def test_1d_shift(self):
x = [1, 2, 3, 4, 5, 6]
xp1 = np.roll(x, 1)
xn1 = np.roll(x, -1)
xp5 = np.roll(x, 5)
xn5 = np.roll(x, -5)
with self.test_session() as sess:
x_tf = tf.constant(x)
self.assertAllEqual(Utils.shift_1d(x_tf, 0).eval(), x)
self.assertAllEqual(Utils.shift_1d(x_tf, 1).eval(), xp1)
self.assertAllEqual(Utils.shift_1d(x_tf, -1).eval(), xn1)
self.assertAllEqual(Utils.shift_1d(x_tf, 5).eval(), xp5)
self.assertAllEqual(Utils.shift_1d(x_tf, -5).eval(), xn5)
def map_reward(self, state, electrode_weights=[], Fs=250, L=1000):
with tf.name_scope("Generate_Reward"):
reward = Utils.extract_frequency_bins(state, self.mLOWALPHA,
self.mHIGHALPHA, 1)
reward_flat = tf.reshape(reward, [-1], name="flatten_fft")
reward_summed = tf.reduce_sum(reward_flat, name="alphapow_summing")
return reward_summed
def bod(l_idx, specgram):
w_tmp = Utils.shift_1d(w_var, tf.cast(w_shift * l_idx, tf.int32))
s_tmp = tf.multiply(s_var, w_tmp)
fft_tmp = tf.fft(tf.cast(s_tmp, tf.complex64))
fft_padded = tf.pad(tf.expand_dims(
fft_tmp, 0), [[l_idx, tf.shape(specgram)[0] - l_idx - 1], [0, 0]])
#Update loop variables
l_idx = l_idx + 1
specgram = tf.add(specgram, fft_padded)
return [l_idx, specgram]
def bod(lcl_idx, lcl_r_var):
w_tmp = Utils.shift_1d(w_var, tf.cast(w_overlap * lcl_idx,
tf.int32))
s_tmp = tf.multiply(s_var, w_tmp)
fft_tmp = tf.fft(tf.cast(s_tmp, tf.complex64))
fft_padded = tf.pad(
tf.expand_dims(fft_tmp, 0),
[[lcl_idx, tf.shape(lcl_r_var)[0] - lcl_idx - 1], [0, 0]])
lcl_idx = lcl_idx + 1
lcl_r_var = tf.add(lcl_r_var, fft_padded)
return [lcl_idx, lcl_r_var]
def test_2d_array(self):
with self.test_session() as sess:
expected_val = np.int64([[500, 0, 0, 0, 0], [0, 0, 0, 0, 500]])
Sin1 = [np.sin(2 * np.pi * (x / 250)) for x in range(1000)]
Sin20 = [np.sin(2 * np.pi * (x / 250) * 20) for x in range(1000)]
a = tf.constant([Sin1, Sin20], dtype=tf.complex64)
b = Utils.extract_frequency_bins(a, 0, 25, 5)
c = tf.cast(b, dtype=tf.int64)
#Write graph to file
writer = tf.summary.FileWriter(".\\Logs\\", sess.graph)
#Actual check
self.assertAllEqual(c.eval(), expected_val)
def bod(l_idx, specgram):
w_tmp = Utils.shift_2d(w_var, tf.cast(w_shift * l_idx, tf.int32), 1)
s_tmp = tf.multiply(s_var, w_tmp)
fft_tmp = tf.fft(tf.cast(s_tmp, tf.complex64))
#Subtract one from the padding because fft takes up one row
fft_padded = tf.pad(tf.expand_dims(fft_tmp, 0),
[[l_idx,
(specgram_len - 1) - l_idx], [0, 0], [0, 0]])
#Update loop variables
specgram = tf.add(specgram, fft_padded)
with tf.control_dependencies([specgram]):
l_idx = l_idx + 1
return [l_idx, specgram]
def run(self, _buf):
with tf.name_scope("RUN_FIRFilt"):
tf.assert_rank(_buf['data'],
2,
message="JCR: Input must be rank 2 tensor")
asserts = [
tf.assert_equal(
tf.shape(_buf['data'])[0],
self.mNCHAN,
message="JCR: Input Dim-0 must equal number of channels")
]
with tf.control_dependencies(asserts):
dout = Utils.multi_ch_conv(_buf['data'], self.mCOEFFS)
return {
'data': dout,
'summaries': _buf['summaries'],
'updates': _buf['updates']
}
def run(self, _buf):
with tf.name_scope("RUN_IAFPower"):
tf.assert_rank(_buf['data'], 2, message="JCR: Input must be rank 2 tensor")
asserts= [
tf.assert_equal(tf.shape(_buf['data'])[0], self.mNCHAN, message="JCR: Input Dim-0 must equal number of channels")
]
with tf.control_dependencies(asserts):
s_len = tf.shape(_buf['data'])[1]
