def read(self):
"""Reads from the channels in the task."""
array_size_samps = self.sample_rate * self.samples_per_channel
fillmode = daqmx.parameters[self.fillmode]
if self.task_type == "analog input":
self.newdata, self.samples_per_channel_received = daqmx.ReadAnalogF64(
self.handle, self.samples_per_channel, self.timeout, fillmode,
array_size_samps, len(self.channels))
if self.append_data:
self.newdf = pd.DataFrame()
if self.time_array:
newtime = np.arange(self.samples_per_channel_received +
1) / self.sample_rate
if len(self.data["time"]) == 0:
self.newdf["time"] = newtime[:-1]
else:
last_time = self.data["time"].values[-1]
newtime += last_time
self.newdf["time"] = newtime[1:]
for n, channel in enumerate(self.channels):
self.newdf[channel.name] = self.newdata[:, n]
self.data = self.data.append(self.newdf, ignore_index=False)
if self.autolog:
self.do_autolog()
if self.autotrim:
self.autotrim_dataframe()
return self.newdata, self.samples_per_channel_received
def EveryNCallback_py(taskHandle, everyNsamplesEventType, nSamples,
callbackData_ptr):
callbackdata = daqmx.get_callbackdata_from_id(callbackData_ptr)
data, npoints = daqmx.ReadAnalogF64(taskHandle, buffsize, 10.0,
daqmx.Val_GroupByChannel, buffsize, 1)
callbackdata.extend(data.tolist())
print "Acquired %d samples" % len(data)
saved_data = np.load(filename)
new_saved_data = np.append(saved_data, np.asarray(data))
np.save(filename, new_saved_data)
return 0 # The function should return an integer
def get_data(self):
while True:
# Parameters for analog read
samps_per_chan = int(self.sr/10)
timeout = 10
fillmode = daqmx.Val_GroupByChannel
array_size_samps = 1000
nchan = 1
data1, spc = daqmx.ReadAnalogF64(self.th, samps_per_chan,
timeout, fillmode,
array_size_samps, nchan)
self.data = np.append(self.data, data1[:,0])
def EveryNCallback_py(taskHandle, everyNsamplesEventType, nSamples,
callbackData_ptr):
"""Function called every N samples"""
callbackdata = daqmx.get_callbackdata_from_id(callbackData_ptr)
data, npoints = daqmx.ReadAnalogF64(taskHandle, self.nsamps, 10.0,
daqmx.Val_GroupByChannel,
self.nsamps,
len(self.analogchans))
callbackdata.extend(data.tolist())
self.data["torque_trans"] = np.append(self.data["torque_trans"],
data[:, 0],
axis=0)
self.data["torque_arm"] = np.append(self.data["torque_arm"],
data[:, 1],
axis=0)
self.data["drag_left"] = np.append(self.data["drag_left"],
data[:, 2],
axis=0)
self.data["drag_right"] = np.append(self.data["drag_right"],
data[:, 3],
axis=0)
self.data["t"] = np.arange(len(self.data["torque_trans"]),
dtype=float) / self.sr
carpos, cpoints = daqmx.ReadCounterF64(self.carpostask,
self.nsamps, 10.0,
self.nsamps)
self.data["carriage_pos"] = np.append(self.data["carriage_pos"],
carpos)
turbang, cpoints = daqmx.ReadCounterF64(self.turbangtask,
self.nsamps, 10.0,
self.nsamps)
self.data["turbine_angle"] = np.append(self.data["turbine_angle"],
turbang)
self.data["turbine_rpm"] \
= ts.smooth(fdiff.second_order_diff(self.data["turbine_angle"],
self.data["t"])/6.0, 50)
return 0 # The function should return an integer
def get_observation(self):
read_t, _ = daqmx.ReadAnalogF64(self.AItask,self.i_size,self.timeout,self.mode,self.i_size,self.num_i)
return read_t / self.sens_coff
daqmx.CreateAOVoltageChan(AOtaskHandle, "Dev1/ao0", "", -10.0, 10.0,
