def trend(year, dh):
nt = len(year)
_, ny, nx = dh.shape
dhdt = np.zeros((ny, nx), 'f8') * np.nan
for i in range(ny):
for j in range(nx):
ii, = np.where(~np.isnan(dh[:, i, j]))
if len(ii) < 8:
continue
# TREND
m, c = ap.linear_fit(year, dh[:, i, j], return_coef=True)
dhdt[i, j] = m
if m == 0:
continue
return dhdt
def trend(year, dh):
nt = len(year)
_, ny, nx = dh.shape
dhdt = np.zeros((ny, nx), "f8") * np.nan
for i in range(ny):
for j in range(nx):
ii, = np.where(~np.isnan(dh[:, i, j]))
if len(ii) < 8:
continue
# TREND
m, c = ap.linear_fit(year, dh[:, i, j], return_coef=True)
dhdt[i, j] = m
if m == 0:
continue
return dhdt
def plot_ts(time2, lon, lat, dh_mean_cor, dh_mean, dg_mean, R, S, diff=True):
if np.alltrue(np.isnan(dh_mean[1:])):
return None
#time2 = y2dt(time2)
R = np.mean(R)
S = np.mean(S)
# use only non-null and non-zero entries for correlation
ind, = np.where((~np.isnan(dh_mean)) & (~np.isnan(dg_mean)) & \
(dh_mean!=0) & (dg_mean!=0))
t = np.arange(len(dh_mean))
if not diff:
x, y = ap.linear_fit(dg_mean[ind], dh_mean[ind])
x2, y2 = ap.linear_fit_robust(dg_mean[ind], dh_mean[ind])
fig = plt.figure()
ax = fig.add_subplot((111))
plt.plot(dg_mean[ind], dh_mean[ind], 'o')
plt.plot(x, y, linewidth=2, label='lstsq fit')
plt.plot(x2, y2, linewidth=2, label='robust fit')
plt.legend(loc=2).draw_frame(False)
plt.xlabel('dAGC (dB)')
plt.ylabel('dh (m)')
plt.title('Mixed-term sensitivity')
else:
dh_mean2 = np.diff(dh_mean)
dg_mean2 = np.diff(dg_mean)
dh_mean2 = np.append(dh_mean2, np.nan)
dg_mean2 = np.append(dg_mean2, np.nan)
x, y = ap.linear_fit(dg_mean2[ind], dh_mean2[ind])
x2, y2 = ap.linear_fit_robust(dg_mean2[ind], dh_mean2[ind])
fig = plt.figure()
ax = fig.add_subplot((111))
plt.plot(dg_mean2[ind], dh_mean2[ind], 'o')
plt.plot(x, y, linewidth=2, label='lstsq fit')
plt.plot(x2, y2, linewidth=2, label='robust fit')
plt.legend(loc=2).draw_frame(False)
plt.xlabel('$\Delta$dAGC (dB)')
plt.ylabel('$\Delta$dh (m)')
plt.title('Short-term sensitivity')
ax1 = viz.add_inner_title(ax, 'corrcoef: R = %.2f' % R, 3)
ax1 = viz.add_inner_title(ax, 'slope: S = %.2f' % S, 4)
plt.savefig('corr.png')
#-----------------
if not diff:
fig = plt.figure()
ax2 = plt.subplot((211))
plt.plot(time2, dh_mean, 'b', linewidth=2, label='dh')
plt.plot(time2, dh_mean_cor, 'r', linewidth=2, label='dh$_{COR}$')
#plt.legend().draw_frame(False)
viz.add_inner_title(ax2, 'dh', 2)
viz.add_inner_title(ax2, 'dh$_{COR}$', 3)
plt.title('lon = %.2f, lat = %.2f' % (lon, lat))
plt.ylabel('m')
#plt.xlim(1992, 2012.1)
#plt.gca().xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))
ax3 = plt.subplot((212))
plt.plot(time2, dg_mean, 'g', linewidth=2, label='dAGC')
#plt.legend().draw_frame(False)
viz.add_inner_title(ax3, 'dAGC', 3)
plt.ylabel('dB')
#plt.xlim(1992, 2012.1)
#plt.gca().xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))
else:
fig = plt.figure()
ax2 = plt.subplot((311))
plt.plot(time2, dh_mean, 'b', linewidth=2, label='dh')
