def plot_voltage_after(ax):
data = core.get_data('fig3/2019-02-02_membrane_potential_psth_end.npy')[10]
data_teach=core.get_data("fig3/2019-02-02_membrane_potential_psth_teacher.npy")[10]
time_params=core.get_data("fig3/time_parameters.npy").item()
#ax.set_xlim([0.,.5])
time=np.arange(0,500)*time_params["dt"]
#subtract resting potential
data=data-70.
data_teach=data_teach-70.
ax.plot(time,data,color=core.label_to_color["natural"],linewidth=core.linewidth["voltage"])
ax.plot(time,data_teach,color=core.label_to_color["teacher"],linewidth=core.linewidth["voltage"])
core.show_axis(ax)
core.make_spines(ax)
#plotf.trace_plot(data)
ax.set_xlabel(core.x_labels["voltage_trace"])
ax.set_ylabel(core.y_labels["voltage_trace"])
#ax.legend(("student","teacher"),loc=0,fontsize=8.,labelspacing=0.1,borderaxespad=0.1,borderpad=0.,handlelength=0.5,frameon=False)
pass
def plot_var_result_example(ax):
ax.set_xlim([-0.0,5.5])
data_hr_ht = core.get_data('fig6/variance_hr_ht.npy')
data_lr_ht = core.get_data('fig6/variance_lr_ht.npy')
data_hr_lt = core.get_data('fig6/variance_hr_lt.npy')
time_params=core.get_data('fig6/time_parameters.npy').item()
weighttrace=np.concatenate(((data_hr_ht,),(data_lr_ht,),(data_hr_lt,)))
core.show_axis(ax)
core.make_spines(ax)
ax.set_prop_cycle(cycler('color',[core.label_to_color["hr_ht"],core.label_to_color["lr_ht"],core.label_to_color["hr_lt"]] ))
plotf.trace_plot(ax,weighttrace,time_params,linewidth=core.linewidth["normal"])
ax.axvline(x=5,color=core.label_to_color["aux_lines"],linewidth=core.linewidth["aux_lines"],ls="dashed")
ax.text(5.,75.,r"$t_0$")
ax.set_xlabel(core.x_labels["trace_plot"])
ax.set_ylabel(core.y_labels["trace_plot"])
ax.legend((r"$w_1$",r"$w_2$",r"$w_3$"),loc=2,fontsize=8.,handlelength=0.5,frameon=False)
pass
def plot_heatplot_heterosyn(ax):
data = core.get_data('fig5/Delta_w_het_grid.npy')
data_aux = core.get_data('fig5/Delta_w_hom_grid.npy')
vmin = min(np.amin(data), np.amin(data_aux))
vmax = max(np.amax(data), np.amax(data_aux))
w_range = core.get_data('fig5/w_range.npy')
core.show_axis(ax)
core.make_spines(ax)
label = r"$\Delta w_{\mathrm{ust}}$ [%]"
labelpad = -8.
plotf.hplot(ax, data, vmin, vmax, w_range, label, labelpad)
ax.axhline(y=1.85,
color=core.label_to_color["aux_line_heat"],
linewidth=core.linewidth["aux_lines"],
ls="dashed")
ax.set_xlabel(core.x_labels["heat_plot"],
labelpad=core.linewidth["lp_heat"])
ax.set_ylabel(core.y_labels["heat_plot"],
labelpad=core.linewidth["lp_heat"])
pass
def plot_var_hist_hr_ht(ax):
data_usp = core.get_data('fig6/histogram_hr_ht.npy')
ax.set_ylim([0.,20.])
ax.set_xlim([0.,5.])
counts,bins,patches=ax.hist(data_usp,bins=np.arange(0.,25.,.5),log=1,orientation="horizontal",density=1,rwidth=1.,color=core.label_to_color["hr_ht"])
ax.set_xlim(ax.get_xlim()[::-1])
ax.spines['left'].set_position(('axes', 1.))
ax.tick_params(axis="y",direction="in", pad=-15.)
ax.tick_params(axis="x",direction="out", pad=-0.3)
core.show_axis(ax)
core.make_spines(ax)
tick_formatter =ticker.LogFormatterExponent()
ax.xaxis.set_major_formatter(tick_formatter)
ax.set_xlabel(core.x_labels["hist"],labelpad=-2.)
pass
def plot_var_lowr_hightau(ax):
data = core.get_data('fig6/usp_trace_lr_ht.npy')[0]
ax.set_ylim([0.,20.])
