def create_N(_n,_xmax,_ymax): _nodes = [] for i in range(_n): _tmpx = pylab.rand()*_xmax _tmpy = pylab.rand()*_ymax _nodes.append((_tmpx, _tmpy)) return _nodes
def __init__(self, phase_potrait, network, info=None, position=None):
self.system = phase_potrait
self.network = network
self.CYCLES = 10
self.info = info
self.initial_condition = self.system.load_initial_condition(pl.rand(), pl.rand(), pl.rand())
self.fig = pl.figure('Voltage Traces', figsize=(6, 2), facecolor='#EEEEEE')
self.ax = self.fig.add_subplot(111, frameon=False, yticks=[])
self.li_b, = self.ax.plot([], [], 'b-', lw=2.)
self.li_g, = self.ax.plot([], [], 'g-', lw=2.)
self.li_r, = self.ax.plot([], [], 'r-', lw=2.)
self.li_y, = self.ax.plot([], [], 'y-', lw=2.)
self.ax.set_xlabel(r'time (sec.)', fontsize=20)
self.ax.set_xticklabels(np.arange(0., 1., 0.1), fontsize=15)
self.ax.set_yticklabels(np.arange(0., 1., 0.1), fontsize=15)
self.ax.set_xlim(0., 100.)
self.ax.set_ylim(-0.06-0.18, 0.04)
self.fig.tight_layout()
self.key_func_dict = dict(u=traces.increase_cycles, i=traces.decrease_cycles)
self.fig.canvas.mpl_connect('key_press_event', self.on_key)
self.fig.canvas.mpl_connect('axes_enter_event', self.focus_in)
if not position == None:
try:
self.fig.canvas.manager.window.wm_geometry(position)
except:
pass
def check():
gamma = 0.01
X = matrix(rand(64,10))
S = matrix(rand(64,10))
args = (S, X, gamma)
x0 = rand(4096,)
return check_grad(f_l2_wd, g_l2_wd, x0, *args)
def __init__(self, phase_potrait, info=None, position=None):
win.window.__init__(self, position)
self.system = phase_potrait
self.info = info
self.CYCLES = 8
self.running = False
self.pulsed = 0
self.num_osci = 8
self.state = np.random.randn(self.num_osci*model.N_EQ1)
self.initial_condition = self.system.load_initial_condition(pl.rand(), pl.rand())
self.ax = self.fig.add_subplot(111, frameon=False, yticks=[])
self.li = [self.ax.plot([], [], 'k-', lw=2.)[0] for i in xrange(self.num_osci)]
self.ax.set_xlabel(r'time (sec.)', fontsize=20)
self.ax.set_xticklabels(np.arange(0., 1., 0.1), fontsize=15)
self.ax.set_yticklabels(np.arange(0., 1., 0.1), fontsize=15)
self.ax.set_xlim(0., 100.)
self.ax.set_ylim(-1.5-self.num_osci*2, 1.5)
self.key_func_dict.update(dict(u=traces.increase_cycles, i=traces.decrease_cycles))
self.fig.canvas.mpl_connect('button_press_event', self.on_click)
self.fig.canvas.mpl_connect('axes_enter_event', self.focus_in)
def normal_test_case():
'''
Runs a test case with simulated data from a normal distribution.
'''
obs, fa, dur = [], [], []
for n in range(15):
d, f, o = make_test_data(
5, split=min(plt.rand()*50+120, 170),
intercept=plt.rand()*50 + 225,
slope1=1 + plt.randn()/0.75, slope2=plt.randn()/.75)
obs.append(o+n)
fa.append(f)
dur.append(d)
plt.plot(f, d, 'o', alpha=0.1)
dur, fa, obs = (np.hstack(dur)[:, np.newaxis],
np.hstack(fa)[:, np.newaxis],
np.hstack(obs)[:, np.newaxis])
dur_mean = dur.mean()
dur_std = dur.std()
dur = (dur-dur_mean)/dur_std
m = normal_model(dur, fa, obs)
trace = sample_model(m, 5000)
predict(trace, 5, 2500, {'mean': dur_mean, 'std': dur_std})
plt.figure()
traceplot(trace, 2, 2500)
return dur, fa, obs, (dur_mean, dur_std), trace
def computeTraces(self, initial_condition=None, plotit=True): if initial_condition == None: initial_condition = self.system.load_initial_condition(pl.rand(), pl.rand()) V_i = fh.integrate_three_rk4( initial_condition, self.network.coupling_strength, self.system.dt/float(self.system.stride), self.system.N_output(self.CYCLES), self.system.stride) t = self.system.dt*np.arange(V_i.shape[0]) if plotit: ticks = np.asarray(t[::t.size/10], dtype=int) xscale, yscale = t[-1], 2. for (i, li) in enumerate([self.li_b, self.li_g, self.li_r]): tj, Vj = tl.adjustForPlotting(t, V_i[:, i], ratio=xscale/yscale, threshold=0.05*xscale) li.set_data(tj, Vj-i*2) self.ax.set_xticks(ticks) self.ax.set_xticklabels(ticks) self.ax.set_xlim(t[0], t[-1]) self.fig.canvas.draw() return t, V_i
def computeTraces(self, initial_condition=None, plotit=True): if initial_condition == None: initial_condition = self.system.load_initial_condition(pl.rand(), pl.rand()) V_i = fh.integrate_three_rk4( initial_condition, self.network.coupling_strength, self.system.dt/float(self.system.stride), self.system.N_output(self.CYCLES), self.system.stride) t = self.system.dt*np.arange(V_i.shape[0]) if plotit: ticks = np.asarray(t[::t.size/10], dtype=int) self.li_b.set_data(t, V_i[:, 0]) self.li_g.set_data(t, V_i[:, 1]-2.) self.li_r.set_data(t, V_i[:, 2]-4.) self.ax.set_xticks(ticks) self.ax.set_xticklabels(ticks) self.ax.set_xlim(t[0], t[-1]) self.fig.canvas.draw() return t, V_i
def __init__(self, phase_potrait, network, info=None, position=None):
win.window.__init__(self, position)
self.system = phase_potrait
self.network = network
self.info = info
self.CYCLES = 10
self.initial_condition = self.system.load_initial_condition(pl.rand(), pl.rand())
self.ax = self.fig.add_subplot(111, frameon=False, yticks=[])
self.li_b, = self.ax.plot([], [], 'b-', lw=2.)