plt.plot(time2, dh_mean_cor, 'r', linewidth=2, label='dh$_{COR}$')
#plt.legend().draw_frame(False)
viz.add_inner_title(ax2, 'dh', 2)
viz.add_inner_title(ax2, 'dh$_{COR}$', 3)
plt.title('lon = %.2f, lat = %.2f' % (lon, lat))
plt.ylabel('m')
#plt.xlim(1992, 2012.1)
#plt.gca().xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))
ax3 = plt.subplot((312))
plt.plot(time2, dh_mean2, 'm', linewidth=2, label='$\Delta$dh')
#plt.legend().draw_frame(False)
viz.add_inner_title(ax3, '$\Delta$dh', 3)
plt.ylabel('m')
#plt.xlim(1992, 2012.1)
#plt.gca().xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))
ax4 = plt.subplot((313))
plt.plot(time2, dg_mean2, 'c', linewidth=2, label='$\Delta$dAGC')
#plt.legend().draw_frame(False)
viz.add_inner_title(ax4, '$\Delta$dAGC', 3)
plt.ylabel('dB')
#plt.xlim(1992, 2012.1)
#plt.gca().xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))
fig.autofmt_xdate()
plt.savefig('ts.png')
return fig
def main():
fname_in = sys.argv[1]
din = GetData(fname_in, 'a')
satname = din.satname
time = change_day(din.time, 15) # change all days (e.g. 14,15,16,17) to 15
ts = getattr(din, VAR_TO_CALIBRATE)
err = din.dh_error
n_ad = din.n_ad
n_da = din.n_da
lon = din.lon
lat = din.lat
din.file.close()
t = ap.num2year(time)
if SUBSET: # get subset
ts, lon2, lat2 = ap.get_subset(ap.amundsen, ts, lon, lat)
err, lon2, lat2 = ap.get_subset(ap.amundsen, err, lon, lat)
n_ad, lon2, lat2 = ap.get_subset(ap.amundsen, n_ad, lon, lat)
n_da, lon2, lat2 = ap.get_subset(ap.amundsen, n_da, lon, lat)
lon, lat = lon2, lat2
xx, yy = np.meshgrid(lon, lat)
nt, ny, nx = ts.shape
offset_12 = np.full((ny, nx), np.nan)
offset_23 = np.full((ny, nx), np.nan)
print 'cross-calibrating time series:', VAR_TO_CALIBRATE
isfirst = True
if SAT_NAMES is None:
satnames = np.unique(din.satname)
# iterate over every grid cell (all times)
no_overlap_12 = 0
no_overlap_23 = 0
for i in xrange(ny):
for j in xrange(nx):
if 0:
i, j = ap.find_nearest2(xx, yy, (LON, LAT))
i -= 0
j += 0
print 'grid-cell:', i, j
ts_ij = ts[:, i, j]
if np.isnan(ts_ij).all(): continue
# get all time series (all sats) in one df (per grid-cell)
var = create_df_with_sats(time, ts_ij, satname, SAT_NAMES)
if DETREND:
var = var.apply(detrend)
if FILTER:
var = var.apply(ap.hp_filt, lamb=7, nan=True)
if PLOT_TS and (var.count().sum() > 10):
print 'grid-cell:', i, j
var.plot(linewidth=3, figsize=(9, 3), legend=False)
plt.title('Elevation change, dh (lon=%.2f, lat=%.2f)' %
(xx[i, j], yy[i, j]))
plt.ylabel('m')
plt.show()
# compute offset (if ts overlap)
#---------------------------------------------------
x = pd.notnull(var)
overlap_12 = x['ers1'] & x['ers2']
overlap_23 = x['ers2'] & x['envi']
if np.sometrue(overlap_12):
if SAT_BIAS:
s1 = var['ers1'][overlap_12]
s2 = var['ers2'][overlap_12]
if LINEAR_FIT:
# using linear fit
s1 = s1[s1.notnull() & s2.notnull()]
s2 = s2[s1.notnull() & s2.notnull()]
if len(s1) > 1 and len(s2) > 1:
s1.index, s1[:] = ap.linear_fit(