ax.set_xlim([0.,0.5])
time_params=core.get_data('fig6/time_parameters.npy').item()
time=np.arange(0,time_params["learning_time"],time_params["dt"])
ax.text(0.3,22,r"$r_2=5\rm{ Hz }$")
ax.text(0.3,17,r"$\tau_2=15\rm{ ms}$")
core.show_axis(ax)
core.make_spines(ax)
ax.plot(time,data,color=core.label_to_color["lr_ht"],linewidth=core.linewidth["normal"])
ax.set_xlabel(core.x_labels["usp_trace"])
ax.set_ylabel(core.y_labels["usp_trace"])
pass
def plot_attenuation_example_abs(ax):
data_prox = core.get_data('fig6/attenuation_proximal_synapse.npy')
data_dist = core.get_data('fig6/attenuation_distal_synapse.npy')
time_params=core.get_data('fig6/time_parameters.npy').item()
ax.set_prop_cycle(cycler('color',[core.label_to_color["attenuation_ldist"],core.label_to_color["attenuation_hdist"]] ))
weighttrace=np.concatenate(((data_prox,),(data_dist,)))
core.show_axis(ax)
core.make_spines(ax)
plotf.trace_plot(ax,weighttrace,time_params,perc=False,linewidth=core.linewidth["normal"])
ax.set_xlabel(core.x_labels["trace_plot"])
ax.set_ylabel(core.y_labels["trace_plot_dend_abs"])
# ax.legend((r"$d=3\mu m$",r"$d=7\mu m$"),loc=2,fontsize=8.,handlelength=0.5,frameon=False)
pass
def plot_spikes_post(ax):
data = core.get_data('fig5/teacher_spikes_illu.npy')
ax.plot(data[10, 0:500],
color=core.label_to_color["aux_lines"],
linewidth=core.linewidth["spikes"])
core.hide_axis(ax)
pass
def plot_attenuation_summary_abs(ax):
ax.set_xlim([0.,11.])
data_wchange = core.get_data('fig6/att_weight_change_abs.npy')
data_dist = core.get_data('fig6/distances.npy')
color=[core.label_to_color["attenuation_im"]]*np.shape(data_dist)[0]
color[2]=core.label_to_color["attenuation_ldist"]
color[-3]=core.label_to_color["attenuation_hdist"]
core.show_axis(ax)
core.make_spines(ax)
ax.scatter(data_dist,data_wchange,color=color,s=core.linewidth["scatter"])
ax.set_xlabel(r"$d[\mu m]$")
ax.set_ylabel(core.y_labels["weightchange_dend_abs"])
pass
def plot_phase3(ax):
data_hom = core.get_data('fig5/w_hom_phase3.npy')
data_het = core.get_data('fig5/w_het_phase3.npy')
time_params = core.get_data('fig5/time_parameters.npy').item()
ax.set_xlim([-.5, 5.5])
x_ticks = np.array((0., 2.5, 5.))
ax.xaxis.set_ticks(x_ticks)
weighttrace = np.concatenate(((data_hom, ), (data_het, )))
ax.set_prop_cycle(
cycler('color',
[core.label_to_color["stim"], core.label_to_color["unstim"]]))
core.show_axis(ax)
core.make_spines(ax)
plotf.trace_plot(ax,
weighttrace,
time_params,
linewidth=core.linewidth["normal"])
ax.axvline(x=0,
color=core.label_to_color["aux_lines"],
linewidth=core.linewidth["aux_lines"],
ls="dashed")
ax.text(0., 99.05, r"$t_0$")
ax.axvline(x=5,
color=core.label_to_color["aux_lines"],
linewidth=core.linewidth["aux_lines"],
ls="dashed")
ax.text(5., 99.05, r"$t_1$")
ax.set_xlabel(core.x_labels["trace_plot"])
ax.set_ylabel(core.y_labels["trace_plot"])
ax.legend((r"$w_{\rm st}$", r"$w_{\rm ust}$"),
loc=0,
fontsize=8.,
labelspacing=0.1,
borderaxespad=0.1,
borderpad=0.,
handlelength=0.5,
frameon=False)
pass
def plot_spikes_stim2(ax):
data = core.get_data('fig5/inputspikes_stim_illu.npy')
ax.plot(data[20, 0:500],
color=core.label_to_color["aux_lines"],
linewidth=core.linewidth["spikes"])
core.hide_axis(ax)
pass
def plot_heatplot_hom_het_comparison(ax):
data = core.get_data('fig5/Sign_delta_w_comparison.npy')
w_range = core.get_data('fig5/w_range.npy')
core.show_axis(ax)
core.make_spines(ax)
labelpad=0.
label=r"$\rm{Sign}(\Delta w_{\mathrm{st}}\times\Delta w_{\mathrm{unst}})$"
plotf.hplot(ax,data,-1.,1.,w_range,label=label,labelpad=labelpad)
ax.axhline(y=1.85,color=core.label_to_color["aux_line_heat"],linewidth=core.linewidth["aux_lines"],ls="dashed")
ax.set_xlabel(core.x_labels["heat_plot"],labelpad=core.linewidth["lp_heat"])
ax.set_ylabel(core.y_labels["heat_plot"],labelpad=core.linewidth["lp_heat"])
pass
def plot_var_result_summary(ax):
ax.set_xlim([0.,31.])