self.li_g, = self.ax.plot([], [], 'g-', lw=2.)
self.li_r, = self.ax.plot([], [], 'r-', lw=2.)
self.ax.set_xlabel(r'time (sec.)', fontsize=20)
self.ax.set_xticklabels(np.arange(0., 1., 0.1), fontsize=15)
self.ax.set_yticklabels(np.arange(0., 1., 0.1), fontsize=15)
self.ax.set_xlim(0., 100.)
self.ax.set_ylim(-5.5, 1.5)
#self.fig.tight_layout()
self.key_func_dict = dict(u=traces.increase_cycles, i=traces.decrease_cycles)
self.fig.canvas.mpl_connect('key_press_event', self.on_key)
self.fig.canvas.mpl_connect('axes_enter_event', self.focus_in)
def __init__(self, system, network, info=None, position=None):
win.window.__init__(self, position)
self.system = system
self.network = network
self.info = info
self.CYCLES = 8
self.state = system.load_initial_condition( pl.rand(), pl.rand() )
self.initial_condition = self.system.load_initial_condition(pl.rand(), pl.rand())
self.running = False
self.pulsed = 0
self.ax = self.fig.add_subplot(111, frameon=False, yticks=[])
self.li_b, = self.ax.plot([], [], 'b-', lw=1.)
self.li_g, = self.ax.plot([], [], 'g-', lw=1.)
self.li_r, = self.ax.plot([], [], 'r-', lw=1.)
self.ax.set_xlabel(r'time (sec.)', fontsize=20)
self.ax.set_xticklabels(np.arange(0., 1., 0.1), fontsize=15)
self.ax.set_yticklabels(np.arange(0., 1., 0.1), fontsize=15)
self.ax.set_xlim(0., 100.)
self.ax.set_ylim(-0.06-0.12, 0.04)
self.key_func_dict = dict(u=traces.increase_cycles, i=traces.decrease_cycles)
self.fig.canvas.mpl_connect('button_press_event', self.on_click)
self.fig.canvas.mpl_connect('axes_enter_event', self.focus_in)
def datagen(N):
"""
Produces N pairs of training data and desired output;
each sample of training data contains -1 in its first position,
this corresponds to the interpretation of the threshold as first
element of the weight vector
"""
fun1 = lambda x1,x2: -2*x1**3-x2+.5*x1**2
fun2 = lambda x1,x2: x1**2*x2+2*x1*x2+1
fun3 = lambda x1,x2: .5*x1*x2**2+x2**2-2*x1**2
rarr1 = rand(1,N)
rarr2 = rand(1,N)
teacher = sign(rand(1,N)-.5)
idplus = (teacher<0)
idminus = -idplus
rarr1[idplus] = rarr1[idplus]-1
y1=fun1(rarr1,rarr2)
y2=fun2(rarr1,rarr2)
y3=fun3(rarr1,rarr2)
x=transpose(concatenate((-ones((1,N)),y1,y2)))
return x, teacher[0]
def step(self):
# if not tumbling, pick random number. If less than RUN_P, move RUN_R in direction of orientation. else, start tumbling.
# if tumbling, pick random number. If greater than TUMBLE_P, rotate by TUMBLE_R. else, stop tumbling.
# matplotlib has (0,0) in the upper left - adapt trig accordingly...
if not self.TUMBLE:
p = rand()
if p < self.run_p:
self.xy[0] = (self.xy[0] + RUN_R*np.sin(self.th*np.pi/180.))%self.frame_lim
self.xy[1] = (self.xy[1] - RUN_R*np.cos(self.th*np.pi/180.))%self.frame_lim
else:
self.TUMBLE = True
if self.TUMBLE:
p = rand()
if p > self.tumble_p:
q = rand()
if q > 0.5:
self.th = (self.th + TUMBLE_R)%360
else:
self.th = (self.th - TUMBLE_R)%360
else:
self.TUMBLE = False
def test_add_out_of_center_l(plot=False, color=1):
# use the same data for both experiments:
num_imgs = 200
color_mult = [1, 3][color]
img_res = 15
padding = 4
targ_pix = img_res ** 2 * color_mult
img_pix = (img_res + 2 * padding) ** 2 * color_mult
inputs = rand(num_imgs, img_pix) < 0.1
targets = rand(num_imgs, targ_pix) * 0.4
inputs[1:] *= 0
# targets = g.rand(num_imgs,targ_pix)
ans = [None] * 2
print "a"
for (i, gx) in enumerate([g, gc]):
conv.g = gx # wonderful. This seems to work. At least.