daqmx.Val_Volts, None)
daqmx.CfgSampClkTiming(AOtaskHandle, "", sr, daqmx.Val_Rising,
daqmx.Val_ContSamps, 1000)
daqmx.CreateAIVoltageChan(AItaskHandle, "Dev1/ai1", "", daqmx.Val_Diff, -10.0,
10.0, daqmx.Val_Volts, None)
daqmx.CfgSampClkTiming(AItaskHandle, "", sr, daqmx.Val_Rising,
daqmx.Val_ContSamps, 1000)
written = daqmx.WriteAnalogF64(AOtaskHandle, 1000, 0, 10.0,
daqmx.Val_GroupByChannel, data)
dataread, nread = daqmx.ReadAnalogF64(AItaskHandle, 1000, 10.0,
daqmx.Val_GroupByChannel, 1000, 1)
daqmx.StartTask(AOtaskHandle)
daqmx.StartTask(AItaskHandle)
print written, "samples written"
print nread, "samples read"
plt.plot(dataread)
daqmx.StopTask(AItaskHandle)
daqmx.ClearTask(AItaskHandle)
daqmx.StopTask(AOtaskHandle)
daqmx.ClearTask(AOtaskHandle)
def main():
# Task parameters
taskhandle = daqmx.TaskHandle()
daqmx.CreateTask("", taskhandle)
phys_chan = b"Dev1/ai0"
nchan = 1
sr = 100.0
dt = 1 / sr
totaltime = 3
nloops = int(sr * totaltime / 10)
# Analog read parameters
samps_per_chan = 10
timeout = 10
fillmode = daqmx.Val_GroupByChannel
array_size_samps = 1000
daqmx.CreateAIVoltageChan(taskhandle, phys_chan, "", daqmx.Val_Diff, -10.0,
10.0, daqmx.Val_Volts, None)
daqmx.CfgSampClkTiming(taskhandle, "", sr, daqmx.Val_Rising,
daqmx.Val_ContSamps, 1000)
# Parameters for plotting
plotting = True
plot_dynamic = True
if plotting == True and plot_dynamic == True:
plt.ioff()
fig = plt.figure()
ax = fig.add_subplot(111)
line, = ax.plot([], )
fig.show()
daqmx.StartTask(taskhandle, fatalerror=False)
v = np.array([])
for n in range(nloops):
data, spc = daqmx.ReadAnalogF64(taskhandle, samps_per_chan, timeout,
fillmode, array_size_samps, nchan)
v = np.append(v, data[:, 0])
time.sleep(0.001)
t = np.float64(range(len(v))) * dt
if plotting == True and plot_dynamic == True:
line.set_xdata(t)
line.set_ydata(v)
ax.relim()
ax.autoscale_view()
fig.canvas.draw()
daqmx.ClearTask(taskhandle)
print("Task complete")
if plotting == True:
if plot_dynamic == False:
plt.plot(t, v)
plt.show()
return t, data
def run(self, Nepisode, targetQN, train=True):
self.setup_DAQmx()
# Parameters for plotting
# initiate
seq = np.zeros((self.num,self.NICH))
read = np.zeros((0,self.NCH))
state_t = np.zeros((self.SPC,self.NICH))
b = np.ones(self.num_mave)/self.num_mave
tilen = int(self.intispn / 0.01)
# start read analog
daqmx.StartTask(self.AItask, fatalerror=False)
# first loop
# you read
# you do not act
# this loop is for caliburation of reward
# you need to satrt writeanalog in first loop
# 'cause the function takes time to start
for n in range(self.nloops):
read_t, _ = daqmx.ReadAnalogF64(self.AItask,self.SPC,self.TO,self.MODE,self.ASS,self.NCH)
read_t = read_t/self.sensor_coff
# moving average filter
for ich in range(self.NICH):
state_t[self.num_mave-1:,ich] = np.convolve(read_t[:,self.IAIs[ich]], b, mode='vaild')
state_ave = ( np.average(np.split(state_t,int(self.num/tilen),0),axis=1) + self.dynamic_pressre )/self.dynamic_pressre
seq[self.num-int(self.num/tilen):self.num,:] = np.round(state_ave,2)
# action
# adopt output timing and action zero
if n!=10 or n!=20 or n!=40 or n!=50 or n!=60 or n!=80 or n!=90:
daqmx.WriteAnalogF64(self.AOtask,self.OB,1,self.TO,daqmx.Val_GroupByChannel,np.zeros(self.OB))
