lock.acquire() x = np.array(self.frames[len(self.frames) - self.select:]) self.lock.release() x = x.reshape((1, x.size)) # print('X shape:', x.shape) # res = model.train_on_batch(x=x, y=y) predict = model.predict(x=x)[0] predict = predict.tolist() res = predict.index(max(predict)) res = self.ACTIONS[res] # print('predict:', res) self.logger_test.push(UiLogger.Item(UiLogger.LEVEL_INFO, 'predict %.2f' % (time.time() - start), '%s' % res))
class MCHandleTrainer: ACTION_NONE = '无动作' ACTION_FORWARD = '前进' ACTION_JUMP = '起跳' ACTION_DOWN = '下降' ACTION_HIT = '打击' ACTION_PUT = '放置' ACTIONS = [ACTION_NONE, ACTION_FORWARD, ACTION_JUMP, ACTION_DOWN, ACTION_HIT, ACTION_PUT] def __init__(self, root=None): self.init_top = Tk() self.port_left = 'COM4' self.port_right = 'COM5' self.init_bps = StringVar() self.init_bps.set('115200') self.init_com_left = StringVar() self.init_com_left.set(self.port_left) self.init_com_right = StringVar() self.init_com_right.set(self.port_right) self.init_communication() self.bps = 115200 self.comm = None self.n = 512 self.select = 24 self.frames = [[0 for i in range(12)] for j in range(self.n)] self.raw = [[[0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]] for j in range(self.n)] # 建立网络 self.model_file = 'mc_actions.h5' # 建立网络的过程放在线程2 # self.model = model # print(self.model.get_config()) self.comm_left = BaseComm(self.init_com_left.get(), self.bps) self.comm_right = BaseComm(self.init_com_right.get(), self.bps) self.root = root if self.root is None: self.root = Tk() self.root.title("MC手柄训练器") self.panel = Label(self.root) self.panel.pack(side=TOP, expand=1, fill=X) frame = Frame(self.root) Button(frame, text='切换模式', command=self.predict_mode).grid(row=1, column=1, sticky=W + E) Button(frame, text='前进', command=self.action_forward).grid(row=1, column=2, sticky=W + E) Button(frame, text='上跳', command=self.action_jump).grid(row=1, column=3, sticky=W + E) Button(frame, text='下降', command=self.action_down).grid(row=1, column=4, sticky=W + E) Button(frame, text='打击', command=self.action_hit).grid(row=1, column=5, sticky=W + E) Button(frame, text='放置', command=self.action_put).grid(row=1, column=6, sticky=W + E) Button(frame, text='无动作', command=self.action_none).grid(row=1, column=7, sticky=W + E) Button(frame, text='保存模型', command=self.save_model).grid(row=1, column=8, sticky=W + E) Label(frame, text='正在训练:').grid(row=1, column=9, sticky=W + E) self.var_training = StringVar() self.var_training.set('...') Label(frame, textvariable=self.var_training).grid(row=1, column=10, sticky=W + E) frame.pack(side=BOTTOM, expand=1, fill=X) self.logger_test = UiLogger(self.root, title='程序日志', simplify=False, height=10) self.logger_test.logger().pack(side=BOTTOM, expand=1, fill=X) self.lock = threading.Lock() self.training = self.ACTION_NONE self.will_save_model = False self.train_mode = True self.t1 = 0 self.t2 = 0 t = threading.Thread(target=self.read_thread) t.setDaemon(True) t.start() t = threading.Thread(target=self.parse_thread) t.setDaemon(True) t.start() def predict_mode(self): if self.train_mode is True: self.train_mode = False self.logger_test.push(UiLogger.Item(UiLogger.LEVEL_WARNING, 'switch', '切换到预测模式')) self.t2 = 0 else: self.train_mode = True self.logger_test.push(UiLogger.Item(UiLogger.LEVEL_WARNING, 'switch', '切换到训练模式')) self.t2 = 0 def action_forward(self): if self.training == self.ACTION_FORWARD: self.training = self.ACTION_NONE else: self.training = self.ACTION_FORWARD def action_jump(self): if self.training == self.ACTION_JUMP: self.training = self.ACTION_NONE else: self.training = self.ACTION_JUMP def action_down(self): if self.training == self.ACTION_DOWN: self.training = self.ACTION_NONE else: self.training = self.ACTION_DOWN def action_hit(self): if self.training == self.ACTION_HIT: self.training = self.ACTION_NONE else: self.training = self.ACTION_HIT def action_put(self): if self.training == self.ACTION_PUT: self.training = self.ACTION_NONE else: self.training = self.ACTION_PUT def action_none(self): self.training = self.ACTION_NONE def save_model(self): self.will_save_model = True def init_communication(self): top = self.init_top frame = LabelFrame(top, text="连接设置") Label(frame, text="左手柄").grid(row=1, column=1) Entry(frame, textvariable=self.init_com_left).grid(row=1, column=2) Label(frame, text="右手柄").grid(row=2, column=1) Entry(frame, textvariable=self.init_com_right).grid(row=2, column=2) Label(frame, text="波特率").grid(row=3, column=1) Entry(frame, textvariable=self.init_bps).grid(row=3, column=2) frame.grid(row=1, columnspan=3, column=1) Button(top, text="测试", command=self.init_communication_test).grid(row=2, column=1, sticky=W+E) Button(top, text="刷新", command=self.init_communication_refresh).grid(row=2, column=2, sticky=W+E) Button(top, text="确定", command=self.init_communication_ok).grid(row=2, column=3, sticky=W+E) top.mainloop() def init_communication_ok(self): try: bps = int(self.init_bps.get()) except ValueError: messagebox.showerror("错误", '数值错误!') return self.bps = bps self.port_left = self.init_com_left.get() self.port_right = self.init_com_right.get() if self.init_communication_test(show=False) is False: messagebox.showerror("错误", '手柄测试不通过!') return self.init_top.destroy() def mainloop(self): self.root.mainloop() def init_communication_test(self, show=True): try: bps = int(self.init_bps.get()) except ValueError: messagebox.showerror("错误", '数值错误!') return res = True print('测试左手柄') comm = BaseComm(self.init_com_left.get(), bps) if not comm.test(): if show is True: messagebox.showerror("错误", '测试左手柄失败') res = False comm.close() print('测试右手柄') comm = BaseComm(self.init_com_right.get(), bps) if not comm.test(): if show is True: messagebox.showerror("错误", '测试右手柄失败') res = False comm.close() return res def init_communication_refresh(self): pass # 单个手柄数据读取 def read_data(self, comm: BaseComm, q: queue.Queue): q.put(comm.read1epoch()) # 第二个线程,负责读取 def read_thread(self): while True: time.sleep(0.01) q_left = queue.Queue() q_right = queue.Queue() # data_left = self.comm_left.read1epoch() # data_right = self.comm_right.read1epoch() thread_left = threading.Thread(target=self.read_data, args=(self.comm_left, q_left)) thread_right = threading.Thread(target=self.read_data, args=(self.comm_right, q_right)) thread_left.setDaemon(True) thread_right.setDaemon(True) thread_left.start() thread_right.start() thread_left.join(5) thread_right.join(5) if q_left.empty() or q_right.empty(): print('WARING: 数据读取失败!') continue data_left = q_left.get() data_right = q_right.get() self.lock.acquire() self.raw.append([data_left, data_right]) if len(self.raw) > self.n: self.raw = self.raw[1:-1] self.lock.release() # frames添加数据 ann = data_left[0:6] ann.extend(data_right[0:6]) self.lock.acquire() self.frames.append(ann) if len(self.frames) > self.n: self.frames = self.frames[1:-1] self.lock.release() # print('ANN DATA:', ann) def parse_thread(self): # 建模 try: model = load_model(self.model_file) except OSError: print("Can't find", self.model_file) model = Sequential() model.add(Dense(self.select * 12, activation='tanh', input_dim=self.select * 12)) model.add(Dense(self.select * 24, activation='tanh')) model.add(Dense(self.select * 32, activation='tanh')) model.add(Dense(self.select * 48, activation='tanh')) model.add(Dense(self.select * 32, activation='tanh')) model.add(Dense(self.select * 24, activation='tanh')) model.add(Dense(self.select * 12, activation='tanh')) model.add(Dense(self.select, activation='tanh')) model.add(Dense(6, activation='softmax')) model.compile(loss='binary_crossentropy', optimizer='adam') start = time.time() while True: self.var_training.set(self.training) # 只需要MPU数据 self.lock.acquire() data_left = self.raw[-1][0][:6] data_right = self.raw[-1][1][:6] self.lock.release() # data_left = self.comm_left.read1epoch()[:6] # data_right = self.comm_right.read1epoch()[:6] # print(data) data = data_left data.extend(data_right) # print(data) # self.lock.acquire() # self.frames.append(data) # if len(self.frames) > self.n: # self.frames = self.frames[1:-1] # self.lock.release() if self.t1 == 5: im = self.draw() imp = ImageTk.PhotoImage(image=im) self.panel.configure(image=imp) self.panel.image = imp self.t1 = 0 self.t1 += 1 # 开始训练 if self.t2 == 5 and self.train_mode is True: self.lock.acquire() x = np.array(self.frames[len(self.frames) - self.select:]) self.lock.release() x = x.reshape((1, x.size)) # print('X shape:', x.shape) one = [0 for i in range(6)] one[self.ACTIONS.index(self.training)] = 1 y = np.array(one) y = y.reshape((1, 6)) # print('Y shape:', y.shape) self.t2 = 0 res = model.train_on_batch(x=x, y=y) # print('train:', res) self.logger_test.push(UiLogger.Item(UiLogger.LEVEL_INFO, 'training', '%s' % res)) self.t2 += 1 if self.will_save_model is True: print('保存模型...') self.lock.acquire() model.save(self.model_file) self.will_save_model = False self.lock.release() # 预测模式 if self.t2 == 5 and self.train_mode is False: self.t2 = 0 self.lock.acquire() x = np.array(self.frames[len(self.frames) - self.select:]) self.lock.release() x = x.reshape((1, x.size)) # print('X shape:', x.shape) # res = model.train_on_batch(x=x, y=y) predict = model.predict(x=x)[0] predict = predict.tolist() res = predict.index(max(predict)) res = self.ACTIONS[res] # print('predict:', res) self.logger_test.push(UiLogger.Item(UiLogger.LEVEL_INFO, 'predict %.2f' % (time.time() - start), '%s' % res)) def draw(self): width = 1 height = 32 colors = [ 'red', 'orange', 'yellow', 'green', 'cyan', 'blue', 'purple', 'red', 'orange', 'yellow', 'green', 'cyan', 'blue', 'purple', ] size = (width * self.n, height * 6) im = Image.new("RGB", size, color='white') draw = ImageDraw.Draw(im) for i in range(self.n - 2): for j in range(12): draw.line((width * i, self.frames[i][j] + size[1] / 2, width * (i + 1), self.frames[i + 1][j] + size[1] / 2), fill=colors[j]) sx = size[0] - width * self.select draw.line((sx, 0, sx, size[1]), fill='red') return im