conv._cu = gx._cudamat
a = gx.garray(inputs)
ans[i] = conv.add_out_of_center_l(a, gx.garray(targets), color=color).asarray()
print "b"
print abs(ans[0] - ans[1]).max()
if plot:
if not color:
from pylab import show, subplot
subplot(221)
show(inputs[0])
subplot(223)
show(ans[0][0])
subplot(224)
show(ans[1][0])
else:
from pylab import show, subplot
r = img_pix / color_mult
subplot(331)
show(inputs[0][:r])
r = ans[0].shape[1] / 3
subplot(332)
show(ans[0][0][:r])
subplot(334)
show(ans[1][0][:r])
subplot(335)
show(ans[1][0][r : 2 * r])
subplot(336)
show(ans[1][0][2 * r : 3 * r])
def sim_time(i):
n = pylab.randint(N_MIN, N_MAX)
alpha = pylab.rand()
net = random_network(n)
r = ne_capacity(net)*((1-MIN_DEMAND)*pylab.rand() + MIN_DEMAND)
tic = time.clock()
optimal_stackelberg(net,r,alpha)
val = (n,time.clock() - tic)
print val
return val
def make_2DLinearSeparable_Dataset(n):
xb = (rand(n)*2-1)/2-0.5
yb = (rand(n)*2-1)/2+0.5
xr = (rand(n)*2-1)/2+0.5
yr = (rand(n)*2-1)/2-0.5
inputs = []
for i in range(len(xb)):
inputs.append([xb[i],yb[i],1])
inputs.append([xr[i],yr[i],-1])
return inputs
def genererDonnees(n):
xb=(pl.rand(n)*2-1)/2-0.5
yb=(pl.rand(n)*2-1)/2+0.5
xr=(pl.rand(n)*2-1)/2+0.5
yr=(pl.rand(n)*2-1)/2-0.5
donnees=[]
for i in range(len(xb)):
donnees.append(((xb[i],yb[i]),-1))
donnees.append(((xr[i],yr[i]),1))
return donnees
def add_scatter():
ax = fig.add_axes([0.6, 0.125, 0.15, 0.4])
ax.axesPatch.set_alpha(axalpha)
N = 40
volume = 100 * rand(N)
color = 256 * rand(N)
darkgray = [0.2] * 3
plt.scatter(rand(N), rand(N), c=color, s=volume, alpha=0.75, edgecolor=darkgray)
plt.axis("tight")
ax.set_yticks([])
ax.set_xticks([])
def genererDonnees(n):
"Generer un jeu de donnees 2D lineairement separable de taille n"
xb=(rand(n)*2-1)/2-0.5
yb=(rand(n)*2-1)/2+0.5
xr=(rand(n)*2-1)/2+0.5
yr=(rand(n)*2-1)/2-0.5
donnees=[]
for i in range (len(xb)):
donnees.append(((xb[i],yb[i]),False))
donnees.append(((xr[i],yr[i]),True))
return donnees
def gen_data(n):
xb = (rand(n)*2-1)/2-0.5
yb = (rand(n)*2-1)/2+0.5
inputs = [[xb[i], yb[i]] for i in xrange(len(xb))]
targets = [[i] for i in repeat(1, len(xb))]
xr = (rand(n)*2-1)/2+0.5
yr = (rand(n)*2-1)/2-0.5
inputs = inputs + [[xr[i], yr[i]] for i in xrange(len(xr))]
targets = targets + [[i] for i in repeat(0, len(xr))]
return np.array(inputs), np.array(targets)
def choose_patches(IMAGES, L, batch_size=1000):
sz = int(sqrt(L))
imsz = shape(IMAGES)[0]
num_images = shape(IMAGES)[2]
BUFF = 4
X = matrix(zeros([L,batch_size],'d'))
for i in range(batch_size):
j = int(floor(num_images * rand()))
r = sz/2+BUFF+int(floor((imsz-sz-2*BUFF)*rand()))
c = sz/2+BUFF+int(floor((imsz-sz-2*BUFF)*rand()))
X[:,i] = reshape(IMAGES[r-sz/2:r+sz/2,c-sz/2:c+sz/2,j],[L,1])
return X
def generateData(m, type_):
data = []
if type_ == "own-ns":
feature_1_c1 = (rand(m)*3-1)/2 - 0.1
feature_2_c1 = (rand(m)*3-1)/4 + 0.2
feature_1_c2 = (rand(m)*2-1)/2 + 0.2
feature_2_c2 = (rand(m)*3-1)/2 - 0.1
for i in range(m):
data.append([feature_1_c1[i], feature_2_c1[i], 1])
for i in range(m):
data.append([feature_1_c2[i], feature_2_c2[i], -1])
elif type_ == "own-s":
feature_1_c1 = (rand(m)*2-1)/2 - 0.6
feature_2_c1 = (rand(m)*2-1)/2 + 0.6
feature_1_c2 = (rand(m)*2-1)/2 + 0.6
feature_2_c2 = (rand(m)*2-1)/2 - 0.6
for i in range(m):
data.append([feature_1_c1[i], feature_2_c1[i], 1])
for i in range(m):
data.append([feature_1_c2[i], feature_2_c2[i], -1])
return data
def make2DLinearSeparableDataset(n): """ generates a 2D linearly separable dataset with n samples. The third element of the sample is the label """ xb = (rand(n)*2-1)/2-0.5 yb = (rand(n)*2-1)/2+0.5 xr = (rand(n)*2-1)/2+0.5 yr = (rand(n)*2-1)/2-0.5 inputs = [] for i in range(len(xb)): inputs.append([xb[i],yb[i],1]) inputs.append([xr[i],yr[i],-1]) return inputs
def generateData(n): """ generates a 2D linearly separable dataset with n samples. The third element of the sample is the label """ xb = (rand(n)*2-1)/2-0.5 yb = (rand(n)*2-1)/2+0.5 xr = (rand(n)*2-1)/2+0.5 yr = (rand(n)*2-1)/2-0.5 inputs = [] for i in range(len(xb)): inputs.append(((xb[i],yb[i]),0)) inputs.append(((xr[i],yr[i]),1)) return inputs
def testActivityMap(self):
self.spk.dimensions = [5, 10]
self.spk.activity_map(t_start = 1000, t_stop = 2000, display=pylab.subplot(211), kwargs={'interpolation':'bicubic'})
positions = pylab.rand(2, 50)
self.spk.activity_map(float_positions = positions, display=pylab.subplot(212))
pylab.savefig("Plots/SpikeList_activitymaps.png")
pylab.close()
def __init__(self):
'''
perceptron initialization
'''
self.w = rand(2)*2 -1
self.learningRate = 0.05
self.test_steps = 10
def randomwalk(sf, N=None):
"""
Generates a randomwalk in a ndarray of given shape sf and length N
--------------------------------------------------------------------------
Usage:
Call: w = randomwalk(sf, N=None)
Input: sf size of ndarray w
N length of randomwalk
Output: ndarray w containing randomwalk of length N.
