def test():
filepath = '../pygview_test_data/se_ne5np4_test.cam.h0.0000-01-01-00000.nc'
dslice = dataslice( filepath, 'T', time_ndx=0, level_ndx=8 )
print ' filepath : '+filepath
print 'dslice.data.shape : ',dslice.data.shape
print 'dslice.structured : ',dslice.structured
print ' dslice.units : ',dslice.units
print ' dslice.lon.size : ',dslice.lon.size
print ' dslice.lat.size : ',dslice.lat.size
print ' dslice.lon : ',dslice.lon
print ' dslice.lat : ',dslice.lat
print ' dslice.lon.min, dslice.lon.max : ',dslice.lon.min(), dslice.lon.max()
print ' dslice.lon.min, dslice.lon.max : ',numpy.amin(dslice.lon), numpy.amax(dslice.lon)
print ' dslice.lat.min, dslice.lat.max : ',numpy.amin(dslice.lat), numpy.amax(dslice.lat)
print ' dslice.data.min,dslice.data.max : ',numpy.amin(dslice.data), numpy.amax(dslice.data)
filepath = '../pygview_test_data/fv_10x15_test.cam.h0.0000-01-01-00000.nc'
dslice = dataslice( filepath, 'V', time_ndx=0, level_ndx=18 )
print ' filepath : '+filepath
print 'dslice.data.shape : ',dslice.data.shape
print 'dslice.structured : ',dslice.structured
print ' dslice.units : ',dslice.units
print ' dslice.lon : ',dslice.lon
print ' dslice.lat : ',dslice.lat
print ' dslice.data : ',dslice.data.min(),dslice.data.max()
def predictSoftmax(theta,data,label, numClasses,inputSize):
theta = theta.reshape(numClasses,inputSize+1)
y = np.zeros((data.shape[0], numClasses))
for i in range(len(label)):
k = np.zeros(numClasses)
k[label[i,0]] = 1
y[i] = (y[i] + k).astype(int)
theta_1 = np.dot(data, theta.T)
theta_1 = theta_1 - np.amax(theta_1, axis = 1).reshape(data.shape[0],1)
prob = np.exp(theta_1) # 10000*724 * 724*10 = 10000*10
sum_prob = np.sum(prob, axis = 1).reshape(data.shape[0],1) # 10000*1
prob = prob/sum_prob #10000*10
predict = prob/np.amax(prob, axis = 1).reshape(data.shape[0],1)
predict = (predict == 1.0 ).astype(float)
k = 0
for i in range(len(label)):
if np.array_equal(predict[i,:],y[i,:]):
k = k+1
correctness = k/(len(label))
return correctness
def test_more_known_parametrization_together():
R = 1
P = 1
toll = 7.e-3
intervals = 5
vs_order = 2
n = (intervals*(vs_order)+1-1)
#n = 18
ii = np.linspace(0,1,n+1)
n_1 = 2
n_2 = 4
control_points_3d = np.asarray(np.zeros([n+1,n_1,n_2,3]))#[np.array([R*np.cos(5*i * np.pi / (n + 1)), R*np.sin(5*i * np.pi / (n + 1)), P * i]) for i in range(0, n+1)]
for k in range(n_1):
for j in range(n_2):
control_points_3d[:,k,j,0] = np.array([R*np.cos(5*i * np.pi / (n + 1))for i in ii])
control_points_3d[:,k,j,1] = np.array([R*np.sin(5*i * np.pi / (n + 1))for i in ii])
control_points_3d[:,k,j,2] = np.array([(k+j+1)*P*i for i in range(n+1)])
#vsl = IteratedVectorSpace(UniformLagrangeVectorSpace(vs_order+1), np.linspace(0,1,intervals+1))
vsl = AffineVectorSpace(UniformLagrangeVectorSpace(n+1),0,1)
arky = ArcLengthParametrizer(vsl, control_points_3d)
new_control_points_3d = arky.reparametrize()
#print control_points_3d.shape, new_control_points_3d.shape
tt = np.linspace(0,1,128)
for k in range(n_1):
for j in range(n_2):
vals = vsl.element(control_points_3d)(tt)
new_vals = vsl.element(new_control_points_3d)(tt)
print (np.amax(np.abs(vals-new_vals))/(k+j+1)/P, (k+j+1))
assert np.amax(np.abs(vals-new_vals))/(k+j+1)/P < toll
def get_peaks_cf(data, win_size):
"""
data: audio as numpy array to be analyzed
win_size: value in samples to create the blocks for analysis
Used in calc_crest_factor, this function returns an array of peak levels
for each window.
return: array of peak audio levels
"""
if len(data) == 2:
# Seperate left and right channels
data_l = data[0,:]
data_r = data[1,:]
# Buffer up the data
data_matrix_l = librosa.util.frame(data_l, win_size, win_size)
data_matrix_r = librosa.util.frame(data_r, win_size, win_size)
# Get peaks for left and right channels
peaks_l = np.amax(np.absolute(data_matrix_l), axis=0)
peaks_r = np.amax(np.absolute(data_matrix_r), axis=0)
return np.maximum(peaks_l, peaks_r)
else:
data_matrix = librosa.util.frame(data, win_size, win_size)
return np.amax(np.absolute(data_matrix), axis=0)
def Seuil_var(img):
"""
This fonction compute threshold value. In first the image's histogram is calculated. The threshold value is set to the first indexe of histogram wich respect the following criterion : DH > 0, DH(i)/H(i) > 0.1 , H(i) < 0.01 % of the Norm.
In : img : ipl Image : image to treated
Out: seuil : Int : Value of the threshold
"""
dim=255
MaxValue=np.amax(np.asarray(img[:]))
Norm = np.asarray(img[:]).shape[0]*np.asarray(img[:]).shape[1]
scale=MaxValue/dim
Wdim=dim*scale
MaxValue=np.amax(np.asarray(img[:]))
bins= [float(x) for x in range(dim)]
hist,bin_edges = np.histogram(np.asarray(img[:]), bins)
Norm = Norm -hist[0]
median=np.median(hist)
mean=0
var=0
i=1
som = 0
while (som < 0.8*Norm and i <len(hist)-1):
som = som + hist[i]
i=i+1
while ((hist[i]-hist[i-1] < 0 or (hist[i]-hist[i-1])/hist[i-1]>0.1 or hist[i]> 0.01*Norm ) and i < len(hist)-1):
i=i+1
if( i == len(hist)-1):
seuil=0
seuil = i
var = 0
return seuil
def set_data(self, zname, zdata, zcolor):
if zdata!=None:
if self.overall_plot_type=="polygon":
if zname not in self.clts: #plottables['plotted']:#self.pd.list_data():
clt=PolyCollection(zdata, alpha=0.5, antialiased=True)#, rasterized=False, antialiased=False)
clt.set_color(colorConverter.to_rgba(zcolor))
self.clts[zname]=clt
self.axe.add_collection(self.clts[zname], autolim=True)
else:
self.clts[zname].set_verts(zdata)
if self.overall_plot_type=="XY":
if zname not in self.clts:
clt = LineCollection(zdata)#, offsets=offs)
clt.set_color(colors)
#print dir(clt)
self.clts[zname]=clt
self.axe.add_collection(self.clts[zname], autolim=True)
self.axe.autoscale_view()
else:
self.clts[zname].set_segments(zdata)
if self.overall_plot_type=="img":
if zname not in self.clts:
axeimg=self.axe.imshow( Magvec,
vmin=amin(Magvec),
vmax=0.001, #amax(Magvec),
aspect="auto", origin="lower",
extent=[amin(yoko),amax(yoko), amin(freq),amax(freq)],
#cmap='RdBu'
)
self.fig.colorbar(axeimg)
def test_known_parametrization():
R = 1
P = 1
toll = 2.e-3
n = 10
ii = np.linspace(0,1,n+1)
control_points_3d = np.asarray(np.zeros([n+1,3]))#[np.array([R*np.cos(5*i * np.pi / (n + 1)), R*np.sin(5*i * np.pi / (n + 1)), P * i]) for i in range(0, n+1)]
print (control_points_3d.shape)
control_points_3d[:,0] = np.array([R*np.cos(5*i * np.pi / (n + 1))for i in ii])
control_points_3d[:,1] = np.array([R*np.sin(5*i * np.pi / (n + 1))for i in ii])
control_points_3d[:,2] = np.array([P*i for i in range(n+1)])
vsl = AffineVectorSpace(UniformLagrangeVectorSpace(n+1),0,1)
arky = ArcLengthParametrizer(vsl, control_points_3d)
new_control_points_3d = arky.reparametrize()
#new_arky = ArcLengthParametrizer(vsl, new_control_points_3d)
#new_new_control_points_3d = arky.reparametrize()
tt = np.linspace(0, 1, 128)
vals = vsl.element(control_points_3d)(tt)
#print vals
new_vals = vsl.element(new_control_points_3d)(tt)
#print vals.shape, new_vals.shape
print (np.amax((np.abs(vals-new_vals))))
assert np.amax(np.abs(control_points_3d-new_control_points_3d))/P < toll
def xyzrgb_segment_kmeans(pc,colors,histograms,hist_bin_size=4):
"""Input:
- pc is a numpy array of points of size N * 3
- colors is an array of colors of size N * 3
- historgrams is a list of uv histograms
Output (labels,M,cost):
- labels: an N*1 vector of integers 0,...,M
- M: the number of histograms
- cost: a number denoting cost of the histogram.
better values are lower.
"""
assert pc.shape[1]==3
xscale = 2.0
yscale = 2.0
zscale = 1.0
hscale = np.amax(np.amax(pc,0)-np.amin(pc,0))
assert colors.shape[1]==3
features = np.zeros((pc.shape[0],3+len(histograms)))
features[:,0] = pc[:,0]*xscale
features[:,1] = pc[:,1]*yscale
features[:,2] = pc[:,2]*zscale
h = []
for i in xrange(pc.shape[0]):
uv = rgb_to_yuv(*colors[i,:])[1:3]
hi = [eval_uv_hist(hj,uv,hist_bin_size)*hscale for hj in histograms]
h.append(hi)
features[:,3:] = np.array(h)
naive_labeling = []
for i in xrange(pc.shape[0]):
hi,index = max((v,j) for (j,v) in enumerate(h[i]))
naive_labeling.append(index)
labels,quality = kmeans(features,len(histograms),initial=naive_labeling)
return (labels,len(histograms),quality)
def CAOSpy_run(tstart,tstop,mc,pdyn,particles,leftover,drained):
timenow=tstart
#loop through time
while timenow < tstop:
print 'time:',timenow
[thS,npart]=pdyn.gridupdate_thS(particles.lat,particles.z,mc)
#define dt as Curant criterion
dt_D=(mc.mgrid.vertfac.values[0])**2 / (2*np.amax(mc.D[np.amax(thS),:]))
dt_ku=-mc.mgrid.vertfac.values[0]/np.amax(mc.ku[np.amax(thS),:])
dt=np.amin([dt_D,dt_ku])
#INFILT
p_inf=cinf.pmx_infilt(timenow,precTS,mc,dt,leftover)
#print timenow
#print p_inf
particlesnow=pd.concat([particles,p_inf])
#p_backup=particlesnow.copy()
#DIFFUSION
[particlesnow,thS,npart,phi_mx]=pdyn.part_diffusion_split(particlesnow,npart,thS,mc,dt,True,10)
#ADVECTION
particlesnow=pdyn.mac_advection(particlesnow,mc,thS,dt)
#drained particles
drained=drained.append(particlesnow[particlesnow.flag==len(mc.maccols)+1])
particlesnow=particlesnow[particlesnow.flag!=len(mc.maccols)+1]
#MX-MAC-INTERACTION
pdyn.mx_mp_interact(particlesnow,npart,thS,mc,dt)
pondparts=(particlesnow.z<0.)
leftover=np.count_nonzero(-pondparts)
particles=particlesnow[pondparts]
timenow=timenow+dt
return(particles,npart,thS,leftover,drained,timenow)
def plot_Nhden(elem,N,hcol,hden,bounds=False):
for i in to_plot[elem]:
plt.clf()
x = np.array(hden,dtype=np.float)
y = np.array(N[i])
#x,y,hcol = trim(x,y,hcol)
y = hcol[0] - y
xlims=[0.75*np.amin(x), 1.25*np.amax(x)]
ylims=[0.75*np.amin(y), 1.25*np.amax(y)]
try:
if bounds:
l = minNHI - observed[elem][i]["column"][2]
if observed[elem][i]["column"][0]==-30.:
u=maxNHI
else:
u = maxNHI - observed[elem][i]["column"][0]
plt.fill([-30.,30., 30., -30.], [l,l,u,u], '0.50', alpha=0.2, edgecolor='b')
#plt.fill_between(np.arange(xlims[0],xlims[1]),lower,upper,color='0.50')
except KeyError:
pass
plt.plot(x, y, color_map[i],label=ion_state(i,elem))
plt.ylabel(r"log $N_{HI}/N_{%s}$"%(str(elem)+str(roman[i])))
plt.xlabel("log $n_{H}$")
plt.minorticks_on()
makedir('hden')
f=os.path.join(paths["plot_path"],"hden", elem+roman[i]+"N_Nhden.png")
plt.xlim([-3.,0.])
