def getcatch(filen,dp,wvar,numhauls):
#where "dp" is the site code
#where "wvar" can be "kept", "total","shorts", or "eggers"
#returns haul,hauls,datetc where "hauls" is the smoothed version and "datetc" is the datetime
catch=csv.reader(open(filen,'r'), delimiter=',', quotechar='|',skipinitialspace=True)
haul,datetc=[],[]
for row in catch:
if (row[2]==dp) and len(row):
if wvar=='kept':
haul.append(float(row[11])/float(row[10]))# normalizes by number of traps per trawl (comma separated data)
elif wvar=='total': # as needed in Bill Doherty's case, for example
if row[12]=='':
row[12]=0
if row[13]=='':
row[13]=0
haul.append((float(row[11])+float(row[12])+float(row[13]))/float(row[10]))# normalizes by number of traps per trawl (comma separated data)
dd=row[5] #datetime
datetc.append(dt(int(dd[0:4]),int(dd[5:7]),int(dd[8:10]),int(dd[11:13]),int(dd[14:16]),0))
#sort this according to time?
haul=numpy.array(haul).transpose()
if numhauls>2:
haul_smooth=utilities.smooth(haul,numhauls,'hanning')
difflen=len(haul_smooth)-len(haul)
hauls=haul_smooth[difflen/2:-difflen/2]
else:
hauls=haul
return haul,hauls,datetc
def find_fault_time(self, window1=1, window2=4):
window_len1 = int(window1/self.hsw.step_len)
window_len2 = int(window2/self.hsw.step_len)
para_fault_id = np.array(self.para_fault_id)
para_fault_id = smooth(para_fault_id, 100)
for i in range(window_len1, len(para_fault_id)):
if (np.array(para_fault_id[i-window_len1:i])==0).all() and \
(np.array(para_fault_id[i:i+window_len2])!=0).all():
return (i+1)*self.hsw.step_len
return 0
def estimate_mode(self):
n_obs = self.obs / self.obs_scale
t_len, o_len = n_obs.shape
obs = n_obs.reshape(1, t_len, o_len)
obs = np2tensor(obs, use_cuda=False)
mode = self.mode_detector(obs)
mode = mode.detach().numpy()[0]
mode = np.argmax(mode, axis=1)
mode[:100] = self.mode0
mode = smooth(mode, 50)
self.mode = mode
def find_dominant_orientations(self):
''' '''
# flipud: why? I'm sure it won't work otherwise, but dont know why
# It should happen in both "find_dominant_orientations" & "find_grid_lines"
image = np.flipud(self.data['image'])
### computing orientations
if image is None:
orientations = np.array([])
else:
### smoothing
image = cv2.blur(image, (9, 9))
image = cv2.GaussianBlur(image, (9, 9), 0)
### oriented gradient of the image
# is it (dx - 1j*dy) or (dx + 1j*dy)
# this is related to "flipud" problem mentioned above
dx = cv2.Sobel(image, cv2.CV_64F, 1, 0, ksize=9)
dy = cv2.Sobel(image, cv2.CV_64F, 0, 1, ksize=9)
grd = dx - 1j * dy
### weighted histogram of oriented gradients (over the whole image)
hist, binc = utilities.wHOG(grd, NumBin=180 * 5, Extension=False)
hist = utilities.smooth(hist, window_len=21)
### finding peaks in the histogram
orthogonal_orientations = True
if orthogonal_orientations:
print(
'\t *** WARNING: currently auto-orientation detection assumes perpendicularity...'
)
# if dominant orientations are orthogonal, find only one and add pi/2 to it
peak_idx = np.argmax(hist)
orientations = [binc[peak_idx], binc[peak_idx] + np.pi / 2]
# shrinking the range to [-pi/2, pi/2]
for idx in range(len(orientations)):
if orientations[idx] < -np.pi / 2:
orientations[idx] += np.pi
elif np.pi / 2 <= orientations[idx]:
orientations[idx] -= np.pi
### setting orientations into places
self.data['dominant_orientation'] = np.array(orientations)
# note: that the internal unit of the orientation angles is radian
# it is converted to degree (and back) only for the user interaction
string = [
'{:.1f}'.format(a * 180 / np.pi)
for a in self.data['dominant_orientation']
]
self.ui.textEdit_dominant_orientations.setText(', '.join(string))
def track(self, mode, state_mu, state_sigma, obs, N):
msg = 'Tracking hybrid states...'