Note, that w is normalized to w.sum() = 1
--------------------------------------------------------------------------
Copyright (C) 2011 Michael Hirsch
"""
if N == None:
N = np.floor(np.prod(sf)/100)
w = np.zeros(sf);
ndims = len(np.shape(w))
center = np.ceil(np.array(sf)/2)
w[tuple(center)] = 1.
loc = center
for i in np.arange(N):
loc += ((pylab.rand(ndims)-0.5)*2).round()
loc = clip(loc,1,np.array(sf).min()-1)
w[tuple(loc)] += 1
w /= w.sum()
return w
def randu(*shape):
"""Generate uniformly random values in the range (-1,1).
This can usually be used as a drop-in replacement for `randn`
resulting in a different distribution for weight initializations.
Empirically, the choice of randu/randn can make a difference
for neural network initialization."""
return 2*rand(*shape)-1
def make_fiber(l, a, stepsize=0.5):
angles = np.random.standard_cauchy(l) * a
angles[0] += 2 * pi * pylab.rand()
angles = add.accumulate(angles)
coss = add.accumulate(cos(angles) * stepsize)
sins = add.accumulate(sin(angles) * stepsize)
return array([coss, sins]).transpose(1, 0)
def __init__(self):
""" perceptron initialization """
self.w = rand(2) * 2 - 1 # weights
self.learningRate = 0.01
self.momentum = 0.8
self.d = 3072
self.k = 500 # number of hidden layers
self.n = 10000
self.nTest = 2000
self.W1 = np.random.random_sample(
(self.d + 1, self.k)
) # [[rand(1)*2-1 for x in range(self.k)] for x in range(self.d+1)] #for W10
self.W2 = np.random.random_sample(
(self.k + 1, 1)
) #[[rand(1)*2-1 for x in range(1)] for x in range(self.k+1)]#for W20
self.Z1 = np.zeros((self.k + 1, 1))
self.Z2 = 0
self.f1 = np.zeros((self.k + 1, 1))
self.f2 = 0
self.maxIteration = 10000
self.globalError = np.zeros((self.maxIteration, 1))
self.globalTestError = np.zeros((self.maxIteration, 1))
self.batchsize = 1
self.dict = cPickle.load(open("cifar_2class_py2.p", "rb"))
self.shuffledData = np.random.permutation(self.n)
self.D1 = np.zeros((self.d + 1, self.k))
self.D2 = np.zeros((self.k + 1, 1))
self.G1 = np.zeros((self.d + 1, self.k))
self.G2 = np.zeros((self.k + 1, 1))
self.dictTraindata = np.zeros((self.n, self.d))
self.dictTrainLables = np.zeros((self.n, 1))
self.dictTestdata = np.zeros((self.nTest, self.d))
self.dictTestLables = np.zeros((self.nTest, 1))
def plotReachSet_norm1(self, NUM, figname):
fig = p.figure()
for j in range(n):
ax = fig.add_subplot(2,2,j+1 , aspect='equal')
ax.set_xlim(0, 4)
ax.set_ylim(0, 1)
ax.set_xlabel('$x_'+str(j+1)+'$')
ax.set_ylabel('$y_'+str(j+1)+'$')
for trace in self:
for i in [int(floor(k*len(trace.T)/NUM)) for k in range(NUM)]:
verts = [(trace.x[i][j] + 1/trace.d_norm1[i][2*j], trace.y[i][j] ),
(trace.x[i][j] , trace.y[i][j] - 1/trace.d_norm1[i][2*j+1]),
(trace.x[i][j] - 1/trace.d_norm1[i][2*j], trace.y[i][j] ),
(trace.x[i][j] , trace.y[i][j] + 1/trace.d_norm1[i][2*j+1])]
# poly = Ellipse((trace.x[i][j],trace.y[i][j]), width=trace.d1[i], height=trace.d2[i], angle=trace.theta[i])
poly = Polygon(verts, facecolor='0.8', edgecolor='k')
ax.add_artist(poly)
poly.set_clip_box(ax.bbox)
poly.set_alpha(1)
if i==0:
poly.set_facecolor('r')
else:
poly.set_facecolor(p.rand(3))
#for trace in self:
#e = Ellipse((trace.x[0][j],trace.y[0][j]), width=trace.d1[0], height=trace.d2[0], angle=trace.theta[0])
#ax.add_artist(e)
#e.set_clip_box(ax.bbox)
#e.set_alpha(1)
#e.set_facecolor('r')
#e.set_edgecolor('r')
p.savefig(figname)
def test_matrix_to_grid(plot=False):
# use the same data for both experiments:
num_imgs = 200
img_res = 20
square_size = 10
regions_per_square = (img_res / square_size) ** 2
img_pix = img_res ** 2
inputs = rand(num_imgs * regions_per_square, square_size ** 2) < 0.1
ans = [None] * 2
print "a"
for (i, gx) in enumerate([g, gc]):
conv.g = gx # wonderful. This seems to work. At least.
conv._cu = gx._cudamat
a = gx.garray(inputs)
ans[i] = conv.matrix_to_grid(a, img_res).asarray()
print "b"
print abs(ans[0] - ans[1]).max()
# quite successful, indeed.
if plot:
from pylab import show, subplot
subplot(221)
show(inputs[0])
subplot(223)
show(ans[0][0])
subplot(224)
show(ans[1][0])
def test_block_hankel(self):
"""
Block hankel function.