#plt.ylim(ylims)
plt.savefig(f)
plt.show()
plt.close()
def run_sim(R_star, transit_duration, bodies):
"""Run 3-body sim and convert results to TTV + TDV values in [minutes]"""
# Run 3-body sim for one full orbit of the outermost moon
loop(bodies, orbit_duration)
# Move resulting data from lists to numpy arrays
ttv_array = numpy.array([])
ttv_array = ttv_list
tdv_array = numpy.array([])
tdv_array = tdv_list
# Zeropoint correction
middle_point = numpy.amin(ttv_array) + numpy.amax(ttv_array)
ttv_array = numpy.subtract(ttv_array, 0.5 * middle_point)
ttv_array = numpy.divide(ttv_array, 1000) # km/s
# Compensate for barycenter offset of planet at start of simulation:
planet.px = 0.5 * (gravity_firstmoon + gravity_secondmoon)
stretch_factor = 1 / ((planet.px / 1000) / numpy.amax(ttv_array))
ttv_array = numpy.divide(ttv_array, stretch_factor)
# Convert to time units, TTV
ttv_array = numpy.divide(ttv_array, R_star)
ttv_array = numpy.multiply(ttv_array, transit_duration * 60 * 24) # minutes
# Convert to time units, TDV
oldspeed = (2 * R_star / transit_duration) * 1000 / 24 / 60 / 60 # m/sec
newspeed = oldspeed - numpy.amax(tdv_array)
difference = (transit_duration - (transit_duration * newspeed / oldspeed)) * 24 * 60
conversion_factor = difference / numpy.amax(tdv_array)
tdv_array = numpy.multiply(tdv_array, conversion_factor)
return ttv_array, tdv_array
def fit_min_max(args,p,max_iter,proj_list,proj_axis):
mins = numpy.array([])
maxs = numpy.array([])
kind = [item for item in args.kind.split(' ')]
for i in xrange(int(args.fmin)+int(args.step),max_iter+1,int(args.step)):
args.proj = proj_list[p]
axis = proj_axis[args.proj]
dat = load_map(args,p,i)
unit_l, unit_d, unit_t, unit_m = load_units(i, args)
if kind[p] == 'dens':
dat *= unit_d # in g/cc
if kind[p] in ['vx','vy','vz']:
dat *= (unit_l/unit_t)/1e5 # in km/s
if kind[p] in ['stars','dm']:
dat += 1e-12
if args.logscale:
mins = numpy.append(mins,numpy.log10(numpy.amin(dat)))
maxs = numpy.append(maxs,numpy.log10(numpy.amax(dat)))
else:
mins = numpy.append(mins,numpy.amin(dat))
maxs = numpy.append(maxs,numpy.amax(dat))
ii = range(int(args.fmin)+int(args.step),max_iter+1,int(args.step))
cmin = polyfit(ii,mins,args.poly)
cmax = polyfit(ii,maxs,args.poly)
return p, cmin, cmax
def iff_filter(sig, scale, plot_show = 0):
order = max(sig.size*scale,90)
#order = 80
# Extend signal on both sides for removing boundary effect in convolution
sig_extend = np.ones(sig.size+int(order/2)*2)
sig_extend[int(order/2):(sig.size+int(order/2))] = sig
sig_extend[0:int(order/2)] = sig[(sig.size-int(order/2)):sig.size]
sig_extend[(sig.size+int(order/2)):sig_extend.size] = sig[0:int(order/2)]
# convolve with hamming window and normalize
smooth_sig = np.convolve(sig_extend,np.hamming(order),'same')
smooth_sig = smooth_sig[int(order/2):(sig.size+int(order/2))]
smooth_sig = np.amax(sig)/np.amax(smooth_sig)*smooth_sig
# Plot signal for debug
if(plot_show == 1):
fig, ax = plt.subplots(ncols=2)
ax[0].plot(sig)
ax[0].plot(smooth_sig,'-r')
ax[0].plot(med_sig,'black')
ax[1].loglog(rfft(sig))
ax[1].loglog(rfft(smooth_sig),'-r')
ax[1].loglog(rfft(med_sig),'black')
plt.show()
return smooth_sig
def produce_video(self, interval=200, repeat_delay=2000, filename='video_output.gif', override_min_max=None):
"""
Finalize and save the video of the data.
interval and repeat_delay are the interval between frames and the repeat
delay before restart, both in milliseconds.
filename is the name of the file to save in the present working
directory. At present, only .gifs will implement reliably without
tweaking Python's PATHs.
override_min_max allows the user to set their own maximum and minimum
for the scale on the plot. Use a len-2 tuple, (min, max).
"""
#find the limits for the plot:
if not override_min_max:
self.min_limit = np.amin(self.data_list[0])
self.max_limit = np.amax(self.data_list[0])
assert len(self.data_list) > 1, 'You must include at least two frames to make an animation!'
for i in self.data_list[1:]: #assumes there is more than one frame in the loop
self.min_limit = min((self.min_limit, np.amin(i)))
self.max_limit = max((self.max_limit, np.amax(i)))
else:
self.min_limit=override_min_max[0]
self.max_limit=override_min_max[1]
self.fig.colorbar(self.plotfunc(self.grid, self.data_list[0],limits=(self.min_limit,self.max_limit),allow_colorbar=False, **self.kwds))
ani = animation.FuncAnimation(self.fig, _make_image, frames=self._yield_image, interval=interval, blit=True, repeat_delay=repeat_delay)
ani.save(filename, fps=1000./interval)
def create_histogram (mu, sigma, weights, bin_size, low_spec, high_spec, cu1_accepted, t1_failure_pos):
p1 = figure(title="Normal Distribution",tools = "pan,box_select,box_zoom,xwheel_zoom,reset,save,resize", background_fill="#E8DDCB")
measured = np.random.normal(mu, sigma, 1000)
hist, edges = np.histogram(weights, density=True, bins=bin_size)
x = np.linspace(np.amin(weights), np.amax(weights), 1000)
pdf = 1/(sigma * np.sqrt(2*np.pi)) * np.exp(-(x-mu)**2 / (2*sigma**2))
cdf = (1+scipy.special.erf((x-mu)/np.sqrt(2*sigma**2)))/2
p1.quad(top=hist, bottom=0, left=edges[:-1], right=edges[1:],
fill_color="#036564", line_color="#033649",\
)
sort_weights = sorted(weights)
cu1_yield = round(float(len(cu1_accepted))/(float(len(cu1_accepted)) + float(len(t1_failure_pos))),2)
p1.line(x, pdf, line_color="#D95B43", line_width=8, alpha=0.7, legend="PDF")
p1.line(low_spec, y=[0, np.amax(hist)], line_dash=[4, 4], line_color="orange", line_width=3, alpha=.5)
p1.line(high_spec, y=[0, np.amax(hist)], line_dash=[4, 4], line_color="orange", line_width=3, alpha=.5)
p1.line(weights[0], 0, line_width=1, legend='Mean = ' + str(round(mu, 3))) #daily rejected
p1.line(weights[0], 0, line_width=1, legend='2*Std (Std = ' + str(round(sigma, 3)) + ")") #daily accepted
p1.line(weights[0], 0, line_width=1, legend='Yield: ' + str(cu1_yield)) #daily rejected
p1.line(weights[0], 0, line_width=1, legend='Accepted: ' + str(len(cu1_accepted))) #daily accepted
p1.line(weights[0], 0, line_width=1, legend='Rejected: ' + str(len(t1_failure_pos))) #daily rejected
p1.xaxis.bounds = (np.amin(weights), np.amax(weights))
p1.legend.orientation = "top_left"
p1.xaxis.axis_label = 'Weight (g)'
p1.yaxis.axis_label = 'Pr(x)'
return p1
def c7_eval(self, dis1, dis2):
"""
********************* Peak Displacement *********************
"""
pgd1 = np.amax(dis1)
pgd2 = np.amax(dis2)
return self.eval_func(pgd1, pgd2)
def testRotMatOfExpMap(numpts):
"""Test rotation matrix from axial vector"""
print '* checking case of 1D vector input'
map = numpy.zeros(3)
rmat_1 = rotMatOfExpMap_orig(map)
rmat_2 = rotMatOfExpMap_opt(map)
print 'resulting shapes: ', rmat_1.shape, rmat_2.shape
#
#
map = numpy.random.rand(3, numPts)
map = numpy.zeros([3, numPts])
map[0, :] = numpy.linspace(0, numpy.pi, numPts)
#
print '* testing rotMatOfExpMap with %d random points' % numPts
#
t0 = time.clock()
rmat_1 = rotMatOfExpMap_orig(map)
et1 = time.clock() - t0
#
t0 = time.clock()
rmat_2 = rotMatOfExpMap_opt(map)
et2 = time.clock() - t0
#
print ' timings:\n ... original ', et1
print ' ... optimized', et2
#
drmat = numpy.absolute(rmat_2 - rmat_1)
print 'maximum difference between results'
print numpy.amax(drmat, 0)
return
def c5_eval(self, acc1, acc2):
"""
********************* Peak Acceleration *********************
"""
pga1 = np.amax(acc1)
pga2 = np.amax(acc2)
return self.eval_func(pga1, pga2)
def c6_eval(self, vel1, vel2):
"""
********************* Peak Velocity *********************
"""
pgv1 = np.amax(vel1)
pgv2 = np.amax(vel2)
return self.eval_func(pgv1, pgv2)
def backward_sparse(sparse_transition,reward_f,conv=5,discount = 1.,length = 15,z_states = None):
num_states = sparse_transition.shape[1]; num_actions = sparse_transition.shape[0]/num_states
#if reward_f.shape[0] ==num_actions:
# state_action = True
#else: state_action =False
z_actions = np.zeros(num_actions*num_states)
if z_states==None:
z_states = np.zeros(num_states)
#Backward - - - - - - - - - - - - - - - - - - - - - - - - - -
#print "Caus Ent Backward"
count = 0
delta = 0
reward_temp = reward_f.reshape(num_actions*num_states,order="F")
#while True:
alpha = (1/(num_states*100))*0
for i in range(length):
prev = np.zeros(z_states.shape)
prev += z_states
#print gamma*sparse_transition.dot(z_states)
z_actions = discount*((1-alpha)*sparse_transition.dot(z_states) + alpha*np.sum(z_states) ) +reward_temp
m = np.amax(z_actions.reshape(num_actions,num_states,order="F"),axis = 0)
z_states = m + np.log(np.sum(np.exp(z_actions.reshape(num_actions,num_states,order="F")-m),axis = 0))
count+=1
#Action Probability Computation - - - - - - - - - - - - - - - -
delta = np.amax(np.absolute(prev-z_states))
#print delta
#if count>2 and delta<conv:
#print "Count and delta", count,delta
policy= np.exp(z_actions.reshape(num_actions,num_states,order="F")-z_states)
#print "Policyyy",policy
#break
return policy,np.log(policy),z_states
def caus_ent_backward(transition,reward_f,conv=5,discount = 0.9,z_states = None):
num_actions = transition.tot_actions;num_states = transition.tot_states
if reward_f.shape[0] ==num_actions:
state_action = True
else: state_action =False
z_actions = np.zeros([num_actions,num_states])
if z_states==None:
z_states = np.zeros(num_states)
#Backward - - - - - - - - - - - - - - - - - - - - - - - - - -