# reward
reward_t = np.average(read_t[:,self.CH100])
self.memory.add_local(seq,0,reward_t,0)
read = np.append(read,read_t,axis=0)
seq = shift(seq,[-self.num/tilen,0],cval=0)
# calibulate
self.memory.calc_calibulation()
# second loop
# you read
# you act
for n in range(self.nloops):
read_t, _ = daqmx.ReadAnalogF64(self.AItask,self.SPC,self.TO,self.MODE,self.ASS,self.NCH)
read_t = read_t/self.sensor_coff
# moving average filter
for ich in range(self.NICH):
state_t[self.num_mave-1:,ich] = np.convolve(read_t[:,self.IAIs[ich]], b, mode='vaild')
state_ave = ( np.average(np.split(state_t,int(self.num/tilen),0),axis=1) + self.dynamic_pressre )/self.dynamic_pressre
seq[self.num-int(self.num/tilen):self.num,:] = np.round(state_ave,2)
# action
ai = self.actor.get_action(seq.reshape(-1,self.num,self.NICH), self.QN,train)
daqmx.WriteAnalogF64(self.AOtask,self.OB,1,self.TO,daqmx.Val_GroupByChannel,self.PA[ai,:]*3)
# reward
reward_t = np.average(read_t[:,self.CH100])
self.memory.add_local(seq,ai,reward_t,0)
read = np.append(read,read_t,axis=0)
seq = shift(seq,[-self.num/tilen,0],cval=0)
# third loop
# you read
# you do not act
# make sure PA turn off
for n in range(self.nloops):
read_t, _ = daqmx.ReadAnalogF64(self.AItask,self.SPC,self.TO,self.MODE,self.ASS,self.NCH)
read_t = read_t/self.sensor_coff
# moving average filter
for ich in range(self.NICH):
state_t[self.num_mave-1:,ich] = np.convolve(read_t[:,self.IAIs[ich]], b, mode='vaild')
state_ave = ( np.average(np.split(state_t,int(self.num/tilen),0),axis=1) + self.dynamic_pressre )/self.dynamic_pressre
seq[self.num-int(self.num/tilen):self.num,:] = np.round(state_ave,2)
# action
# action zero
daqmx.WriteAnalogF64(self.AOtask,self.OB,1,self.TO,daqmx.Val_GroupByChannel,np.zeros(self.OB))
# reward
reward_t = np.average(read_t[:,self.CH100])
self.memory.add_local(seq,0,reward_t,0)
read = np.append(read,read_t,axis=0)
seq = shift(seq,[-self.num/tilen,0],cval=0)
# stop DAQmx
self.stop_DAQmx()
# edit experience in buffer
self.memory.edit_experience_local()
self.save(read,Nepisode)
total_reward = self.memory.totalreward()
# move current experience to global buffer
self.memory.add_global(total_reward)
if (self.memory.len() > self.batch_num) and train:
self.loss=self.QN.replay(self.memory,self.batch_num,self.gamma,targetQN)
self.actor.reduce_epsilon()
return total_reward, self.loss
diTaskHandle = daqmx.TaskHandle()
daqmx.CreateTask("", aiTaskHandle)
daqmx.CreateAIVoltageChan(aiTaskHandle, "Dev1/ai0", "", daqmx.Val_Diff, 0.0,
10.0, daqmx.Val_Volts)
daqmx.CfgSampClkTiming(aiTaskHandle, "", 100.0, daqmx.Val_Rising,
daqmx.Val_ContSamps, 1000)
trigname = daqmx.GetTerminalNameWithDevPrefix(aiTaskHandle, "ai/SampleClock")
daqmx.CreateTask("", diTaskHandle)
daqmx.CreateDIChan(diTaskHandle, "Dev1/port0", "", daqmx.Val_ChanForAllLines)
daqmx.CfgSampClkTiming(diTaskHandle, trigname, 100.0, daqmx.Val_Rising,
daqmx.Val_ContSamps, 1000)
daqmx.StartTask(diTaskHandle)
daqmx.StartTask(aiTaskHandle)
adata = daqmx.ReadAnalogF64(aiTaskHandle, 1000, 10.0, daqmx.Val_GroupByChannel,
1000, 1)
ddata = daqmx.ReadDigitalU32(diTaskHandle, 1000, 10.0,
daqmx.Val_GroupByChannel, 1000, 1)
print adata
print ddata
daqmx.StopTask(diTaskHandle)
daqmx.StopTask(aiTaskHandle)
daqmx.ClearTask(diTaskHandle)
daqmx.ClearTask(aiTaskHandle)