"""
y = pl.rand(3, 100)
Y = sysid.subspace.block_hankel(y, 5)
self.assertEqual(Y.shape, (15, 95))
def createCellsFixedNum(self):
''' Create population cells based on fixed number of cells'''
cellModelClass = Cell
cells = []
seed(f.sim.id32('%d'%(f.cfg['randseed']+self.tags['numCells'])))
randLocs = rand(self.tags['numCells'], 3) # create random x,y,z locations
for icoord, coord in enumerate(['x', 'y', 'z']):
if coord+'Range' in self.tags: # if user provided absolute range, convert to normalized
self.tags[coord+'normRange'] = [point / f.net.params['size'+coord.upper()] for point in self.tags[coord+'Range']]
if coord+'normRange' in self.tags: # if normalized range, rescale random locations
minv = self.tags[coord+'normRange'][0]
maxv = self.tags[coord+'normRange'][1]
randLocs[:,icoord] = randLocs[:,icoord] / (maxv-minv) + minv
for i in xrange(int(f.rank), f.net.params['scale'] * self.tags['numCells'], f.nhosts):
gid = f.lastGid+i
self.cellGids.append(gid) # add gid list of cells belonging to this population - not needed?
cellTags = {k: v for (k, v) in self.tags.iteritems() if k in f.net.params['popTagsCopiedToCells']} # copy all pop tags to cell tags, except those that are pop-specific
cellTags['xnorm'] = randLocs[i,0] # set x location (um)
cellTags['ynorm'] = randLocs[i,1] # set y location (um)
cellTags['znorm'] = randLocs[i,2] # set z location (um)
cellTags['x'] = f.net.params['sizeX'] * randLocs[i,0] # set x location (um)
cellTags['y'] = f.net.params['sizeY'] * randLocs[i,1] # set y location (um)
cellTags['z'] = f.net.params['sizeZ'] * randLocs[i,2] # set z location (um)
if 'propList' not in cellTags: cellTags['propList'] = [] # initalize list of property sets if doesn't exist
cells.append(cellModelClass(gid, cellTags)) # instantiate Cell object
if f.cfg['verbose']: print('Cell %d/%d (gid=%d) of pop %s, on node %d, '%(i, f.net.params['scale'] * self.tags['numCells']-1, gid, self.tags['popLabel'], f.rank))
f.lastGid = f.lastGid + self.tags['numCells']
return cells
def _shapeStim(self,
isi=1,
variation=0,
width=0.05,
weight=10,
start=0,
finish=1,
stimshape='gaussian'):
from pylab import r_, convolve, shape, exp, zeros, hstack, array, rand
# Create event times
timeres = 0.001 # Time resolution = 1 ms = 500 Hz (DJK to CK: 500...?)
pulselength = 10 # Length of pulse in units of width
currenttime = 0
timewindow = finish - start
allpts = int(timewindow / timeres)
output = []
while currenttime < timewindow:
# Note: The timeres/2 subtraction acts as an eps to avoid later int rounding errors.
if currenttime >= 0 and currenttime < timewindow - timeres / 2:
output.append(currenttime)
currenttime = currenttime + isi + variation * (rand() - 0.5)
# Create single pulse
npts = int(pulselength * width / timeres)
x = (r_[0:npts] - npts / 2 + 1) * timeres
if stimshape == 'gaussian':
pulse = exp(-2 * (2 * x / width - 1)**
2) # Offset by 2 standard deviations from start
pulse = pulse / max(pulse)
elif stimshape == 'square':
pulse = zeros(shape(x))
pulse[int(npts / 2):int(npts / 2) +
int(width / timeres)] = 1 # Start exactly on time
else:
raise Exception('Stimulus shape "%s" not recognized' % stimshape)
# Create full stimulus
events = zeros((allpts))
events[array(array(output) / timeres, dtype=int)] = 1
fulloutput = convolve(
events, pulse, mode='full'
) * weight # Calculate the convolved input signal, scaled by rate
fulloutput = fulloutput[int(npts / 2 - 1):int(
-npts / 2
)] # Slices out where the convolved pulse train extends before and after sequence of allpts.
fulltime = (r_[0:allpts] * timeres +
start) * 1e3 # Create time vector and convert to ms
fulltime = hstack(
(0, fulltime, fulltime[-1] + timeres *
1e3)) # Create "bookends" so always starts and finishes at zero
fulloutput = hstack(
(0, fulloutput,
0)) # Set weight to zero at either end of the stimulus period
events = hstack((0, events, 0)) # Ditto
stimvecs = deepcopy([fulltime, fulloutput,
events]) # Combine vectors into a matrix
return stimvecs
def randu(*shape):
"""Generate uniformly random values in the range (-1,1).
This can usually be used as a drop-in replacement for `randn`
resulting in a different distribution for weight initializations.