#print "Caus Ent Backward"
count = 0
delta = 0
while True:
prev = np.zeros(z_states.shape)
prev += z_states
for i in range(num_states):
tr = transition.dense_backward[i]
ch = transition.chunks_backward[i]
out = discount*np.array(sum_chunks(tr[2,:]*z_states[map(int,tr[1,:])],ch))
z_actions[:,i] = out +reward_f[:,i]
m = np.amax(z_actions,axis = 0)
z_states = m + np.log(np.sum(np.exp(z_actions-m),axis = 0))
count+=1
#Action Probability Computation - - - - - - - - - - - - - - - -
delta = np.amax(np.absolute(prev-z_states))
if count == 50:
#print "Count and delta", count,delta
z_actions = z_actions
m = np.amax(z_actions,axis = 0)
z_states = m + np.log(np.sum(np.exp(z_actions-m),axis = 0))
policy= np.exp(z_actions-z_states)
break
return policy,np.log(policy),z_states
def CostVariancePlot(funct,args):
pl=args[0]
x=np.array([])
y=np.array([])
f=np.array([])
z=np.array([])
x=np.append(x,funct.rmsSet[:,0])
y=np.append(y,funct.rmsSet[:,3])
f=np.append(f,funct.rmsSet[:,5])
z=np.append(z,funct.rmsSet[:,4])
v=np.array([])
v=np.append(v,[0])
i=0
while i<len(x):
v=np.append(v,(z[i]/f[i])-(y[i]/f[i])*(y[i]/f[i]))
i+=1
v=np.delete(v,0)
if centroidP(x,v):
pl.set_yscale('log')
pl.set_xscale('log')
else:
pl.ticklabel_format(axis='both', style='sci', scilimits=(-2,5),pad=5,direction="bottom")
pl.axis([0, np.amax(x)+(10*np.amax(x)/100), 0, np.amax(v)+(10*np.amax(v)/100)])
pl.set_xlabel("read memory size",fontsize=8)
pl.set_ylabel("cost",fontsize=8)
pl.set_title("Variance Cost",fontsize=14)
pl.grid(True)
pl.tick_params(axis='x', labelsize=7)
pl.tick_params(axis='y', labelsize=7)
sc=pl.scatter(x,v,c=f,s=6,marker = 'o',lw=0.0,cmap=cmap,norm=norm)
pylab.close()
def main(X, Xtest, time):
global cut
global count
cut = 0
count = 0
root = node()
root.trainData = X
root.testData = Xtest
print("shape of xtest in main: ",np.shape(Xtest))
x1 = min(np.amin(X[:,[0]]), np.amin(Xtest[:,[0]]))-.05
x2 = max(np.amax(X[:,[0]]), np.amax(Xtest[:,[0]]))+.1
y1 = min(np.amin(X[:,[1]]), np.amin(Xtest[:,[1]]))-.05
y2 = max(np.amax(X[:,[1]]), np.amax(Xtest[:,[1]]))+.1
plt.figure()
plt.axis([x1,x2,y1,y2])
print("x1 x2 y1 y2: ",x1,x2,y1,y2)
root.coordinates.append([x1,x2])
root.coordinates.append([y1,y2])
leaves = []
MP(root,time,leaves)
point_index = {}
train_key = list(map(tuple,X))
test_key = list(map(tuple,Xtest))
x_shape = np.shape(X)
for i in range(x_shape[0]):
point_index[train_key[i]] = i
Xtest_shape = np.shape(Xtest)
for i in range(0,Xtest_shape[0]):
point_index[test_key[i]] = i
# plt.show()
plt.close()
return feature(leaves, point_index)
def checkcl(cluster_run, verbose = False):
"""Ensure that a cluster labelling is in a valid format.
Parameters
----------
cluster_run : array of shape (n_samples,)
A vector of cluster IDs for each of the samples selected for a given
round of clustering. The samples not selected are labelled with NaN.
verbose : Boolean, optional (default = False)
Specifies if status messages will be displayed
on the standard output.
Returns
-------
cluster_run : array of shape (n_samples,)
The input vector is modified in place, such that invalid values are
either rejected or altered. In particular, the labelling of cluster IDs
starts at zero and increases by 1 without any gap left.
"""
cluster_run = np.asanyarray(cluster_run)
if cluster_run.size == 0:
raise ValueError("\nERROR: Cluster_Ensembles: checkcl: "
"empty vector provided as input.\n")
elif reduce(operator.mul, cluster_run.shape, 1) != max(cluster_run.shape):
raise ValueError("\nERROR: Cluster_Ensembles: checkl: "
"problem in dimensions of the cluster label vector "
"under consideration.\n")
elif np.where(np.isnan(cluster_run))[0].size != 0:
raise ValueError("\nERROR: Cluster_Ensembles: checkl: vector of cluster "
"labellings provided as input contains at least one 'NaN'.\n")
else:
min_label = np.amin(cluster_run)
if min_label < 0:
if verbose:
print("\nINFO: Cluster_Ensembles: checkcl: detected negative values "
"as cluster labellings.")
cluster_run -= min_label
if verbose:
print("\nINFO: Cluster_Ensembles: checkcl: "
"offset to a minimum value of '0'.")
x = one_to_max(cluster_run)
if np.amax(cluster_run) != np.amax(x):
if verbose:
print("\nINFO: Cluster_Ensembles: checkcl: the vector cluster "
"labellings provided is not a dense integer mapping.")
cluster_run = x
if verbose:
print("INFO: Cluster_Ensembles: checkcl: brought modification "
"to this vector so that its labels range "
"from 0 to {0}, included.\n".format(np.amax(cluster_run)))
return cluster_run
def test_make_tone_regular_at_caldb():
fq = 15000
db = 100
fs = 100000
dur = 1
risefall = 0.002
calv = 0.1
caldb = 100
npts = fs*dur
tone, timevals = tools.make_tone(fq, db, dur, risefall, fs, caldb, calv)
assert len(tone) == npts
assert len(timevals) == npts
spectrum = np.fft.rfft(tone)
peak_idx = (abs(spectrum - max(spectrum))).argmin()
freq_idx = np.around(fq*(float(npts)/fs))
assert peak_idx == freq_idx
if tools.USE_RMS is True:
print 'tone max', np.around(np.amax(tone), 5), calv*np.sqrt(2)
assert np.around(np.amax(tone), 5) == np.around(calv*np.sqrt(2),5)
else:
assert np.around(np.amax(tone), 5) == calv
assert timevals[-1] == dur - (1./fs)
def basemap_raster_mercator(lon, lat, grid, cmap = None):
"""
Render a raster in mercator projection. Locations with no values are
rendered transparent.
"""
# longitude/latitude extent
lons = (np.amin(lon), np.amax(lon))
lats = (np.amin(lat), np.amax(lat))
if cmap is None:
cmap = mpl.cm.jet
cmap.set_bad('w', 1.0)
# construct spherical mercator projection for region of interest
m = Basemap(projection='merc',llcrnrlat=lats[0], urcrnrlat=lats[1],
llcrnrlon=lons[0],urcrnrlon=lons[1])
vmin,vmax = np.nanmin(grid),np.nanmax(grid)
masked_grid = np.ma.array(grid,mask=np.isnan(grid))
fig = plt.figure(frameon=False)
plt.axis('off')
m.pcolormesh(lon,lat,masked_grid,latlon=True,cmap=cmap,vmin=vmin,vmax=vmax)
str_io = StringIO.StringIO()
plt.savefig(str_io,bbox_inches='tight',format='png',pad_inches=0,transparent=True)
bounds = [ (lons[0],lats[0]),(lons[1],lats[0]),(lons[1],lats[1]),(lons[0],lats[1]) ]
return str_io.getvalue(), bounds
def _write_data(lock, im, index, outfile, outshape, outtype, rescale_factor, logfilename, cputime, itime):
lock.acquire()
try:
t0 = time()
f_out = getHDF5(outfile, 'a')
f_out_dset = f_out.require_dataset('exchange/data', outshape, outtype, chunks=tdf.get_dset_chunks(outshape[0]))
im = im * rescale_factor
tdf.write_tomo(f_out_dset,index,im.astype(outtype))
# Set minimum and maximum:
if (amin(im[:]) < float(f_out_dset.attrs['min'])):
f_out_dset.attrs['min'] = str(amin(im[:]))
if (amax(im[:]) > float(f_out_dset.attrs['max'])):
f_out_dset.attrs['max'] = str(amax(im[:]))
f_out.close()
t1 = time()
# Print out execution time:
log = open(logfilename,"a")
log.write(linesep + "\ttomo_%s processed (CPU: %0.3f sec - I/O: %0.3f sec)." % (str(index).zfill(4), cputime, t1 - t0 + itime))
log.close()
finally:
lock.release()
def mamPlot(funct,args):
pl=args[0]
x=np.array([])
ymin=np.array([])
yavg=np.array([])
ymax=np.array([])
f=np.array([])
x=np.append(x,funct.rmsSet[:,0])
ymin=np.append(ymin,funct.rmsSet[:,1])
ymax=np.append(ymax,funct.rmsSet[:,2])
t1=funct.rmsSet[:,3]
t2=funct.rmsSet[:,5]
yavg=np.append(yavg,t1/t2)
f=np.append(f,funct.rmsSet[:,5])
if centroidP(x,yavg):
pl.set_yscale('log')
pl.set_xscale('log')
else:
pl.ticklabel_format(axis='both', style='sci', scilimits=(-2,5),pad=5,direction="bottom")
pl.axis([0, np.amax(x)+(2*np.amax(x)/100), 0, np.amax(ymax)+(2*np.amax(ymax)/100)])
pl.set_xlabel('read memory size',fontsize=8)
pl.set_ylabel("cost",fontsize=8)
pl.grid(True)
pl.set_title("Min/Avg/Max Cost",fontsize=14)
pl.tick_params(axis='x', labelsize=7)
pl.tick_params(axis='y', labelsize=7)
sc=pl.scatter(x,ymax,s=7,c='r', marker = 'o',lw=0.0)
sc1=pl.scatter(x,yavg,s=5.5,c='g', marker = 'o',lw=0.0)
sc2=pl.scatter(x,ymin,s=4,c='b', marker = 'o',lw=0.0)
pl.legend((sc2,sc1,sc),("Min","Avg","Max"),scatterpoints=1,ncol=3,bbox_to_anchor=[0.5, mamAdjust],loc="lower center",fontsize=8)
pylab.close()
def statprint(host_per_pg, pg_per_host):
val = pg_per_host.values() # sets val to a list of the values in pg_per_host
mean = numpy.mean(val)
maxvalue = numpy.amax(val)
minvalue = numpy.amin(val)
std = numpy.std(val)
median = numpy.median(val)
variance = numpy.var(val)
print("for placement groups on hosts: ")
print( "the mean is: ", mean)
print( "the max value is: ", maxvalue)
print( "the min value is: ", minvalue)
print( "the standard deviation is: ", std)
print( "the median is: ", median)
print( "the variance is: ", variance)
# prints statements for stats
host_mean = numpy.mean(host_per_pg)
host_max = numpy.amax(host_per_pg)
host_min = numpy.amin(host_per_pg)
host_std = numpy.std(host_per_pg)
host_median = numpy.median(host_per_pg)
host_variance = numpy.var(host_per_pg)
# these are the variables for hosts/pgs
print("hosts per placement group: ")
print("the mean is: ", host_mean)
print("the max value is: ", host_max)
print("the min value is: ", host_min)
print("the standard deviation is: ", host_std)
print("the median is: ", host_median)
print("the variance is: ", host_variance)
def caus_ent_backward_nodisount(transition,reward_f,steps):
num_actions = transition.tot_actions;num_states = transition.tot_states
if reward_f.shape[0] ==num_actions:
state_action = True
else: state_action =False