ap.date2year(s1.index), s1.values)
s2.index, s2[:] = ap.linear_fit(
ap.date2year(s2.index), s2.values)
offset = (s1.values[-1] - s1.values[0]) - (
s2.values[-1] - s2.values[0])
else:
pass
else:
# using absolute values
s1 = ap.referenced(s1, to='first')
s2 = ap.referenced(s2, to='first')
s1[0], s2[0] = np.nan, np.nan # remove first values
offset = np.nanmean(s1 - s2)
#pd.concat((s1, s2), axis=1).plot(marker='o')
else:
offset = np.nanmean(var['ers1'] - var['ers2'])
offset_12[i, j] = offset
else:
no_overlap_12 += 1
if np.sometrue(overlap_23):
if SAT_BIAS:
s2 = var['ers2'][overlap_23]
s3 = var['envi'][overlap_23]
if LINEAR_FIT:
s2 = s2[s2.notnull() & s3.notnull()]
s3 = s3[s2.notnull() & s3.notnull()]
if len(s2) > 1 and len(s3) > 1:
s2.index, s2[:] = ap.linear_fit(
ap.date2year(s2.index), s2.values)
s3.index, s3[:] = ap.linear_fit(
ap.date2year(s3.index), s3.values)
offset = (s2.values[-1] - s2.values[0]) - (
s3.values[-1] - s3.values[0])
else:
pass
else:
s2 = ap.referenced(s2, to='first')
s3 = ap.referenced(s3, to='first')
s2[0], s3[0] = np.nan, np.nan
offset = np.nanmean(s2 - s3)
#pd.concat((s2, s3), axis=1).plot(marker='o')
#plt.show()
else:
offset = np.nanmean(var['ers2'] - var['envi'])
offset_23[i, j] = offset
else:
no_overlap_23 += 1
#---------------------------------------------------
mean_offset_12 = np.nanmean(offset_12)
median_offset_12 = np.nanmedian(offset_12)
mean_offset_23 = np.nanmean(offset_23)
median_offset_23 = np.nanmedian(offset_23)
if SAVE_TO_FILE:
fout = tb.open_file(FNAME_OUT, 'w')
fout.create_array('/', 'lon', lon)
fout.create_array('/', 'lat', lat)
fout.create_array('/', 'offset_12', offset_12)
fout.create_array('/', 'offset_23', offset_23)
fout.close()
if PLOT:
plt.figure()
plt.subplot(211)
offset_12 = ap.median_filt(offset_12, 3, 3)
plt.imshow(offset_12,
origin='lower',
interpolation='nearest',
vmin=-.5,
vmax=.5)
plt.title('ERS1-ERS2')
plt.colorbar(shrink=0.8)
plt.subplot(212)
offset_23 = ap.median_filt(offset_23, 3, 3)
plt.imshow(offset_23,
origin='lower',
interpolation='nearest',
vmin=-.5,
vmax=.5)
plt.title('ERS2-Envisat')
#plt.colorbar(shrink=0.3, orientation='h')
plt.colorbar(shrink=0.8)
plt.figure()
plt.subplot(121)
o12 = offset_12[~np.isnan(offset_12)]
plt.hist(o12, bins=100)
plt.title('ERS1-ERS2')
ax = plt.gca()
ap.intitle('mean/median = %.2f/%.2f m' %
(mean_offset_12, median_offset_12),
ax=ax,
loc=2)
plt.xlim(-1, 1)
plt.subplot(122)
o23 = offset_23[~np.isnan(offset_23)]
plt.hist(o23, bins=100)
plt.title('ERS2-Envisat')
ax = plt.gca()
ap.intitle('mean/median = %.2f/%.2f m' %
(mean_offset_23, median_offset_23),
ax=ax,
loc=2)
plt.xlim(-1, 1)
plt.show()
print 'calibrated variable:', VAR_TO_CALIBRATE
print 'no overlaps:', no_overlap_12, no_overlap_23
print 'mean offset:', mean_offset_12, mean_offset_23
print 'median offset:', median_offset_12, median_offset_23
print 'out file ->', FNAME_OUT
def plot_ts(time2, lon, lat, dh_mean_cor, dh_mean, dg_mean, R, S, diff=True):