Empirically, the choice of randu/randn can make a difference
for neural network initialization."""
return 2*rand(*shape)-1
def main():
plt.ion()
fil = FletcherFilter()
Niter = 12
logp = plt.zeros((Niter,2))
for k in range(Niter):
while True:
#print k
p = plt.rand(2)
if not fil.dominated(p):
break
logp[k] = p
fil.add(p, 0.0, 0.0)
ff = fil.values[fil.valid]
ff = plt.r_[[[1e-6,1]], ff[plt.argsort(ff[:,0])], [[1,1e-6]]]
ww = plt.zeros((ff.shape[0] * 2 - 1, 2))
ww[::2] = ff
ww[1::2,0] = ff[1:,0]
ww[1::2,1] = ff[:-1,1]
plt.loglog(ww[:,0], ww[:,1], '-')
plt.loglog(logp[:,0], logp[:,1], 'ys-', lw=2)
plt.axis([0,1,0,1])
plt.axis('equal')
plt.grid()
code.interact()
def __init__(self): """ inicialización del perceptron """ self.w = rand(2)*2-1 # pesos self.tasaAprendizaje = 0.1
def make_fibrous_image(shape,
nfibers=300,
l=300,
a=0.2,
stepsize=0.5,
limits=(0.1, 1.0),
blur=1.0):
h, w = shape
lo, hi = limits
result = zeros(shape)
for i in range(nfibers):
v = pylab.rand() * (hi - lo) + lo
fiber = make_fiber(l, a, stepsize=stepsize)
y, x = randint(0, h - 1), randint(0, w - 1)
fiber[:, 0] += y
fiber[:, 0] = clip(fiber[:, 0], 0, h - .1)
fiber[:, 1] += x
fiber[:, 1] = clip(fiber[:, 1], 0, w - .1)
for y, x in fiber:
result[int(y), int(x)] = v
result = ndi.gaussian_filter(result, blur)
result -= amin(result)
result /= amax(result)
result *= (hi - lo)
result += lo
return result
def test_json():
sc.heading('Testing JSON read/write functions')
not_jsonifiable = sc.Blobject(
) # Create an object that can't be JSON serialized
print('Testing jsonifying a NON-jsonifiable object:')
notjson = sc.jsonify(not_jsonifiable,
die=False) # Will return a string representation
sc.sanitizejson(not_jsonifiable,
die=True) # Will still not die thanks to jsonpickle
jsonifiable = sc.objdict().make(keys=['a', 'b'], vals=pl.rand(10))
json_obj = sc.jsonify(jsonifiable)
json_str = sc.jsonify(jsonifiable, tostring=True,
indent=2) # kwargs are passed to json.dumps()
print('Not-a-JSON as sanitized object:')
print(notjson)
print('JSON as sanitized object:')
print(json_obj)
print('JSON as string:')
print(json_str)
return json_str
def norm(pars, noise=0.0, optimum='min', delay=None):
if delay:
pl.pause(delay * (0.5 + 0.5 * pl.rand())) # Add a noticeable delay
err = pl.linalg.norm(pars)
err = addnoise(err, noise)
if optimum == 'max':
err = -err
return err
def _sample_posteriors_noloop(self, true_N_A, p_A, N_samples):
true_N_B = self.N_u - true_N_A
N_values = pl.shape(true_N_A)[0]
posteriors = pl.zeros((N_samples, N_values))
for (i, (t_N_A, t_N_B)) in enumerate(zip(true_N_A, true_N_B)):
A_probs = pl.ones((N_samples, self.N_u))
A_probs[:, :t_N_A] *= self.p_uA_given_A
A_probs[:, t_N_A:] *= self.p_uA_given_B
B_probs = pl.ones((N_samples, self.N_u))
B_probs[:, :t_N_A] *= self.p_uB_given_A
B_probs[:, t_N_A:] *= self.p_uB_given_B
N_A = pl.sum(A_probs > pl.rand(N_samples, self.N_u), 1)
N_B = pl.sum(B_probs > pl.rand(N_samples, self.N_u), 1)
posteriors[:, i] = self._p_A_given_N_A(N_A, p_A, N_B)
return pl.mean(posteriors, 0)
def bounded_gaussian_noise(shape, sigma, maxdelta):
n, m = shape
deltas = pylab.rand(2, n, m)
deltas = ndi.gaussian_filter(deltas, (0, sigma, sigma))
deltas -= np.amin(deltas)
deltas /= np.amax(deltas)
deltas = (2 * deltas - 1) * maxdelta
return deltas
def make_noise_at_scale(shape, scale):
h, w = shape
h0, w0 = int(h / scale + 1), int(w / scale + 1)
data = pylab.rand(h0, w0)
with np.warnings.catch_warnings():
np.warnings.simplefilter("ignore")
result = ndi.zoom(data, scale)
return result[:h, :w]
def boot_p(pc, nsamp, bootstraps=2000):
"""Given a probability value p and sample size n, return us bootstraps
number of probability values obtained by random resampling based on p."""
r = pylab.rand(nsamp, bootstraps)
z = pylab.zeros((nsamp, bootstraps))
idx = pylab.find(r < pc)
z.flat[idx] = 1
booted_p = z.mean(axis=0)
return booted_p
def natural_selection(selected_mate, survival_rate):
'''
randomly kill individuals with low survival rate
killed individuals will have genotype -1
'''
r = pylab.rand(len(selected_mate))
selected_mate[survival_rate[selected_mate] < r] = -1
#print 'killed',sum(selected_mate == -1),'individuals'
return selected_mate
def __init__(self, eta, max_iterations, dimension):
""" Construtor do objeto Perceptron, recebe a taxa de
aprendizado (eta) e o número máximo de iterações
que o algoritmo pode executar durante um treinamento. """
self.w = rand(dimension) * 2 - 1 # weights
self.learningRate = eta
self.max_iterations = max_iterations
self.history = [list(self.w)]
self.dimension = dimension
def permute(a):
"""
Randomly permute the elements in array a
"""
for n in range(len(a)):
m = int(pylab.rand() * (len(a) - n)) + n
t = a[m]
a[m] = a[n]
a[n] = t
def funcNbofAttr(P, G):
endDir = 'G_%.3f_P_%.5f.npy' %(G,P)
try:
pattsx = data2array(dir_prix + '/patterns_' + endDir, mmap_mode="r+")
pattsA = data2array(dir_priA + '/patterns_' + endDir, mmap_mode="r+")
return len(pattsx)
except:
try:
pattsx = data2array(dir_prix + '/allPatt_' + endDir)
pattsA = data2array(dir_priA + '/allPatt_' + endDir)
except:
N = data2array(dco).shape[0]
pattsx = zeros((mmax,N))
pattsA = zeros((mmax,N))
conn = {'connAd': dco,
'normType': '1'}
noise = {'colors': None}
model = {'model': 'HopfieldBasedStatic',
'threshold': 'local',
'tauT': 0,
'P': P,
'G': G}
out = []
other = {'init': 'rand',
'dens': rand(),} #p2,!!!!!!!!!!!!!! RAND
for d in range(mmax):
eva = main.evaCure(evaCon=conn, evaNoi=noise, evaMod=model, out=out, **other)
eva.toEquilibrium()
pattsx[d] = eva.evaMod.x.copy()
pattsA[d] = eva.evaMod.A.copy()
array2data(pattsx, dir_prix + '/allPatt_' + endDir)
array2data(pattsA, dir_priA + '/allPatt_' + endDir)
patts = pattsx # !!!!!!!!!