gamma = discount
z_actions = np.zeros([num_actions,num_states])
z_states = np.zeros(num_states)
#Backward - - - - - - - - - - - - - - - - - - - - - - - - - -
print "Caus Ent Backward"
count = 0
delta = 0
for j in range(steps):
prev = np.zeros(z_states.shape)
prev += z_states
for i in range(num_states):
tr = transition.dense_backward[i]
ch = transition.chunks_backward[i]
out = gamma*np.array(sum_chunks(tr[2,:]*z_states[map(int,tr[1,:])],ch))
z_actions[:,i] = out +reward_f[:,i]
m = np.amax(z_actions)
z_states = m + np.log(np.sum(np.exp(z_actions-m),axis = 0))
count+=1
#Action Probability Computation - - - - - - - - - - - - - - - -
delta = np.sum(np.sum(np.absolute(prev-z_states)))
#delta +=1
#print "DElta cause",delta,delta2
if j==steps-1:
z_actions = z_actions
m = np.amax(z_actions)
z_states = m + np.log(np.sum(np.exp(z_actions-m),axis = 0))
policy= np.exp(z_actions-z_states)
return policy,np.log(policy),z_states
atm[-1] -= center_about
'''
File Output
'''
f = open('FEASST_' + TRAPPE_fname.split('_')[-1].split('.')[0] + '.in', "w")
#Heading
f.write('# FEASST data file \n# %s \n# %s \n# %s \n' %
(TRAPPE_name[0], TRAPPE_fname, SMILES))
#Preface
f.write('\n%s atoms \n%s bonds \n\n%s atom types \n%s bond types \n' %
(len(TRAPPE_atoms[:, 0]), len(TRAPPE_bonds[:, 0]),
len(TRAPPE_atoms[:, 1]), len(TRAPPE_bonds[:, 2])))
f.write('\n%s %s xlo xhi \n%s %s ylo yhi \n%s %s zlo zhi \n' % (
np.amin(psuedo_atoms.positions[:, 0]), np.amax(
psuedo_atoms.positions[:, 0]), np.amin(psuedo_atoms.positions[:, 1]),
np.amax(psuedo_atoms.positions[:, 1]), np.amin(
psuedo_atoms.positions[:, 2]), np.amax(psuedo_atoms.positions[:, 2])))
#Masses
f.write('\nMasses\n\n')
for i, atm in enumerate(organized_ASE_atoms):
f.write('%s %s \n' % (atm[0], atm[1]))
#Pair Coeffs
f.write('\nPair Coeffs\n\n')
for i, line in enumerate(TRAPPE_atoms):
f.write('%s %s %s \n' % (TRAPPE_atoms[i, 0], float(TRAPPE_atoms[i, 3]) *
kB[0] * NA[0] / 1000, TRAPPE_atoms[i, 4]))
#Bond Coeffs
def run_MLP_pipeline(args, train_adata, test_adata, result_dir, prefix=""):
''' Run MLP pipeline for left out cell types
@arg: argparse object
- args.leaveout: indicate up to how many cell types will be randomly left out
- args.threshold: use threshold to decide unassigned cells
'''
batch_size = 128
celltype_cols = "cell.type"
## Hyperparameters for network
if train_adata.shape[0] >= 5000:
## === parameters for mousebrain (high cell number)
dims = [128, 32]
MLP_dims = [128, 64, 32, 16, 8]
else:
## === parameters for PBMC datasets (low cell number)
dims = [16]
MLP_dims = [64, 16]
## OneHotEncoding the celltypes
enc = OneHotEncoder(handle_unknown='ignore')
if scipy.sparse.issparse(train_adata.X):
x_train = train_adata.X.toarray()
else:
x_train = train_adata.X
y_train = enc.fit_transform(train_adata.obs[[celltype_cols]]).toarray()
if scipy.sparse.issparse(test_adata.X):
x_test = test_adata.X.toarray()
else:
x_test = test_adata.X
### --- run MLP
print("\n\n=== MLP\n")
start = time.time()
y_pred = method_utils.run_MLP(x_train, y_train, x_test,
dims=MLP_dims, batch_size=batch_size, seed=RANDOM_SEED) ## run MLP
end = time.time()
print("\n\n=== Run time:", end-start)
### draw unassigned cells lower than a certain threshold
thres = args.threshold
test_adata.obs['pred_celltypes'] = 'unassigned'
assigned_idx = np.where(np.amax(y_pred, 1) >= thres)[0]
unassigned_idx = np.where(np.amax(y_pred, 1) < thres)[0]
## print out unassigned cells number
print(test_adata.obs[celltype_cols].value_counts())
print("Unassigned cells number:", len(unassigned_idx))
print(test_adata[unassigned_idx].obs[celltype_cols].value_counts())
## get assigned cell labels
assigned_labels = y_pred[assigned_idx].argmax(1)
n_clusters = len(set(train_adata.obs[celltype_cols]))
assigned_onehot = np.zeros((assigned_labels.size, n_clusters))
assigned_onehot[np.arange(assigned_labels.size), assigned_labels] = 1
assigned_celltypes = enc.inverse_transform(assigned_onehot)
assigned_cells = test_adata[assigned_idx].obs_names
test_adata.obs.loc[assigned_cells, 'pred_celltypes'] = assigned_celltypes
test_adata.obs.to_csv(result_dir+os.sep+prefix+"_predicted_obs.csv")
## visualization of unassigned cells
plt.figure(args.leaveout+1)
sc.pl.tsne(test_adata, color=[celltype_cols, "pred_celltypes"], size=15)
plt.savefig(result_dir+os.sep+prefix+'_'+str(args.threshold)+"_prediction_result.png")
print("=== Finish visualizing..")
return y_pred
acc_flr = flrs_results.mean()
acc_bf = (blds_results*flrs_results).mean()
# rfps_results = (np.equal(np.argmax(test_labels[:, 8:118], axis=1), np.argmax(preds[:, 8:118], axis=1))).astype(int)
# acc_rfp = rfps_results.mean()
# acc = (blds_results*flrs_results*rfps_results).mean()
# calculate positioning error when building and floor are correctly estimated
mask = np.logical_and(blds_results, flrs_results) # mask index array for correct location of building and floor
x_test_utm = x_test_utm[mask]
y_test_utm = y_test_utm[mask]
blds = blds[mask]
flrs = flrs[mask]
rfps = (preds[mask])[:, 8:118]
n_success = len(blds) # number of correct building and floor location
blds = np.greater_equal(blds, np.tile(np.amax(blds, axis=1).reshape(n_success, 1), (1, 3))).astype(int) # set maximum column to 1 and others to 0 (row-wise)
flrs = np.greater_equal(flrs, np.tile(np.amax(flrs, axis=1).reshape(n_success, 1), (1, 5))).astype(int) # ditto
n_loc_failure = 0
sum_pos_err = 0.0
sum_pos_err_weighted = 0.0
idxs = np.argpartition(rfps, -N)[:, -N:] # (unsorted) indexes of up to N nearest neighbors
threshold = scaling*np.amax(rfps, axis=1)
for i in range(n_success):
xs = []
ys = []
ws = []
for j in idxs[i]:
rfp = np.zeros(110)
rfp[j] = 1
rows = np.where((train_labels == np.concatenate((blds[i], flrs[i], rfp))).all(axis=1)) # tuple of row indexes
import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
x = np.load('xdata400.dat')
y = np.load('ydata400.dat')
z = np.load('zdata400.dat')
mag = np.load('magdata400.dat')
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
for a in range(0, len(mag)):
for b in range(0, len(mag)):
for c in range(0, len(mag)):
ax.scatter(x[a][b][c],y[a][b][c],z[a][b][c],marker='o',
alpha=(mag[a][b][c]/np.amax(mag)))
plt.show()
def activelearning(x_train, y_train, x_test, y_test, start_num_samples, end_num_samples , step_size, epochs, n_average, criteria ):
print("===================================")
submission = pd.DataFrame(columns=['#','accuracy', 'best_accuracy'])
cur_x = x_train[0:start_num_samples,:,:]
cur_y = y_train[0:start_num_samples]
#iterative retrain after applying the network to trained model
back_x = x_train[start_num_samples:,:,:]
back_y = y_train[start_num_samples:]
n_iteration = (end_num_samples-start_num_samples)//step_size
for j in range(n_iteration):
avg_acc = 0
best_acc = 0
for k in range(n_average):
model = get_model()
callbacks = build_callbacks()
model.fit(cur_x, cur_y, callbacks=[callbacks], epochs=epochs, validation_data=(x_test, y_test), verbose = 0)
model =tf.keras.models.load_model("params.h5")
score, acc = model.evaluate(x_test, y_test, verbose=0)
avg_acc += acc
if best_acc < acc :
best_acc = acc
bestmodel = tf.keras.models.load_model("params.h5")
submission = submission.append(pd.DataFrame({'#': cur_x.shape[0], 'accuracy': [avg_acc/n_average] , 'best_accuracy':[best_acc] }))
# use the remaining trainning datasets
predictions = bestmodel.predict (back_x)
prob = tf.nn.softmax(predictions)
if criteria == "en":
index = entropy_sampling(prob)
elif criteria == "rs":
index = random_sampling(prob)
elif criteria == "ms":
index = margin_sampling(prob)
elif criteria == "lc":
index = least_confidence(prob)
else:
raise ValueError("Unknow criteria value ")
# pick up step_size worst samples to train
cur_x = np.concatenate((cur_x, back_x[index[0:step_size], :,:]) )
cur_y = np.concatenate((cur_y, back_y[index[0:step_size]]) )
back_x = back_x[index[step_size:],:,:]
back_y = back_y[index[step_size:]]
# autolabel : use the remaining data
# back_x, back_y
size = back_x.shape[0]
n_iteration = size//step_size
# reduce the size to save time for quick testing, add in the end , we can add during the sampling steps above
n_iteration = min(n_iteration, 32)
for j in range(n_iteration):
predictions = bestmodel.predict (back_x)
prob = tf.nn.softmax(predictions)
y_autolabeled = np.argmax(prob, axis=1)
pmax = np.amax(prob, axis=1)
pidx = np.argsort(pmax)
back_x = back_x[pidx]
y_autolabeled = y_autolabeled[pidx]
cur_x = np.concatenate([cur_x, back_x[-step_size:]])
cur_y = np.concatenate([cur_y, y_autolabeled[-step_size:]])
back_x = back_x[:-step_size]
avg_acc = 0
best_acc = 0
for k in range(n_average):
model = get_model()
callbacks = build_callbacks()
model.fit(cur_x, cur_y, callbacks=[callbacks], epochs=epochs, validation_data=(x_test, y_test), verbose = 0)
model =tf.keras.models.load_model("params.h5")
score, acc = model.evaluate(x_test, y_test, verbose=0)
avg_acc += acc
if best_acc < acc :
best_acc = acc
bestmodel = tf.keras.models.load_model("params.h5")
submission = submission.append(pd.DataFrame({'#': cur_x.shape[0], 'accuracy': [avg_acc/n_average] , 'best_accuracy':[best_acc] }))
submission.to_csv('method_' + criteria + ".csv", index=False)
def perturb(self, params):
"""
Unlike in C++, this takes a numpy array of parameters as input,
and modifies it in-place. The return value is still logH.