pphz = tf.realdiv(tf.cast(s_len, tf.float32) , tf.cast(self.mFS, tf.float32))
#This value (in Hz) is used to determine the peak frequency - larger window uses more surrounding values to calculate IAF
IAF_Peak_Window_Size = tf.realdiv(self.mPEAK_WINDOW, 2.0)
asserts = [
tf.assert_greater_equal(IAF_Peak_Window_Size, tf.realdiv(1.0, pphz), message="JCR: Invalid number of Hz/Window")
]
with tf.control_dependencies(asserts):
IAF_Window = tf.ones([tf.cast(tf.multiply(IAF_Peak_Window_Size,pphz),tf.int32)])
data_fft = tf.fft(tf.cast(_buf['data'],tf.complex64))
#Convolve the window over the FFT to strengthen peak
data_fft_neighbor_avg = Utils.multi_ch_conv(tf.cast(tf.abs(data_fft),tf.float32),IAF_Window)
half_size = tf.cast((tf.shape(data_fft_neighbor_avg)[1] / 2), tf.int32)
dout1 = tf.argmax(data_fft_neighbor_avg[:, 0:half_size], axis=1)
dout = tf.realdiv( tf.cast(dout1, tf.float32), tf.realdiv(tf.cast(half_size,tf.float32), tf.realdiv(tf.cast(self.mFS,tf.float32), 2.0)))
asserts = [
tf.assert_equal(tf.shape(dout)[0], self.mNCHAN, message="JCR: Input/output shape mismatch",name='FinalCheck')
]
with tf.control_dependencies(asserts):
return {
'data':dout,
'summaries':_buf['summaries'],
#pass dout shape to updates so that the assert gets evaluated
'updates': _buf['updates'] + [tf.shape(dout)[0]]
}
def run(self, _buf):
with tf.name_scope("RUN_BandPower"):
tf.assert_rank(_buf['data'],
2,
message="JCR: Input must be rank 2 tensor")
asserts = [
tf.assert_equal(
tf.shape(_buf['data'])[0],
self.mNCHAN,
message="JCR: Input Dim-0 must equal number of channels")
]
with tf.control_dependencies(asserts):
hz_per_point = self.mFS / self.mSIGLEN
points_per_bin = np.maximum(self.mBIN_WIDTH // hz_per_point,
1.0)
total_number_bins = int(self.mSIGLEN // points_per_bin)
total_number_points = int(total_number_bins * points_per_bin)
f_start = 0
f_end = total_number_points * self.mFS / self.mSIGLEN
dout = Utils.extract_frequency_bins(_buf['data'], f_start,
f_end, total_number_bins,
self.mSIGLEN, self.mFS)
asserts = [
tf.assert_equal(tf.shape(dout)[0],
self.mNCHAN,
message="JCR: Input/output shape mismatch",
name='FinalCheck')
]
with tf.control_dependencies(asserts):
return {
'data': dout,
'summaries': _buf['summaries'],
#pass dout shape to updates so that the assert gets evaluated
'updates': _buf['updates'] + [tf.shape(dout)[0]]
}
return tf.greater(specgram_len, tf.cast(l_idx, tf.int32))
return tf.while_loop(cond, bod, loopvars, parallel_iterations=1000)
tf.reset_default_graph()
##SPECGRAM TST
if True:
with tf.Session(config=tf.ConfigProto(log_device_placement=True)) as sess:
raw_data_tensor = tf.constant(np.asarray(np.transpose(rd)),
dtype=tf.float32)
b_coeffs = tf.constant(b, dtype=tf.float32)
data_bp_filt_tmp = Utils.multi_ch_conv(raw_data_tensor, b_coeffs)
data_bp_filt = tf.slice(data_bp_filt_tmp, [0, 300],
[-1, tf.shape(data_bp_filt_tmp)[1] - 600])
raw_data_tensor = data_bp_filt
z = specgram_tf_2d(raw_data_tensor, 500, 480)
init = tf.global_variables_initializer()
print("Global variables init, ret: ", sess.run(init))
writer = tf.summary.FileWriter(".\\Logs\\", sess.graph)
global spectro
run_options = tf.RunOptions(trace_level=tf.RunOptions.SOFTWARE_TRACE)
run_metadata = tf.RunMetadata()
spectro = sess.run([z], options=run_options, run_metadata=run_metadata)
writer.add_run_metadata(run_metadata, 'S: %d' % 1)