self.log_msg(msg)
self.mode0, self.state_mu0, self.state_sigma0, self.obs, self.N = mode, state_mu, state_sigma, obs, N
length = len(obs)
with progressbar.ProgressBar(max_value=length * self.hsw.step_len,
redirect_stdout=True) as bar:
i = 0
while i < length:
obs = self.obs[i]
particle = self.last_particle()
self.step(particle, obs)
bar.update(float('%.2f' % ((i + 1) * self.hsw.step_len)))
i += 1
self.mode = smooth(np.array(self.mode), 50)
def check_Z(self, window1=1, window2=2):
if self.fault_time>0:
return
window_len1 = int(window1/self.hsw.step_len)
window_len2 = int(window2/self.hsw.step_len)
if len(self.Z)<=(window_len1+window_len2):
return
Z = np.array(self.Z)
Z = smooth(Z, 100)
Z1 = np.array(Z[-(window_len1+window_len2):-window_len2])
Z2 = np.array(Z[-window_len2:])
Z1 = np.mean(Z1!=0, 0)
Z2 = np.mean(Z2!=0, 0)
if (Z1==0).all() and (Z2==1).any():
self.fault_time = self.t - window2
msg = 'A fault occurred at {}s, estimated its magnitude at 100s, fault parameters are mu=[0 0 0 ], sigma=[1 1 1 ].'\
.format(round(self.fault_time, 2)) # add some extra insignificant txt so that we can process log script conveniently.
self.log_msg(msg)
self.break_flag = True
def plot_mode(self, file_name=None):
data = np.array(self.mode)
data = smooth(data, 50)
self.hsw.plot_modes(data, file_name)
#la,lo=conversions.dm2dd([t[j,3]],[t[j,4]]) #changed this in May 2015
la,lo=conversions.dm2dd(df['lat'].values[0],df['lon'].values[0])
lat.append(la)
lon.append(lo)
if df['year_day'].values[j]+1<366.0:
#datet.append(num2date(t[j,1]+1).replace(year=int(t[j,0])))
datet.append(num2date(df['year_day'].values[j]+1).replace(year=int(df['year'].values[j])))
else:
#datet.append(num2date(t[j,1]+1).replace(year=int(t[j,0])+1))
datet.append(num2date(df['year_day'].values[j]+1).replace(year=int(df['year'].values[j]+1)))
#wd=t[:,5]
#temp=t[:,2]
#temp_smooth=utilities.smooth(temp,24*7,'hanning')
wd=df['depth'].values
temp=df['temp'].values
temp_smooth=utilities.smooth(temp,hrs,smoothing_method)
ax1.plot_date(datet,temp,'k-')
ax1.plot_date(datet,temp_smooth[hrs/2-1:-hrs/2],'r-',linewidth=4)
ax1.set_ylabel(str(days)+' day running average temperature (degF)',color='r',fontsize=16)
ax1.set_ylim(ylim)
haul,haul_smooth,datetc=getcatch(fileprefix+sc+'.csv',sc+depset,wvar,numhauls)
ax2=ax1.twinx()
ax2.plot(datetc,haul_smooth,'g-',linewidth=3)
ax2.set_ylabel(str(numhauls)+' haul average catch/pot',color='g',fontsize=16)
#ax2.set_ylim(xlim)
#ax2.xaxis.set_major_locator(yrsl)
#ax2.xaxis.set_major_formatter(majfmt)
#ax2.xaxis.set_major_formatter(Fmt)
#plt_fullmoon(datetc[-1].year,min(haul_smooth),max(haul_smooth))
index=range(len(obsTime)))
data = data.sort_index(by='obsTime')
data.index = range(len(obsTime))
Date = []
for i in data.index:
Date.append(data['obsTime'][i])
ave_obs = round(np.mean(obsILD), 1)
ave_wea_wind_speed = round(np.mean(wea_wind_speed), 1)
t = [data, Date, ave_obs, ave_wea_wind_speed]
T.append(t)
print 'e'
O, W2 = [], [] # smooth the model and observation ILD
for i in range(4):
num = 6 # smooth by 6 is the best
ild2_smooth = smooth(T[i][0]['obsILD'], num, 'hanning')
difflen2 = len(ild2_smooth) - len(T[i][0]['obsILD'])
ilds2 = ild2_smooth[difflen2 / 2:-difflen2 / 2]
wea_wind_smooth = smooth(T[i][0]['wea_wind_speed'], num, 'hanning')
difflen4 = len(wea_wind_smooth) - len(T[i][0]['wea_wind_speed'])
wind2 = wea_wind_smooth[difflen4 / 2:-difflen4 / 2]
O.append(ilds2)
W2.append(wind2)
fig = plt.figure()
for i in range(4):
ax1 = fig.add_subplot(
2,
2,
def detect_events(data, cell_index, stimulus, debug_plots=False):
stimulus_table = data.get_stimulus_table(stimulus)
dff_trace = data.get_dff_traces()[1][cell_index, :]
k_min = 0
k_max = 10
delta = 3
dff_trace = smooth(dff_trace, 5)
var_dict = {}
debug_dict = {}
for ii, fi in enumerate(stimulus_table['start'].values):
if ii > 0 and stimulus_table.iloc[ii].start == stimulus_table.iloc[
ii - 1].end:
offset = 1
else:
offset = 0
if fi + k_min >= 0 and fi + k_max <= len(dff_trace):
trace = dff_trace[fi + k_min + 1 + offset:fi + k_max + 1 + offset]
xx = (trace - trace[0])[delta] - (trace - trace[0])[0]
yy = max((trace - trace[0])[delta + 2] - (trace - trace[0])[0 + 2],
(trace - trace[0])[delta + 3] - (trace - trace[0])[0 + 3],
(trace - trace[0])[delta + 4] - (trace - trace[0])[0 + 4])
var_dict[ii] = (trace[0], trace[-1], xx, yy)
debug_dict[fi + k_min + 1 + offset] = (ii, trace)
xx_list, yy_list = [], []
for _, _, xx, yy in var_dict.itervalues():
xx_list.append(xx)
yy_list.append(yy)
mu_x = np.median(xx_list)
mu_y = np.median(yy_list)
xx_centered = np.array(xx_list) - mu_x
yy_centered = np.array(yy_list) - mu_y
std_factor = 1
std_x = 1. / std_factor * np.percentile(
np.abs(xx_centered), [100 * (1 - 2 * (1 - sps.norm.cdf(std_factor)))])
std_y = 1. / std_factor * np.percentile(
np.abs(yy_centered), [100 * (1 - 2 * (1 - sps.norm.cdf(std_factor)))])
curr_inds = []
allowed_sigma = 4
for ii, (xi, yi) in enumerate(zip(xx_centered, yy_centered)):
if np.sqrt(((xi) / std_x)**2 + ((yi) / std_y)**2) < allowed_sigma:
curr_inds.append(True)
else:
curr_inds.append(False)
curr_inds = np.array(curr_inds)
data_x = xx_centered[curr_inds]
data_y = yy_centered[curr_inds]
Cov = np.cov(data_x, data_y)
Cov_Factor = np.linalg.cholesky(Cov)
Cov_Factor_Inv = np.linalg.inv(Cov_Factor)
#===================================================================================================================
noise_threshold = max(allowed_sigma * std_x + mu_x,
allowed_sigma * std_y + mu_y)
mu_array = np.array([mu_x, mu_y])
yes_set, no_set = set(), set()
for ii, (t0, tf, xx, yy) in var_dict.iteritems():
xi_z, yi_z = Cov_Factor_Inv.dot((np.array([xx, yy]) - mu_array))
# Conditions in order:
# 1) Outside noise blob
# 2) Minimum change in df/f
# 3) Change evoked by this trial, not previous
# 4) At end of trace, ended up outside of noise floor
if np.sqrt(xi_z**2 + yi_z**2
) > 4 and yy > .05 and xx < yy and tf > noise_threshold / 2:
yes_set.add(ii)
else:
no_set.add(ii)
assert len(var_dict) == len(stimulus_table)
b = np.zeros(len(stimulus_table), dtype=np.bool)
for yi in yes_set:
b[yi] = True
if debug_plots == True:
import matplotlib.pyplot as plt
fig, ax = plt.subplots(1, 2)
# ax[0].plot(dff_trace)
for key, val in debug_dict.iteritems():
ti, trace = val
if ti in no_set:
ax[0].plot(np.arange(key, key + len(trace)), trace, 'b')
elif ti in yes_set:
ax[0].plot(np.arange(key, key + len(trace)),
trace,
'r',
linewidth=2)
else:
raise Exception
for ii in yes_set:
ax[1].plot([var_dict[ii][2]], [var_dict[ii][3]], 'r.')
for ii in no_set:
ax[1].plot([var_dict[ii][2]], [var_dict[ii][3]], 'b.')
print('number_of_events:', b.sum())
plt.show()
return b
data = data.sort_index(by='obsTime')
data.index = range(len(obsTime))
Date = []
for i in data.index:
Date.append(data['obsTime'][i])
ave_obs = round(np.mean(obsILD), 1)
ave_mod = round(np.mean(modILD), 1)
ave_hycom = round(np.mean(hycom_ILD), 1)
t = [data, Date, ave_obs, ave_mod, ave_hycom, data1, Date1, ave_fvcom]
T.append(t)
print e
H, M, O, F = [], [], [], [] # smooth the model and observation ILD
for i in range(2):
num = 6 # smooth by 6 is the best
ild0_smooth = utilities.smooth(T[i][0]['hycom_ILD'], num, 'hanning')
difflen0 = len(ild0_smooth) - len(T[i][0]['hycom_ILD'])
ilds0 = ild0_smooth[difflen0 / 2:-difflen0 / 2]
ild1_smooth = utilities.smooth(T[i][0]['modILD'], num, 'hanning')
difflen1 = len(ild1_smooth) - len(T[i][0]['modILD'])
ilds1 = ild1_smooth[difflen1 / 2:-difflen1 / 2]
ild2_smooth = utilities.smooth(T[i][0]['obsILD'], num, 'hanning')
difflen2 = len(ild2_smooth) - len(T[i][0]['obsILD'])
ilds2 = ild2_smooth[difflen2 / 2:-difflen2 / 2]
ild3_smooth = utilities.smooth(T[i][5]['fvcom_ILD'], num, 'hanning')
difflen3 = len(ild3_smooth) - len(T[i][5]['fvcom_ILD'])
ilds3 = ild3_smooth[difflen3 / 2:-difflen3 / 2]