"""
logH = 0.0
reps = 1;
if(rng.rand() < 0.5):
reps += np.int(np.power(100.0, rng.rand()));
# print "going to perturb %d reps" % reps
for i in range(reps):
# print " rep iteration %d" % i
which = rng.randint(len(params))
if which == 0:
rad_idx = 0
theta_idx = 2
theta = params[theta_idx]
#FIND THE MAXIMUM RADIUS STILL INSIDE THE DETECTOR
theta_eq = np.arctan(detector.detector_length/detector.detector_radius)
theta_taper = np.arctan(detector.taper_length/detector.detector_radius)
# print "theta: %f pi" % (theta / np.pi)
if theta <= theta_taper:
z = np.tan(theta)*(detector.detector_radius - detector.taper_length) / (1-np.tan(theta))
max_rad = z / np.sin(theta)
elif theta <= theta_eq:
max_rad = detector.detector_radius / np.cos(theta)
# print "max rad radius: %f" % max_rad
else:
theta_comp = np.pi/2 - theta
max_rad = detector.detector_length / np.cos(theta_comp)
# print "max rad length: %f" % max_rad
#AND THE MINIMUM (from PC dimple)
#min_rad = 1./ ( np.cos(theta)**2/detector.pcRad**2 + np.sin(theta)**2/detector.pcLen**2 )
min_rad = np.amax([detector.pcRad, detector.pcLen])
total_max_rad = np.sqrt(detector.detector_length**2 + detector.detector_radius**2 )
params[which] += total_max_rad*dnest4.randh()
params[which] = dnest4.wrap(params[which] , min_rad, max_rad)
elif which ==2: #theta
rad_idx = 0
rad = params[rad_idx]
# print "rad: %f" % rad
if rad < np.amin([detector.detector_radius - detector.taper_length, detector.detector_length]):
max_val = np.pi/2
min_val = 0
# print "theta: min %f pi, max %f pi" % (min_val, max_val)
else:
if rad < detector.detector_radius - detector.taper_length:
#can't possibly hit the taper
# print "less than taper adjustment"
min_val = 0
elif rad < np.sqrt(detector.detector_radius**2 + detector.taper_length**2):
#low enough that it could hit the taper region
# print "taper adjustment"
a = detector.detector_radius - detector.taper_length
z = 0.5 * (np.sqrt(2*rad**2-a**2) - a)
min_val = np.arcsin(z/rad)
else:
#longer than could hit the taper
# print " longer thantaper adjustment"
min_val = np.arccos(detector.detector_radius/rad)
if rad < detector.detector_length:
max_val = np.pi/2
else:
max_val = np.pi/2 - np.arccos(detector.detector_length/rad)
# print "theta: min %f pi, max %f pi" % (min_val, max_val)
params[which] += np.pi/2*dnest4.randh()
params[which] = dnest4.wrap(params[which], min_val, max_val)
# if which == 0:
# params[which] += (detector.detector_radius)*dnest4.randh()
# params[which] = dnest4.wrap(params[which] , 0, detector.detector_radius)
elif which == 1:
max_val = np.pi/4
params[which] += np.pi/4*dnest4.randh()
params[which] = dnest4.wrap(params[which], 0, max_val)
if params[which] < 0 or params[which] > np.pi/4:
print "wtf phi"
#params[which] = np.clip(params[which], 0, max_val)
# elif which == 2:
# params[which] += (detector.detector_length)*dnest4.randh()
# params[which] = dnest4.wrap(params[which] , 0, detector.detector_length)
elif which == 3: #scale
min_scale = wf.wfMax - 0.01*wf.wfMax
max_scale = wf.wfMax + 0.005*wf.wfMax
params[which] += (max_scale-min_scale)*dnest4.randh()
params[which] = dnest4.wrap(params[which], min_scale, max_scale)
# print " adjusted scale to %f" % ( params[which])
elif which == 4: #t0
params[which] += 1*dnest4.randh()
params[which] = dnest4.wrap(params[which], min_maxt, max_maxt)
elif which == 5: #smooth
params[which] += 0.1*dnest4.randh()
params[which] = dnest4.wrap(params[which], 0, 25)
# print " adjusted smooth to %f" % ( params[which])
# elif which == 6: #wf baseline slope (m)
# logH -= -0.5*(params[which]/1E-4)**2
# params[which] += 1E-4*dnest4.randh()
# logH += -0.5*(params[which]/1E-4)**2
# elif which == 7: #wf baseline incercept (b)
# logH -= -0.5*(params[which]/1E-2)**2
# params[which] += 1E-2*dnest4.randh()
# logH += -0.5*(params[which]/1E-2)**2
# params[which] += 0.01*dnest4.randh()
# params[which]=dnest4.wrap(params[which], -1, 1)
# print " adjusted b to %f" % ( params[which])
else: #velocity or rc params: cant be below 0, can be arb. large
print "which value %d not supported" % which
exit(0)
return logH
quantized_model.eval()
with torch.no_grad():
student_output_img = quantized_model(input_img.cpu())
# save imgs:
epoch = 0
images = {
'input_img': input_img,
'teacher_output_img': teacher_output_img,
'student_output_img': student_output_img
}
for key in images:
img_np = images[key].detach().cpu().numpy()
img_np = np.moveaxis(img_np, 1, -1)
img_np = (img_np + 1) / 2 # (-1,1) -> (0,1)
img_big = fourD2threeD(img_np, n_row=1)
print(key, img_big.shape, np.amax(img_big), np.amin(img_big))
imsave(os.path.join(img_dir, 'epoch%d_%s.png' % (epoch, key)),
img_as_ubyte(img_big))
###### Training ######
print('dataloader:', len(dataloader)) # 1334
for epoch in range(start_epoch, args.epochs):
epoch += 1
if args.lq and epoch == args.start_lq:
netG.apply(learn_quant.enable_param_learning)
netG.cuda()
start_time = time.time()
netG.train(), netD.train()
# define average meters:
loss_G_meter, loss_G_perceptual_meter, loss_G_GAN_meter, loss_D_meter = \
def show3dlidar(pointpaht, detections, calib):
pointcloud = np.fromfile(pointpaht, dtype=np.float32).reshape(-1, 4)
x = pointcloud[:, 0] # x position of point
xmin = np.amin(x, axis=0)
xmax = np.amax(x, axis=0)
y = pointcloud[:, 1] # y position of point
ymin = np.amin(y, axis=0)
ymax = np.amax(y, axis=0)
z = pointcloud[:, 2] # z position of point
zmin = np.amin(z, axis=0)
zmax = np.amax(z, axis=0)
d = np.sqrt(x**2 + y**2) # Map Distance from sensor
vals = 'height'
if vals == "height":
col = z
else:
col = d
fig = mayavi.mlab.figure(bgcolor=(0, 0, 0), size=(640, 500))
mayavi.mlab.points3d(
x,
y,
z,
col, # Values used for Color
mode="point",
colormap='Blues', # 'bone', 'copper', 'gnuplot'
# color=(0, 1, 0), # Used a fixed (r,g,b) instead
figure=fig,
)
mayavi.mlab.points3d(0,
0,
0,
color=(1, 1, 1),
mode="sphere",
scale_factor=0.2)
print(detections.shape)
detections[:, 1:] = lidar_to_camera_box(detections[:, 1:], calib.V2C,
calib.R0, calib.P2)
for i in range(detections.shape[0]):
h = float(detections[i][4])
w = float(detections[i][5])
l = float(detections[i][6])
x = float(detections[i][1])
y = float(detections[i][2])
z = float(detections[i][3])
x_corners = [
l / 2, l / 2, -l / 2, -l / 2, l / 2, l / 2, -l / 2, -l / 2
]
y_corners = [0, 0, 0, 0, -h, -h, -h, -h]
z_corners = [
w / 2, -w / 2, -w / 2, w / 2, w / 2, -w / 2, -w / 2, w / 2
]
#print(x_corners)
#print(detections[i])
R = roty(float(detections[i][7]))
corners_3d = np.dot(R, np.vstack([x_corners, y_corners, z_corners]))
# print corners_3d.shape
#corners_3d = np.zeros((3,8))
corners_3d[0, :] = corners_3d[0, :] + x
corners_3d[1, :] = corners_3d[1, :] + y
corners_3d[2, :] = corners_3d[2, :] + z
corners_3d = np.transpose(corners_3d)
box3d_pts_3d_velo = project_rect_to_velo(corners_3d)
#x1, y1, z1 = box3d_pts_3d_velo[0, :]
#x2, y2, z2 = box3d_pts_3d_velo[1, :]
if detections[i][0] == 1.0:
draw_gt_boxes3d([box3d_pts_3d_velo], 1, color=(1, 0, 0), fig=fig)
else:
draw_gt_boxes3d([box3d_pts_3d_velo], 1, color=(0, 1, 0), fig=fig)
mayavi.mlab.show()
def onehot(y):
y_onehot = np.array(y)
oh = np.zeros((len(y_onehot) ,np.amax(y_onehot)+1))
oh[np.arange(y_onehot.size),y_onehot]=1
print(oh)
return oh
import numpy
from matplotlib.colors import LogNorm
from velociraptor.tools.lines import binned_median_line as bml
import unyt
pyplot.rcParams.update({'font.size': 40})
import adi_gas_spread_density
import adi_gas_coordinate_density
import gas_spread_density
import gas_coordinate_density
import simba_gas_spread_density
import simba_gas_coordinate_density
adi_spread = adi_gas_spread_density.spread
adi_dens = adi_gas_spread_density.density
adi_xbins = numpy.logspace(numpy.log10(numpy.amin(adi_dens)),
numpy.log10(numpy.amax(adi_dens)),
num=50)
adi_ybins = numpy.logspace(numpy.log10(numpy.amin(adi_spread)),
numpy.log10(numpy.amax(adi_spread)),
num=50)
adi_bins = unyt.unyt_array(numpy.logspace(numpy.log10(numpy.amin(adi_dens)),
numpy.log10(numpy.amax(adi_dens)),
num=20),
units=adi_dens.units)
adi_x = adi_gas_coordinate_density.x
adi_y = adi_gas_coordinate_density.y
adic_xbins = numpy.linspace(numpy.amin(adi_x), numpy.amax(adi_x), num=26)
adic_ybins = numpy.linspace(numpy.amin(adi_y), numpy.amax(adi_y), num=26)
adi_centers, adi_med, adi_err = bml(adi_dens, adi_spread, x_bins=adi_bins)
eagle_spread = gas_spread_density.spread
ht = cv2.resize(ht, (0, 0),
fx=float(heatmap_size) / img_size,
fy=float(heatmap_size) / img_size,
interpolation=cv2.INTER_LANCZOS4)
output_heatmaps[:, :, i] = ht
cropping_param = [
int(float(cur_hand_bbox[0])),
int(float(cur_hand_bbox[1])),
int(float(cur_hand_bbox[2])),
int(float(cur_hand_bbox[3])), offset_x, offset_y, img_scale
]
# Create background map
output_background_map = np.ones(
(heatmap_size, heatmap_size)) - np.amax(output_heatmaps, axis=2)
output_heatmaps = np.concatenate((output_heatmaps,
output_background_map.reshape(
(heatmap_size, heatmap_size, 1))),
axis=2)
# coords_set = np.concatenate(
# (np.reshape(cur_hand_joints_x, (num_of_joints, 1)), np.reshape(cur_hand_joints_y, (num_of_joints, 1))), axis=1)
output_image_raw = output_image.astype(np.uint8).tostring()
output_heatmaps_raw = output_heatmaps.flatten().tolist()
# output_coords_raw = coords_set.flatten().tolist()
output_cropping_param_raw = cropping_param
output_R_raw = camera_ref.R.flatten().tolist()
output_C_raw = camera_ref.C.flatten().tolist()
output_K_raw = camera_ref.K.flatten().tolist()
#i_factor = 50 / np.mean(samples[:, 2]) #z_factor = 50 / np.mean(samples[:, 3]) #samples[:, 0] *= np.repeat([g_factor], samples.shape[0], axis = 0) #samples[:, 1] *= np.repeat([r_factor], samples.shape[0], axis = 0) #samples[:, 2] *= np.repeat([i_factor], samples.shape[0], axis = 0) #samples[:, 3] *= np.repeat([z_factor], samples.shape[0], axis = 0) #print(samples[:,0]) #g_diff = np.amax(samples[:,0]) - np.amin(samples[:,0]) #r_diff = np.amax(samples[:,1]) - np.amin(samples[:,1]) #i_diff = np.amax(samples[:,2]) - np.amin(samples[:,2]) #z_diff = np.amax(samples[:,3]) - np.amin(samples[:,3]) g_min = np.amin(samples[:,0]) r_min = np.amin(samples[:,1]) i_min = np.amin(samples[:,2]) z_min = np.amin(samples[:,3]) g_max = np.amax(samples[:,0]) r_max = np.amax(samples[:,1]) i_max = np.amax(samples[:,2]) z_max = np.amax(samples[:,3]) g_min_u = np.amin(old_samples[:,0]) r_min_u = np.amin(old_samples[:,1]) i_min_u = np.amin(old_samples[:,2]) z_min_u = np.amin(old_samples[:,3]) g_max_u = np.amax(old_samples[:,0]) r_max_u = np.amax(old_samples[:,1]) i_max_u = np.amax(old_samples[:,2]) z_max_u = np.amax(old_samples[:,3]) #print(g_min,r_min,i_min,z_min,g_max,r_max,i_max,z_max) #g_min = 0.0
def __start_node(self):
""" Start a node
This functions will start up a node, based on the sliding midpoint rule.