S = sortBy(patts.mean(1) - patts.mean(), inverse=1)[0]
C1, freq = preClustering(patts[S], sim_coef=sim_coef, sim_func=similarity_Euclidean)
C2, freq = preClustering(patts[S][C1], freq=freq, sim_coef=sim_coef, sim_func=fPearsonCorrelation)
SC, freq = sortBy(freq, inverse=1)
array2data(pattsx[S][C1][C2][SC], dir_prix + '/patterns_' + endDir)
array2data(pattsA[S][C1][C2][SC], dir_priA + '/patterns_' + endDir)
array2data(freq, dir_priT + '/tendances_' + endDir)
os.system('rm ' + dir_prix + '/allPatt_' + endDir)
os.system('rm ' + dir_priA + '/allPatt_' + endDir)
return len(pattsx[S][C1][C2][SC])
def main():
zz = nx.zeros([10,10])
print 'tr(zz)=',trace(zz)
oo = nx.ones([4,4],nx.Float)
print 'tr(oo)=',trace(oo)
aa = rand(128,128)
print 'tr(aa)=',trace(aa)
print 'oo:',oo
in_place_mult(3,oo)
print '3*oo:',oo
def stochastic_equations(last_Y, ts, g, b):
Y = last_Y
access_rate = b * (N0 - Y) * Y / N0
denial_rate = g * Y
#generate random numbers
rand1 = pl.rand()
rand2 = pl.rand()
#time until either event occurs
ts = -np.log(rand2) / (denial_rate + access_rate)
if rand1 < (access_rate / (denial_rate + access_rate)):
# access, one more informed agent
Y += 1
else:
# denial, one fewer informed agent
Y -= 1
return [Y, ts]
def _sample_posteriors(self, true_N_A, p_A, N_samples):
true_N_B = self.N_u - true_N_A
N_values = pl.shape(true_N_A)[0]
posteriors = pl.zeros((N_samples, N_values))
for i in range(N_samples):
for (j, (t_N_A, t_N_B)) in enumerate(zip(true_N_A, true_N_B)):
A_given_A = pl.ones(t_N_A) * self.p_uA_given_A
A_given_B = pl.ones(t_N_B) * self.p_uA_given_B
A_probs = pl.hstack((A_given_A, A_given_B))
B_given_A = pl.ones(t_N_A) * self.p_uB_given_A
B_given_B = pl.ones(t_N_B) * self.p_uB_given_B
B_probs = pl.hstack((B_given_A, B_given_B))
N_A = pl.sum(A_probs > pl.rand(self.N_u))
N_B = pl.sum(B_probs > pl.rand(self.N_u))
posteriors[i, j] = self._p_A_given_N_A(N_A, N_B)
return pl.mean(posteriors, 0)
def makestim(isi=1,
variation=0,
width=0.05,
weight=10,
start=0,
finish=1,
stimshape='gaussian'):
from pylab import r_, convolve, shape
# Create event times
timeres = 0.005 # Time resolution = 5 ms = 200 Hz
pulselength = 10 # Length of pulse in units of width
currenttime = 0
timewindow = finish - start
allpts = int(timewindow / timeres)
output = []
while currenttime < timewindow:
if currenttime >= 0 and currenttime < timewindow:
output.append(currenttime)
currenttime = currenttime + isi + variation * (rand() - 0.5)
# Create single pulse
npts = min(pulselength * width / timeres,
allpts) # Calculate the number of points to use
x = (r_[0:npts] - npts / 2 + 1) * timeres
if stimshape == 'gaussian':
pulse = exp(-(x / width * 2 - 2)**
2) # Offset by 2 standard deviations from start
pulse = pulse / max(pulse)
elif stimshape == 'square':
pulse = zeros(shape(x))
pulse[int(npts / 2):int(npts / 2) +
int(width / timeres)] = 1 # Start exactly on time
else:
raise Exception('Stimulus shape "%s" not recognized' % stimshape)
# Create full stimulus
events = zeros((allpts))
events[array(array(output) / timeres, dtype=int)] = 1
fulloutput = convolve(
events, pulse, mode='same'
) * weight # Calculate the convolved input signal, scaled by rate
fulltime = (r_[0:allpts] * timeres +
start) * 1e3 # Create time vector and convert to ms
fulltime = hstack(
(0, fulltime, fulltime[-1] + timeres *
1e3)) # Create "bookends" so always starts and finishes at zero
fulloutput = hstack(
(0, fulloutput,
0)) # Set weight to zero at either end of the stimulus period
events = hstack((0, events, 0)) # Ditto
stimvecs = [fulltime, fulloutput, events] # Combine vectors into a matrix
return stimvecs
def sample(W,g,batch_size, vis_gauss=False):
v,h = W.v, W.h
V = rand(batch_size,v)
for gg in range(g):
H = Rsigmoid(W*V)
if vis_gauss:
V = W.T()*H + randn(batch_size,v)
else:
V = Rsigmoid(W.T()*H)
return V,H
def bar_graph():
k = 8
x = plb.arange(k)
for z in x:
y1 = plb.rand(k) * (1 - x / k)