if np.alltrue(np.isnan(dh_mean[1:])):
return None
#time2 = y2dt(time2)
R = np.mean(R)
S = np.mean(S)
# use only non-null and non-zero entries for correlation
ind, = np.where((~np.isnan(dh_mean)) & (~np.isnan(dg_mean)) & \
(dh_mean!=0) & (dg_mean!=0))
t = np.arange(len(dh_mean))
if not diff:
x, y = ap.linear_fit(dg_mean[ind], dh_mean[ind])
x2, y2 = ap.linear_fit_robust(dg_mean[ind], dh_mean[ind])
fig = plt.figure()
ax = fig.add_subplot((111))
plt.plot(dg_mean[ind], dh_mean[ind], 'o')
plt.plot(x, y, linewidth=2, label='lstsq fit')
plt.plot(x2, y2, linewidth=2, label='robust fit')
plt.legend(loc=2).draw_frame(False)
plt.xlabel('dAGC (dB)')
plt.ylabel('dh (m)')
plt.title('Mixed-term sensitivity')
else:
dh_mean2 = np.diff(dh_mean)
dg_mean2 = np.diff(dg_mean)
dh_mean2 = np.append(dh_mean2, np.nan)
dg_mean2 = np.append(dg_mean2, np.nan)
x, y = ap.linear_fit(dg_mean2[ind], dh_mean2[ind])
x2, y2 = ap.linear_fit_robust(dg_mean2[ind], dh_mean2[ind])
fig = plt.figure()
ax = fig.add_subplot((111))
plt.plot(dg_mean2[ind], dh_mean2[ind], 'o')
plt.plot(x, y, linewidth=2, label='lstsq fit')
plt.plot(x2, y2, linewidth=2, label='robust fit')
plt.legend(loc=2).draw_frame(False)
plt.xlabel('$\Delta$dAGC (dB)')
plt.ylabel('$\Delta$dh (m)')
plt.title('Short-term sensitivity')
ax1 = viz.add_inner_title(ax, 'corrcoef: R = %.2f' % R, 3)
ax1 = viz.add_inner_title(ax, 'slope: S = %.2f' % S, 4)
plt.savefig('corr.png')
#-----------------
if not diff:
fig = plt.figure()
ax2 = plt.subplot((211))
plt.plot(time2, dh_mean, 'b', linewidth=2, label='dh')
plt.plot(time2, dh_mean_cor, 'r', linewidth=2, label='dh$_{COR}$')
#plt.legend().draw_frame(False)
viz.add_inner_title(ax2, 'dh', 2)
viz.add_inner_title(ax2, 'dh$_{COR}$', 3)
plt.title('lon = %.2f, lat = %.2f' % (lon, lat))
plt.ylabel('m')
#plt.xlim(1992, 2012.1)
#plt.gca().xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))
ax3 = plt.subplot((212))
plt.plot(time2, dg_mean, 'g', linewidth=2, label='dAGC')
#plt.legend().draw_frame(False)
viz.add_inner_title(ax3, 'dAGC', 3)
plt.ylabel('dB')
#plt.xlim(1992, 2012.1)
#plt.gca().xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))
else:
fig = plt.figure()
ax2 = plt.subplot((311))
plt.plot(time2, dh_mean, 'b', linewidth=2, label='dh')
plt.plot(time2, dh_mean_cor, 'r', linewidth=2, label='dh$_{COR}$')
#plt.legend().draw_frame(False)
viz.add_inner_title(ax2, 'dh', 2)
viz.add_inner_title(ax2, 'dh$_{COR}$', 3)
plt.title('lon = %.2f, lat = %.2f' % (lon, lat))
plt.ylabel('m')
#plt.xlim(1992, 2012.1)
#plt.gca().xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))
ax3 = plt.subplot((312))
plt.plot(time2, dh_mean2, 'm', linewidth=2, label='$\Delta$dh')
#plt.legend().draw_frame(False)
viz.add_inner_title(ax3, '$\Delta$dh', 3)
plt.ylabel('m')
#plt.xlim(1992, 2012.1)
#plt.gca().xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))
ax4 = plt.subplot((313))
plt.plot(time2, dg_mean2, 'c', linewidth=2, label='$\Delta$dAGC')