Returns
-------
KDTree
self
"""
# Checking if this could be a leaf node
sum = 0
if self.left_indexes is not None:
sum += len(self.left_indexes)
if self.right_indexes is not None:
sum += len(self.right_indexes)
if sum <= self.leaf_size:
self.is_leaf = True
self.left_subtree = None
self.right_subtree = None
if self.left_indexes is None and self.right_indexes is not None:
self.leaf_indexes = self.right_indexes
elif self.left_indexes is not None and self.right_indexes is None:
self.leaf_indexes = self.left_indexes
else:
self.leaf_indexes = list(set().union(self.left_indexes,
self.right_indexes))
else:
# Handling left subtree
if self.left_indexes is not None:
if len(self.left_indexes) > 0:
aux_X = np.asarray(
[self.data[index] for index in self.left_indexes])
maxes = np.amax(aux_X, axis=0)
mins = np.amin(aux_X, axis=0)
# Getting longest side
d = np.argmax(maxes - mins)
maxval = maxes[d]
minval = mins[d]
data = aux_X[:, d]
# Get the split point
split = (maxval + minval) / 2
# Split the indexes between left and right child
left = np.nonzero(data < split)[0]
left = np.asarray([self.left_indexes[k] for k in left])
right = np.nonzero(data >= split)[0]
right = np.asarray([self.left_indexes[k] for k in right])
# If there's a child with no indexes, while the other has more than
# leaf_size indexes, we slide the cutting point towards the 'fat'
# size until there's at least one index on each child.
if (len(right) == 0) or (len(left) == 0):
if (len(right) == 0) and (len(left) > self.leaf_size):
split = np.amax(data[data != np.amax(data)])
left = np.nonzero(data < split)[0]
left = np.asarray(
[self.left_indexes[k] for k in left])
right = np.nonzero(data >= split)[0]
right = np.asarray(
[self.left_indexes[k] for k in right])
elif (len(right)
== 0) and (len(left) <= self.leaf_size):
right = None
elif (len(left)
== 0) and (len(right) > self.leaf_size):
split = np.amin(data[data != np.amin(data)])
left = np.nonzero(data < split)[0]
left = np.asarray(
[self.left_indexes[k] for k in left])
right = np.nonzero(data >= split)[0]
right = np.asarray(
[self.left_indexes[k] for k in right])
elif (len(left)
== 0) and (len(right) <= self.leaf_size):
left = None
# Creates the left subtree
self.left_subtree = KDTreeNode(
data=self.data,
left_indexes=left,
right_indexes=right,
split_axis=d,
split_value=split,
distance_function=self.distance_function,
leaf_size=self.leaf_size,
**self.kwargs)
else:
self.left_subtree = None
else:
self.left_subtree = None
# Handling right subtree
if self.right_indexes is not None:
if len(self.right_indexes) > 0:
aux_X = np.asarray(
[self.data[index] for index in self.right_indexes])
maxes = np.amax(aux_X, axis=0)
mins = np.amin(aux_X, axis=0)
# Getting longest side
d = np.argmax(maxes - mins)
maxval = maxes[d]
minval = mins[d]
data = aux_X[:, d]
# Get the split point
split = (maxval + minval) / 2
# Split the indexes between left and right child
left = np.nonzero(data < split)[0]
left = np.asarray([self.right_indexes[k] for k in left])
right = np.nonzero(data >= split)[0]
right = np.asarray([self.right_indexes[k] for k in right])
# If there's a child with no indexes, while the other has more than
# leaf_size indexes, we slide the cutting point towards the 'fat'
# size until there's at least one index on each child.
if (len(right) == 0) or (len(left) == 0):
if (len(right) == 0) and (len(left) > self.leaf_size):
split = np.amax(data[data != np.amax(data)])
left = np.nonzero(data < split)[0]
left = np.asarray(
[self.right_indexes[k] for k in left])
right = np.nonzero(data >= split)[0]
right = np.asarray(
[self.right_indexes[k] for k in right])
elif (len(right)
== 0) and (len(left) <= self.leaf_size):
right = None
elif (len(left)
== 0) and (len(right) > self.leaf_size):
split = np.amin(data[data != np.amin(data)])
left = np.nonzero(data < split)[0]
left = np.asarray(
[self.right_indexes[k] for k in left])
right = np.nonzero(data >= split)[0]
right = np.asarray(
[self.right_indexes[k] for k in right])
elif (len(left)
== 0) and (len(right) <= self.leaf_size):
left = None
# Creates the right subtree
self.right_subtree = KDTreeNode(
data=self.data,
left_indexes=left,
right_indexes=right,
split_axis=d,
split_value=split,
distance_function=self.distance_function,
leaf_size=self.leaf_size,
**self.kwargs)
else:
self.right_subtree = None
else:
self.right_subtree = None
return self
E = [(0, 1, 1.0), (0, 2, 1.0), (1, 2, 1.0), (3, 2, 1.0), (3, 1, 1.0)]
G = nx.Graph()
G.add_nodes_from(V)
G.add_weighted_edges_from(E)
step_size = 0.1
a_gamma = np.arange(0, np.pi, step_size)
a_beta = np.arange(0, np.pi, step_size)
a_gamma, a_beta = np.meshgrid(a_gamma, a_beta)
F1 = 3 - (np.sin(2 * a_beta)**2 * np.sin(2 * a_gamma)**2 - 0.5 * np.sin(
4 * a_beta) * np.sin(4 * a_gamma)) * (1 + np.cos(4 * a_gamma)**2)
result = np.where(F1 == np.amax(F1))
a = list(zip(result[0], result[1]))[0]
gamma = a[0] * step_size
beta = a[1] * step_size
prog = make_circuit(4)
sample_shot = 5600
writefile = open("../data/startQiskit_QC484.csv", "w")
# prog.draw('mpl', filename=(kernel + '.png'))
IBMQ.load_account()
provider = IBMQ.get_provider(hub='ibm-q')
provider.backends()
backend = provider.get_backend("ibmq_5_yorktown")
circuit1 = transpile(prog, FakeYorktown())
def evaluate_error(gt, pred, criteria):
"""
Calculates various types of error for between ground truth and prediction. The various errors are intended to
showcase different strengths and weaknesses in the predictions.
@param gt: Ground truth depth
@param pred: Predicted depth
@param criteria: dict with criteria as the keys
@return: updated criteria dict
"""
zero_mask = gt > 0
gt = gt[zero_mask]
criteria['n_pixels'] += gt.shape[0]
pred = pred[zero_mask]
gt_rescaled = gt * 80.
pred_rescaled = pred * 80.