y2 = plb.rand(k) * (1 - x / k)
plb.axes([0.075, 0.075, .88, .88])
plb.bar(x, +y1, facecolor='#9922aa', edgecolor='green')
plb.bar(x, -y2, facecolor='#ff3366', edgecolor='green')
for a, b in zip(x, y1):
plb.text(a+0.41, b+0.08, '%.3f' % b, ha='center', va='bottom')
for a, b in zip(x, y2):
plb.text(a+0.41, b+0.08, '%.3f' % b, ha='center', va='top')
plb.xlim(-.5, k), plb.ylim(-1.12, +1.12)
plb.grid(True)
plb.pause(1)
plb.cla()
def testActivityMap(self):
self.spk.dimensions = [5, 10]
self.spk.activity_map(t_start=1000,
t_stop=2000,
display=pylab.subplot(211),
kwargs={'interpolation': 'bicubic'})
positions = pylab.rand(2, 50)
self.spk.activity_map(float_positions=positions,
display=pylab.subplot(212))
pylab.savefig("Plots/SpikeList_activitymaps.png")
pylab.close()
def test_bipartite_matching():
from pylab import plot, rand
from numpy import zeros, arange
N = 10
a = rand(N,2)
a[:,0] = arange(N)
a[:,1] = 0
b = rand(N+2,2)
b[:,0] = arange(N+2)
b[:,1] = 1
d = zeros((a.shape[0],b.shape[0]))
for i,ai in enumerate(a):
for j,bi in enumerate(b):
d[i,j] = ((ai-bi)**2).sum()
assignment,cost = bipartite_matching( d )
print "Matching cost: ", cost.value
plot(a[:,0],a[:,1],'o')
plot(b[:,0],b[:,1],'s')
for i,j in assignment.iteritems():
plot([ a[i,0], b[j,0] ], [ a[i,1], b[j,1] ],'k--')
return assignment
def knockout_uniformly_at_random(in_fname='noisy_data.csv', out_fname='missing_noisy_data.csv', pct=20.):
""" replace data.csv y column with uniformly random missing entries
Parameters
----------
pct : float, percent to knockout
"""
data = pl.csv2rec(in_fname)
for i, row in enumerate(data):
if pl.rand() < pct/100.:
data[i].y = pl.nan
pl.rec2csv(data, out_fname)
def random_distort(images, maxdelta=2.0, sigma=30.0):
n, m = images[0].shape
deltas = pylab.rand(2, n, m)
deltas = ndi.gaussian_filter(deltas, (0, sigma, sigma))
deltas -= np.amin(deltas)
deltas /= np.amax(deltas)
deltas = (2 * deltas - 1) * maxdelta
#print np.amin(deltas), np.amax(deltas)
xy = np.transpose(np.array(np.meshgrid(range(n), range(m))),
axes=[0, 2, 1])
#print(xy.shape, deltas.shape)
deltas += xy
return [ndi.map_coordinates(image, deltas, order=1) for image in images]
def plotPT(self, n):
N = size(self.Asts)
PLtest = self
fig = figure()
ax = fig.add_subplot(1, 1, 1)
for i in range(n):
PLtest.set_w(rand(N))
ax.scatter(PLtest.s(), PLtest.r(), color='black', s=1)
ax.scatter(self.s(), self.r(), color='red', s=10)
xlabel('$\sigma$')
ylabel('r')
grid()
show()
def __init__(self, phase_potrait, network, info=None, position=None):
self.system = phase_potrait
self.network = network
self.CYCLES = 10
self.info = info
self.initial_condition = self.system.load_initial_condition(
pl.rand(), pl.rand(), pl.rand())
self.fig = pl.figure('Voltage Traces',
figsize=(6, 2),
facecolor='#EEEEEE')
self.ax = self.fig.add_subplot(111, frameon=False, yticks=[])
self.li_b, = self.ax.plot([], [], 'b-', lw=2.)
self.li_g, = self.ax.plot([], [], 'g-', lw=2.)
self.li_r, = self.ax.plot([], [], 'r-', lw=2.)
self.li_y, = self.ax.plot([], [], 'y-', lw=2.)
self.ax.set_xlabel(r'time (sec.)', fontsize=20)
self.ax.set_xticklabels(np.arange(0., 1., 0.1), fontsize=15)
self.ax.set_yticklabels(np.arange(0., 1., 0.1), fontsize=15)
self.ax.set_xlim(0., 100.)
self.ax.set_ylim(-8.5, 1.5)
#self.fig.tight_layout()
self.key_func_dict = dict(u=traces.increase_cycles,
i=traces.decrease_cycles)
self.fig.canvas.mpl_connect('key_press_event', self.on_key)
self.fig.canvas.mpl_connect('axes_enter_event', self.focus_in)
if not position == None:
try:
self.fig.canvas.manager.window.wm_geometry(position)
except:
pass
def __init__(self, w_=rand(2) * 2 - 1, tasaApren_=0.1):
"""
Metodo constructor del preceptron,
inicialza los valores por defecto.
Parametros:
w_: array-1d
Pesos actualizados después del ajuste.
tasaApren_: float
Tasa de aprendizaje.
"""
self.w = w_ # Vector w, representa los pesos.
self.tasaApren = tasaApren_ # Tasa de aprendizaje.