#plt.legend().draw_frame(False)
viz.add_inner_title(ax4, '$\Delta$dAGC', 3)
plt.ylabel('dB')
#plt.xlim(1992, 2012.1)
#plt.gca().xaxis.set_major_formatter(plt.FormatStrFormatter('%d'))
fig.autofmt_xdate()
plt.savefig('ts.png')
return fig
x = sin_data['x'].values
y = sin_data['y'].values
X = dmatrix('C(x, Poly)')
N = 5
w = 1/noise
out = ap.lstsq_cv(x, y, cv=10, max_deg=N, weight=w, randomise=True, return_coef=True)
y_wls, coef, deg, mse, var = out
y_ols = ap.lstsq_cv(x, y, cv=10, max_deg=N, weight=None, randomise=True)
a2 = np.polyfit(x, y, 1, w=None)#w)
y_line = np.polyval(a2, x)
m, c = ap.linear_fit(x, y, return_coef=True)
m2, c2 = ap.linear_fit_robust(x, y, return_coef=True)
out = ap.lasso_cv(x, y, cv=10, max_deg=N, return_model=True)
y_lasso, lasso = out
a = np.append(lasso.intercept_, lasso.coef_)
#y_lasso = np.dot(X[:,:N+1], a)
#dy_lasso = a[1] * X[:,0] + 2 * a[2] * X[:,1] # + 3 * a[3] * X[:,2]
dy_lasso = np.gradient(y_lasso, x[2] - x[1])
print a[1]
print 'coef.:', a
print 'slope:', y_lasso[-1] - y_lasso[0]
def main():
fname_in = sys.argv[1]
din = GetData(fname_in, 'a')
satname = din.satname
time = change_day(din.time, 15) # change all days (e.g. 14,15,16,17) to 15
ts = getattr(din, VAR_TO_CALIBRATE)
err = din.dh_error
n_ad = din.n_ad
n_da = din.n_da
lon = din.lon
lat = din.lat
din.file.close()
t = ap.num2year(time)
if SUBSET: # get subset
ts, lon2, lat2 = ap.get_subset(ap.amundsen, ts, lon, lat)
err, lon2, lat2 = ap.get_subset(ap.amundsen, err, lon, lat)
n_ad, lon2, lat2 = ap.get_subset(ap.amundsen, n_ad, lon, lat)
n_da, lon2, lat2 = ap.get_subset(ap.amundsen, n_da, lon, lat)
lon, lat = lon2, lat2
xx, yy = np.meshgrid(lon, lat)
nt, ny, nx = ts.shape
offset_12 = np.full((ny,nx), np.nan)
offset_23 = np.full((ny,nx), np.nan)
print 'cross-calibrating time series:', VAR_TO_CALIBRATE
isfirst = True
if SAT_NAMES is None:
satnames = np.unique(din.satname)
# iterate over every grid cell (all times)
no_overlap_12 = 0
no_overlap_23 = 0
for i in xrange(ny):
for j in xrange(nx):
if 0:
i, j = ap.find_nearest2(xx, yy, (LON,LAT))
i -= 0
j += 0
print 'grid-cell:', i, j
ts_ij = ts[:,i,j]
if np.isnan(ts_ij).all(): continue
# get all time series (all sats) in one df (per grid-cell)
var = create_df_with_sats(time, ts_ij, satname, SAT_NAMES)
if DETREND:
var = var.apply(detrend)
if FILTER:
var = var.apply(ap.hp_filt, lamb=7, nan=True)
if PLOT_TS and (var.count().sum() > 10):
print 'grid-cell:', i, j
var.plot(linewidth=3, figsize=(9, 3), legend=False)
plt.title('Elevation change, dh (lon=%.2f, lat=%.2f)' % (xx[i,j], yy[i,j]))
plt.ylabel('m')
plt.show()
# compute offset (if ts overlap)
#---------------------------------------------------
x = pd.notnull(var)
overlap_12 = x['ers1'] & x['ers2']
overlap_23 = x['ers2'] & x['envi']
if np.sometrue(overlap_12):
if SAT_BIAS:
s1 = var['ers1'][overlap_12]
s2 = var['ers2'][overlap_12]
if LINEAR_FIT:
# using linear fit
s1 = s1[s1.notnull() & s2.notnull()]