# Mean Absolute Relative Error
rel = np.abs(gt - pred) / gt # compute errors
abs_rel_sum = np.sum(rel)
criteria['err_absRel'] += abs_rel_sum
# Square Mean Relative Error
s_rel = ((gt_rescaled - pred_rescaled) * (gt_rescaled - pred_rescaled)) / (
gt_rescaled * gt_rescaled) # compute errors
squa_rel_sum = np.sum(s_rel)
criteria['err_squaRel'] += squa_rel_sum
# Root Mean Square error
square = (gt_rescaled - pred_rescaled)**2
rms_squa_sum = np.sum(square)
criteria['err_rms'] += rms_squa_sum
# Log Root Mean Square error
log_square = (np.log(gt_rescaled) - np.log(pred_rescaled))**2
log_rms_sum = np.sum(log_square)
criteria['err_logRms'] += log_rms_sum
# Scale invariant error
diff_log = np.log(pred_rescaled) - np.log(gt_rescaled)
diff_log_sum = np.sum(diff_log)
criteria['err_silog'] += diff_log_sum
diff_log_2 = diff_log**2
diff_log_2_sum = np.sum(diff_log_2)
criteria['err_silog2'] += diff_log_2_sum
# Mean log10 error
log10_sum = np.sum(np.abs(np.log10(gt) - np.log10(pred)))
criteria['err_log10'] += log10_sum
# Deltas
gt_pred = gt_rescaled / pred_rescaled
pred_gt = pred_rescaled / gt_rescaled
gt_pred = np.reshape(gt_pred, (1, -1))
pred_gt = np.reshape(pred_gt, (1, -1))
gt_pred_gt = np.concatenate((gt_pred, pred_gt), axis=0)
ratio_max = np.amax(gt_pred_gt, axis=0)
delta_1_sum = np.sum(ratio_max < 1.25)
criteria['err_delta1'] += delta_1_sum
delta_2_sum = np.sum(ratio_max < 1.25**2)
criteria['err_delta2'] += delta_2_sum
delta_3_sum = np.sum(ratio_max < 1.25**3)
criteria['err_delta3'] += delta_3_sum
return criteria
def test_pt_tf_model_equivalence(self):
if not is_torch_available():
return
import torch
import transformers
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
pt_model_class_name = model_class.__name__[2:] # Skip the "TF" at the beggining
pt_model_class = getattr(transformers, pt_model_class_name)
config.output_hidden_states = True
tf_model = model_class(config)
pt_model = pt_model_class(config)
# Check we can load pt model in tf and vice-versa with model => model functions
tf_model = transformers.load_pytorch_model_in_tf2_model(tf_model, pt_model, tf_inputs=inputs_dict)
pt_model = transformers.load_tf2_model_in_pytorch_model(pt_model, tf_model)
# Check predictions on first output (logits/hidden-states) are close enought given low-level computational differences
pt_model.eval()
pt_inputs_dict = dict(
(name, torch.from_numpy(key.numpy()).to(torch.long)) for name, key in inputs_dict.items()
)
with torch.no_grad():
pto = pt_model(**pt_inputs_dict)
tfo = tf_model(inputs_dict, training=False)
tf_hidden_states = tfo[0].numpy()
pt_hidden_states = pto[0].numpy()
pt_hidden_states[np.isnan(tf_hidden_states)] = 0
tf_hidden_states[np.isnan(tf_hidden_states)] = 0
pt_hidden_states[np.isnan(pt_hidden_states)] = 0
tf_hidden_states[np.isnan(pt_hidden_states)] = 0
max_diff = np.amax(np.abs(tf_hidden_states - pt_hidden_states))
# Debug info (remove when fixed)
if max_diff >= 2e-2:
print("===")
print(model_class)
print(config)
print(inputs_dict)
print(pt_inputs_dict)
self.assertLessEqual(max_diff, 2e-2)
# Check we can load pt model in tf and vice-versa with checkpoint => model functions
with tempfile.TemporaryDirectory() as tmpdirname:
pt_checkpoint_path = os.path.join(tmpdirname, "pt_model.bin")
torch.save(pt_model.state_dict(), pt_checkpoint_path)
tf_model = transformers.load_pytorch_checkpoint_in_tf2_model(tf_model, pt_checkpoint_path)
tf_checkpoint_path = os.path.join(tmpdirname, "tf_model.h5")
tf_model.save_weights(tf_checkpoint_path)
pt_model = transformers.load_tf2_checkpoint_in_pytorch_model(pt_model, tf_checkpoint_path)
# Check predictions on first output (logits/hidden-states) are close enought given low-level computational differences
pt_model.eval()
pt_inputs_dict = dict(
(name, torch.from_numpy(key.numpy()).to(torch.long)) for name, key in inputs_dict.items()
)
with torch.no_grad():
pto = pt_model(**pt_inputs_dict)
tfo = tf_model(inputs_dict)
tfo = tfo[0].numpy()
pto = pto[0].numpy()
tfo[np.isnan(tfo)] = 0
pto[np.isnan(pto)] = 0
max_diff = np.amax(np.abs(tfo - pto))
self.assertLessEqual(max_diff, 2e-2)
def test_3_fct_tot(data):
sut = data[2]
rms = 0.00012049 # TODO
np.testing.assert_approx_equal(np.amin(sut), 1)
np.testing.assert_approx_equal(np.amax(sut), 4.26672894, significant=2)
a = np.sqrt(WP**2 * (8 * k**2 * v0**2 + WP**2))
omega = np.lib.scimath.sqrt((2 * k**2 * v0**2 + WP**2 - a) / 2)
om_im = omega.imag
f = Figure(figsize=(5, 4), dpi=100)
plt.plot(k, om_im, linewidth=2)
plt.xlabel('k')
plt.ylabel('omega')
plt.show()
gamma_max = np.amax(om_im)
print 'Maximum growth rate: ', gamma_max
idx = np.argmax(om_im)
k_max = k[idx]
print 'k for maximizing growth rate: ', k_max
L = 3 * 2 * np.pi / (np.sqrt(3.0 / 2) / 2.)
mode = 3
k_sim = 2 * np.pi / L * mode
a_sim = np.sqrt(WP**2 * (8 * k_sim**2 * v0**2 + WP**2))
omega_sim = np.lib.scimath.sqrt((2 * k_sim**2 * v0**2 + WP**2 - a_sim) / 2)
print 'Growth rate: ', omega_sim
raw_input('Press enter...')
def test_upwind(data):
sut = data[0]
assert np.amin(sut) == 1
np.testing.assert_approx_equal(np.amax(sut), 4.796, significant=4)
def test_2_fct_iga(data):
sut = data[3]
rms = 0.00026658 # TODO
np.testing.assert_approx_equal(np.amin(sut), 1)
np.testing.assert_approx_equal(np.amax(sut), 4.25518091, significant=2)
def sweep(duration, dt, f,
autocorrelate=True,
return_t=False,
taper='blackman',
**kwargs):
"""
Generates a linear frequency modulated wavelet (sweep). Wraps
scipy.signal.chirp, adding dimensions as necessary.
Args:
duration (float): The length in seconds of the wavelet.
dt (float): is the sample interval in seconds (usually 0.001, 0.002,
or 0.004)
f (ndarray): Any sequence like (f1, f2). A list of lists will create a
wavelet bank.
autocorrelate (bool): Whether to autocorrelate the sweep(s) to create
a wavelet. Default is `True`.
return_t (bool): If True, then the function returns a tuple of
wavelet, time-basis, where time is the range from -duration/2 to
duration/2 in steps of dt.
taper (str or function): The window or tapering function to apply.
To use one of NumPy's functions, pass 'bartlett', 'blackman' (the
default), 'hamming', or 'hanning'; to apply no tapering, pass
'none'. To apply your own function, pass a function taking only
the length of the window and returning the window function.
**kwargs: Further arguments are passed to scipy.signal.chirp. They are
`method` ('linear','quadratic','logarithmic'), `phi` (phase offset
in degrees), and `vertex_zero`.
Returns:
ndarray: The waveform.
"""
t0, t1 = -duration/2, duration/2
t = np.arange(t0, t1, dt)
f = np.asanyarray(f).reshape(-1, 1)
f1, f2 = f
c = [scipy.signal.chirp(t, f1_+(f2_-f1_)/2., t1, f2_, **kwargs)
for f1_, f2_
in zip(f1, f2)]
if autocorrelate:
w = [np.correlate(c_, c_, mode='same') for c_ in c]
w = np.squeeze(w) / np.amax(w)
if taper:
funcs = {
'bartlett': np.bartlett,
'blackman': np.blackman,
'hamming': np.hamming,
'hanning': np.hanning,
'none': lambda x: x,
}
func = funcs.get(taper, taper)
w *= func(t.size)
if return_t:
Sweep = namedtuple('Sweep', ['amplitude', 'time'])
return Sweep(w, t)
else:
return w
def test_2_fct(data):
sut = data[1]
rms = 0.00036731 # TODO
np.testing.assert_approx_equal(np.amin(sut), 1)
np.testing.assert_approx_equal(np.amax(sut), 3.52544410, significant=2)
metric = [param.metric]
txt_col = [METRICS[param.metric]['col']]
results = [np.loadtxt(f, usecols=txt_col) for f in result_files]
# Reduce length of all folds to the minimum length
epochs = [len(r) for r in results]
e_min = min(epochs)
results = [r[:e_min] for r in results]
# res_array.shape = (k , e_min, len(txt_col))
res_array = np.array(results)
# to obtain one measurement for all folds: p = arr[:,:,index_p_in_txt_col]
means = np.mean(res_array, axis=0)
mins = np.amin(res_array, axis=0)
maxs = np.amax(res_array, axis=0)
if len(metric) > 1:
n_cols = 2
n_rows = - (-len(metric) // 2)
fig1, axes = plt.subplots(nrows=n_rows, ncols=n_cols, sharex='all') # squeeze=False??
epochs = np.arange(e_min)
for i in range(n_rows):
for j in range(n_cols):
axes[i, j].plot(epochs, means[:, i * n_cols + j]) # [:, i * n_cols + j]
axes[i, j].fill_between(epochs, mins[:, i * n_cols + j], maxs[:, i * n_cols + j], alpha=.5)
else:
epochs = np.arange(e_min)
fig1 = plt.plot(epochs, means)
plt.fill_between(epochs, mins, maxs, alpha=.5)
def plot_fem_solution(self, kx=0.):
if self.plot[5]: # Determination of the maximum value of the pressure
p_max = 0
p_min = 1e308
for _en in self.entities:
if isinstance(_en, FluidFem):
for _elem in _en.elements:
_, __, p_elem = _elem.display_sol(3)
_max = np.amax(np.abs(p_elem))
_min = np.amin(np.abs(p_elem))
if _max >p_max: p_max = _max
if _min <p_min: p_min = _min
if any(self.plot[3:]):
x, y, u_x, u_y, pr = [], [], [], [], []
for _en in self.entities:
if isinstance(_en, FluidFem):
if any(self.plot[2::3]): # Plot of pressure == True
for ie, _elem in enumerate(_en.elements):
# print(ie/len(_en.elements))
x_elem, y_elem, p_elem = _elem.display_sol(3)
p_elem = p_elem[:, 0]
p_elem *= np.exp(1j*kx*x_elem)
if self.plot[2]:
plt.figure("Pressure")
plt.plot(y_elem, np.abs(p_elem), 'r+')
plt.plot(y_elem, np.imag(p_elem), 'm.')
if self.plot[5]:
triang = mtri.Triangulation(x_elem, y_elem)
plt.figure("Pressure map")
plt.tricontourf(triang, np.abs(p_elem), cmap=cm.jet, levels=np.linspace(p_min, p_max,40))
# x.extend(list(x_elem))
# y.extend(list(y_elem))
# pr.extend(list(p_elem))
elif isinstance(_en, PemFem):
if any(self.plot): # Plot of pressure == True
for _elem in _en.elements:
x_elem, y_elem, f_elem = _elem.display_sol([0, 1, 3])
ux_elem = f_elem[:, 0]*np.exp(1j*kx*x_elem)
uy_elem = f_elem[:, 1]*np.exp(1j*kx*x_elem)
p_elem = f_elem[:, 2]*np.exp(1j*kx*x_elem)
if self.plot[0]:
plt.figure("Solid displacement along x")
plt.plot(y_elem, np.abs(ux_elem), 'r+')
plt.plot(y_elem, np.imag(ux_elem), 'm.')
if self.plot[1]:
plt.figure("Solid displacement along y")
plt.plot(y_elem, np.abs(uy_elem), 'r+')
plt.plot(y_elem, np.imag(uy_elem), 'm.')
if self.plot[2]:
plt.figure("Pressure")
plt.plot(y_elem, np.abs(p_elem), 'r+')
plt.plot(y_elem, np.imag(p_elem), 'm.')
if self.plot[5]:
x.extend(list(x_elem))
y.extend(list(y_elem))
pr.extend(list(p_elem))
elif isinstance(_en, ElasticFem):
if any(self.plot): # Plot of pressure == True
for _elem in _en.elements:
x_elem, y_elem, f_elem = _elem.display_sol([0, 1, 3])
ux_elem = f_elem[:, 0]*np.exp(1j*kx*x_elem)
uy_elem = f_elem[:, 1]*np.exp(1j*kx*x_elem)
if self.plot[0]:
plt.figure("Solid displacement along x")
plt.plot(y_elem, np.abs(ux_elem), 'r+')
plt.plot(y_elem, np.imag(ux_elem), 'm.')
if self.plot[1]:
plt.figure("Solid displacement along y")
plt.plot(y_elem, np.abs(uy_elem), 'r+')
plt.plot(y_elem, np.imag(uy_elem), 'm.')
if any(self.plot[3:]):
# triang = mtri.Triangulation(x, y)
if self.plot[5]:
plt.figure("Pressure map")
# plt.tricontourf(triang, np.abs(pr), 40, cmap=cm.jet)
# self.display_mesh()
# plt.colorbar()
plt.axis("off")
plt.axis('equal')
def extract_particles(self, segmentation):
"""
Saves particle centers into output .star file, after dismissing regions
that are too big to contain a particle.
Args:
segmentation: Segmentation of the micrograph into noise and particle projections.
"""
segmentation = segmentation[self.query_size // 2 - 1:-self.query_size // 2,
self.query_size // 2 - 1:-self.query_size // 2]
labeled_segments, _ = ndimage.label(segmentation, np.ones((3, 3)))
values, repeats = np.unique(labeled_segments, return_counts=True)
values_to_remove = np.where(repeats > self.max_size ** 2)
values = np.take(values, values_to_remove)
values = np.reshape(values, (1, 1, np.prod(values.shape)), 'F')
labeled_segments = np.reshape(labeled_segments, (labeled_segments.shape[0],
labeled_segments.shape[1], 1), 'F')
matrix1 = np.repeat(labeled_segments, values.shape[2], 2)
matrix2 = np.repeat(values, matrix1.shape[0], 0)
matrix2 = np.repeat(matrix2, matrix1.shape[1], 1)
matrix3 = np.equal(matrix1, matrix2)
matrix4 = np.sum(matrix3, 2)
segmentation[np.where(matrix4 == 1)] = 0
labeled_segments, _ = ndimage.label(segmentation, np.ones((3, 3)))
max_val = np.amax(np.reshape(labeled_segments, (np.prod(labeled_segments.shape))))
center = center_of_mass(segmentation, labeled_segments, np.arange(1, max_val))
center = np.rint(center)
img = np.zeros((segmentation.shape[0], segmentation.shape[1]))
img[center[:, 0].astype(int), center[:, 1].astype(int)] = 1
y, x = np.ogrid[-self.moa:self.moa+1, -self.moa:self.moa+1]
element = x*x+y*y <= self.moa * self.moa
img = binary_dilation(img, structure=element)
labeled_img, _ = ndimage.label(img, np.ones((3, 3)))
values, repeats = np.unique(labeled_img, return_counts=True)
y = np.where(repeats == np.count_nonzero(element))
y = np.array(y)
y = y.astype(int)
y = np.reshape(y, (np.prod(y.shape)), 'F')
y -= 1
center = center[y, :]
center = center + (self.query_size // 2 - 1) * np.ones(center.shape)
center = center + (self.query_size // 2 - 1) * np.ones(center.shape)
center = center + np.ones(center.shape)
center = config.apple.mrc_shrink_factor * center
# swap columns to align with Relion
center = center[:, [1, 0]]
# first column is x; second column is y - offset by margins that were discarded from the image
center[:, 0] += config.apple.mrc_margin_left
center[:, 1] += config.apple.mrc_margin_top
if self.output_directory is not None:
basename = os.path.basename(self.filename)
name_str, ext = os.path.splitext(basename)
applepick_path = os.path.join(self.output_directory, "{}_applepick.star".format(name_str))
with open(applepick_path, "w") as f:
np.savetxt(f, ["data_root\n\nloop_\n_rlnCoordinateX #1\n_rlnCoordinateY #2"], fmt='%s')
np.savetxt(f, center, fmt='%d %d')
return center
cols = data.shape[1] X = data.iloc[:,0:cols-1] y = data.iloc[:,cols-1:cols] # convert from data frames to numpy matrices X = np.array(X.values) y = np.array(y.values) y = y.flatten() X_tf = tf.constant(X) Y_tf = tf.constant(y) #Train a Linear Classifier # initialize parameters randomly D = X.shape[1] K = np.amax(y) + 1 # initialize parameters in such a way to play nicely with the gradient-check! #W = 0.01 * np.random.randn(D,K) #b = np.zeros((1,K)) + 1.0 initializer = tf.random_normal_initializer(mean=0.0, stddev=0.01, seed=21131186) #This might differ based on tf.version. # Play with stddev and see how your model performance changes. # Change initialization to Xavier and orthogonal and analysis change in accuracy. #If using other init compare that with Guassian init and report your findings W = tf.Variable(initializer([D, K])) W = tf.cast(W,dtype = tf.double) b = tf.Variable(tf.random.normal([K])) # You can also try tf.zeros and tf.ones, report your findings. theta = (W,b) # some hyperparameters
def check_image(image):
assert isinstance(image, np.ndarray)
assert image.shape == (512, 512)
assert np.amin(image) >= 0 - DELTA and np.amax(image) <= 1 + DELTA
def normalize_coordinates(list_residues_coord):
xyz_max = np.amax(list_residues_coord, axis=0)[0:3]
list_residues_coord[:, 0:3] /= xyz_max
return list_residues_coord
def main():
# Training settings
parser = argparse.ArgumentParser(
description='prediction from cognitive memory')
parser.add_argument('--no-cuda',
action='store_true',
default=False,
help='disables CUDA training')
parser.add_argument('--seed',
type=int,
default=1,
metavar='S',
help='random seed (default: 1)')
parser.add_argument(
'--layer',
type=int,
default=0,
metavar='N',
help=
'select a layer 0-4, which represents the first CNN, 3 composite layers and the final FC'
)
# parser.add_argument('--threshold', type=int, default=0, metavar='N',
# help='threshold values used when cogmemry is generated')
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
device = torch.device("cuda" if use_cuda else "cpu")
kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}
test_loader = torch.utils.data.DataLoader(datasets.CIFAR10(
root='./data',
train=False,
transform=transforms.Compose([transforms.ToTensor(), normalize])),
batch_size=10000,
shuffle=False,
**kwargs)
coglayer = args.layer
#threshold_={0:0.3,1:0.75,2:0.6,3:0.57, 4:0.88}
threshold_0 = [0.18, 0.2, 0.22, 0.24, 0.26, 0.3, 0.32, 0.34]
threshold_1 = [0.64, 0.66, 0.68, 0.70, 0.72, 0.74, 0.76, 0.78]
threshold_2 = [0.48, 0.5, 0.52, 0.54, 0.56, 0.58, 0.6, 0.62]
threshold_3 = [0.5, 0.52, 0.54, 0.56, 0.58, 0.6, 0.62, 0.64]
threshold_4 = [0.8, 0.82, 0.84, 0.86, 0.88, 0.9, 0.92, 0.94]
if coglayer == 0:
thresholds_ = threshold_0
elif coglayer == 1:
thresholds_ = threshold_1
elif coglayer == 2:
thresholds_ = threshold_2
elif coglayer == 3:
thresholds_ = threshold_3
elif coglayer == 4:
thresholds_ = threshold_4
pred_n = torch.load('test_prediction_resnet.pt',
map_location=lambda storage, loc: storage)
print(pred_n.size())
for data, target in test_loader:
labels = target
labels_ = []
for xi in labels:
temp_ = np.zeros(10)
temp_[xi.item()] = 1.0
labels_.append(temp_)
labels_ = np.array(labels_)
print('label.shape', labels_.shape)
if os.path.exists("confusion_sel"):
pass
else:
os.mkdir("confusion_sel")
results = {}
for th in thresholds_:
layer_sel_ = coglayer
temp_dict = {}
roV = load_test_image(layer_sel_)
roV = roV.to(device)
act_map = torch.load('coglayer/map_association_' + str(coglayer) +
'_' + str(th) + '.pt',
map_location=lambda storage, loc: storage)
sel = act_map.map
sel = sel.cpu().numpy()
wm = torch.load('coglayer/wm_' + str(layer_sel_) + '_' + str(th) +
'.pt',
map_location=lambda storage, loc: storage)
wm = wm.to(device)
cog = CogMem_load(wm)
#cog=CogMem_load(wm,label)
cog.forward(roV)
#pred=cog.pred.long()
#pred=cog.pred.long().cpu().numpy()
total_1 = 0
total_2 = 0
total_3 = 0
total_4 = 0
total_5 = 0
cons1 = 0
cons2 = 0
temp = 0
corr = np.zeros((10, 10))
cnt = 0
#mem=[]
#print ('sel shape',sel.shape)
#print (cog.image.size())
for xi, xin in enumerate(pred_n):
cls = xin.item()
label_t = labels[xi].long().item()
v2 = cog.image[:, xi]
#mem.append(v2.cpu().numpy())
idx = torch.argsort(v2).cpu().numpy()
idx = np.flip(idx, 0)[:8]
temp_v = np.zeros(10)
for zin in idx:
temp_v = temp_v + sel[zin, :] * v2[zin].item()
idx2 = np.argmax(temp_v)
idx3 = np.argsort(temp_v)
idx3 = np.flip(idx3, 0)[:3]
sum_v = np.sum(np.exp(temp_v))
#print (xi, idx, cls, idx3, idx2)
# cls: prediction, idx2: max from association, idx3, label from truth, idx_truth: ground truth
if cls == idx2:
total_1 = total_1 + 1
if label_t == cls:
total_2 = total_2 + 1
if label_t == idx2:
total_3 = total_3 + 1
if label_t != cls:
temp = temp + 1
if cls == idx2:
total_4 = total_4 + 1
else:
temp = temp + 1
if cls == idx2:
total_5 = total_5 + 1
if cls in idx3:
cons1 = cons1 + 1
if label_t in idx3:
cons2 = cons2 + 1
if idx2 == label_t:
c1 = idx2
cnt = cnt + 1
for c2 in range(10):
#if c1!=c2:
corr[c1, c2] = corr[c1, c2] + np.exp(temp_v[c2]) / np.exp(
temp_v[c1])
print(cnt)
for c1 in range(10):
corr[c1, :] = corr[c1, :] / float(cnt)
for c1 in range(10):
corr[c1, c1] = 0
max_v = np.amax(corr)
#corr=corr/500.0
print(layer_sel_, total_1, total_2, total_3)
print('cons1', cons1, 'cons2', cons2)
#pylab.figure(ttin+1)
pylab.imshow(corr, cmap='jet', vmax=max_v)
pylab.colorbar()
pylab.savefig('confusion_sel/L' + str(layer_sel_) + '_' + str(th) +
'.png')
pylab.savefig('confusion_sel/L' + str(layer_sel_) + '_' + str(th) +
'.eps')
pylab.close()
temp_dict = {
'max_pred': total_1,
'ref_accuracy': total_2,
'max_accuracy': total_3,
'consist_pred': cons1,
'consist_accuracy': cons2,
'cog_size': wm.size()
}
#mem=np.array(mem)
#np.savetxt('mem_'+str(layer_sel_)+'.txt',mem)
torch.cuda.empty_cache()
del cog, roV, sel, wm, act_map
results[str(th)] = temp_dict
fp = open('confusion_sel/prediction' + str(layer_sel_) + '.json', 'w')
json.dump(results, fp)
fp.close
list_key.extend(re.findall(r'\d+',line))
mean_array_agree = np.array(list_agreement).astype(np.float)
mean_agree = np.mean(mean_array_agree)
print ("Time generating key agreements:")
print(np.sum(mean_array_agree))
print ("Mean generating key agreement:")
print (mean_agree)
print ("Standard deviation of key agreement:")
print (np.std(mean_array_agree))
print ("Fastest time generating key agreement:")
print (np.amin(mean_array_agree))
print ("Slowest time generating key agreement:")
print (np.amax(mean_array_agree))
print ("")
mean_array_key = np.array(list_key).astype(np.float)
mean_key = np.mean(mean_array_key)
print ("Time setting key:")
print(np.sum(mean_array_key))
print ("Mean setting key:")
print (mean_key)
print ("Standard deviation of setting key:")
print (np.std(mean_array_key))