s2 = s2[s1.notnull() & s2.notnull()]
if len(s1) > 1 and len(s2) > 1:
s1.index, s1[:] = ap.linear_fit(ap.date2year(s1.index), s1.values)
s2.index, s2[:] = ap.linear_fit(ap.date2year(s2.index), s2.values)
offset = (s1.values[-1] - s1.values[0]) - (s2.values[-1] - s2.values[0])
else:
pass
else:
# using absolute values
s1 = ap.referenced(s1, to='first')
s2 = ap.referenced(s2, to='first')
s1[0], s2[0] = np.nan, np.nan # remove first values
offset = np.nanmean(s1 - s2)
#pd.concat((s1, s2), axis=1).plot(marker='o')
else:
offset = np.nanmean(var['ers1'] - var['ers2'])
offset_12[i,j] = offset
else:
no_overlap_12 += 1
if np.sometrue(overlap_23):
if SAT_BIAS:
s2 = var['ers2'][overlap_23]
s3 = var['envi'][overlap_23]
if LINEAR_FIT:
s2 = s2[s2.notnull() & s3.notnull()]
s3 = s3[s2.notnull() & s3.notnull()]
if len(s2) > 1 and len(s3) > 1:
s2.index, s2[:] = ap.linear_fit(ap.date2year(s2.index), s2.values)
s3.index, s3[:] = ap.linear_fit(ap.date2year(s3.index), s3.values)
offset = (s2.values[-1] - s2.values[0]) - (s3.values[-1] - s3.values[0])
else:
pass
else:
s2 = ap.referenced(s2, to='first')
s3 = ap.referenced(s3, to='first')
s2[0], s3[0] = np.nan, np.nan
offset = np.nanmean(s2 - s3)
#pd.concat((s2, s3), axis=1).plot(marker='o')
#plt.show()
else:
offset = np.nanmean(var['ers2'] - var['envi'])
offset_23[i,j] = offset
else:
no_overlap_23 += 1
#---------------------------------------------------
mean_offset_12 = np.nanmean(offset_12)
median_offset_12 = np.nanmedian(offset_12)
mean_offset_23 = np.nanmean(offset_23)
median_offset_23 = np.nanmedian(offset_23)
if SAVE_TO_FILE:
fout = tb.open_file(FNAME_OUT, 'w')
fout.create_array('/', 'lon', lon)
fout.create_array('/', 'lat', lat)
fout.create_array('/', 'offset_12', offset_12)
fout.create_array('/', 'offset_23', offset_23)
fout.close()
if PLOT:
plt.figure()
plt.subplot(211)
offset_12 = ap.median_filt(offset_12, 3, 3)
plt.imshow(offset_12, origin='lower', interpolation='nearest', vmin=-.5, vmax=.5)
plt.title('ERS1-ERS2')
plt.colorbar(shrink=0.8)
plt.subplot(212)
offset_23 = ap.median_filt(offset_23, 3, 3)
plt.imshow(offset_23, origin='lower', interpolation='nearest', vmin=-.5, vmax=.5)
plt.title('ERS2-Envisat')
#plt.colorbar(shrink=0.3, orientation='h')
plt.colorbar(shrink=0.8)
plt.figure()
plt.subplot(121)
o12 = offset_12[~np.isnan(offset_12)]
plt.hist(o12, bins=100)
plt.title('ERS1-ERS2')
ax = plt.gca()
ap.intitle('mean/median = %.2f/%.2f m' % (mean_offset_12, median_offset_12),
ax=ax, loc=2)
plt.xlim(-1, 1)
plt.subplot(122)
o23 = offset_23[~np.isnan(offset_23)]
plt.hist(o23, bins=100)
plt.title('ERS2-Envisat')
ax = plt.gca()
ap.intitle('mean/median = %.2f/%.2f m' % (mean_offset_23, median_offset_23),
ax=ax, loc=2)
plt.xlim(-1, 1)
plt.show()
print 'calibrated variable:', VAR_TO_CALIBRATE
print 'no overlaps:', no_overlap_12, no_overlap_23
print 'mean offset:', mean_offset_12, mean_offset_23
print 'median offset:', median_offset_12, median_offset_23
print 'out file ->', FNAME_OUT