data_wchange = core.get_data('fig6/wfinal_com.npy')
data_dist = core.get_data('fig6/usp_var_com.npy')
core.show_axis(ax)
core.make_spines(ax)
color=[core.label_to_color["attenuation_im"]]*np.shape(data_dist)[0]
color[4]=core.label_to_color["lr_ht"]
color[20]=core.label_to_color["hr_lt"]
color[24]=core.label_to_color["hr_ht"]
ax.scatter(data_dist,data_wchange,color=color,s=core.linewidth["scatter"])
ax.set_xlabel(r'$\sigma^2(\mathrm{USP})$')
ax.set_ylabel(core.y_labels["weightchange"])
pass
def plot_spikes_before(ax):
data = core.get_data('fig3/2019-02-02_output_spikes_psth_start.npy')
ax.scatter(np.nonzero(data)[1],np.nonzero(data)[0],color=core.label_to_color["natural"],s=0.01)
core.show_axis(ax)
core.make_spines(ax)
ax.set_xlabel(core.x_labels["spikes"])
ax.set_ylabel(core.y_labels["spikes"])
pass
def plot_histo_after(ax):
data = core.get_data('fig3/2019-02-02_output_spikes_psth_end.npy')
data_teach=core.get_data("fig3/2019-02-02_output_spikes_psth_teacher.npy")
print("data",np.shape(data))
spiketimes=np.nonzero(data)[1]*0.0005
spiketimes_teach=np.nonzero(data_teach)[1]*0.0005
ax.set_prop_cycle(cycler('color',[core.label_to_color["natural"],core.label_to_color["teacher"]] ))
bins=np.linspace(0.,0.25,20,endpoint=False)
ax.hist([spiketimes,spiketimes_teach], bins, label=['x', 'y'],normed=1.)
core.show_axis(ax)
core.make_spines(ax)
ax.set_xlabel(core.x_labels["histo"])
ax.set_ylabel(core.y_labels["histo"])
pass
def plot_weightpath_nat(ax):
import parameter_vectorplot as parameter
N=2
para=parameter.parameter
para["color"]=core.label_to_color["natural"]
data_quiver = core.get_data('fig3/2019-01-07_vectorplot_nat.npy')
data_cost=core.get_data('fig3/2019-01-07_contour_cost.npy')
data_wpath=core.get_data("fig3/2018-08-24_nwpath.npy")
data_wpath=N*data_wpath
#normalize
norm_data=np.sqrt(np.square(data_quiver[:,:,0])+np.square(data_quiver[:,:,1]))
data_quiver[:,:,0]=data_quiver[:,:,0]/norm_data
data_quiver[:,:,1]=data_quiver[:,:,1]/norm_data
plotf.wpathplot(ax,data_quiver,data_wpath,**para)
levels=[0.001,0.003,0.005,0.009,0.015,0.02,0.03,0.04]
plotf.contourplot(ax,data_cost, levels=levels,**para)
core.show_axis(ax)
core.make_spines(ax)
ax.spines['top'].set_visible(True)
ax.spines['right'].set_visible(True)
ax.set_xlabel(core.x_labels["weight_path"])
ax.set_ylabel(core.y_labels["weight_path"])
pass
def plot_phase1(ax):
data_hom = core.get_data('fig5/w_hom_phase1.npy')
data_het = core.get_data('fig5/w_het_phase1.npy')
time_params = core.get_data('fig5/time_parameters.npy').item()
ax.set_xlim([-.5, 5.5])
x_ticks = np.array((0., 2.5, 5.))
ax.xaxis.set_ticks(x_ticks)
weighttrace = np.concatenate(((data_hom, ), (data_het, )))
ax.set_prop_cycle(
cycler('color',
[core.label_to_color["stim"], core.label_to_color["unstim"]]))
core.show_axis(ax)
core.make_spines(ax)
plotf.trace_plot(ax,
weighttrace,
time_params,
linewidth=core.linewidth["normal"])
ax.axvline(x=0,
color=core.label_to_color["aux_lines"],
linewidth=core.linewidth["aux_lines"],
ls="dashed")
ax.text(0., 10., r"$t_0$")
ax.axvline(x=5,
color=core.label_to_color["aux_lines"],
linewidth=core.linewidth["aux_lines"],
ls="dashed")
ax.text(5., 10., r"$t_1$")
ax.set_xlabel(core.x_labels["trace_plot"])
ax.set_ylabel(core.y_labels["trace_plot"])
pass
def plot_learningcurve(ax):
data_nat=core.get_data("fig3/2018-08-24_naturalgradient_KL_errorN100.npy")
data_euc=core.get_data("fig3/2018-08-24_euclideangradient_KL_errorN100.npy")
#data_nat_std=core.get_data("fig3/2018-08-24_naturalgradient_rate_error_std.npy")
#data_euc_std=core.get_data("fig3/2018-08-24_euclideangradient_rate_error_std.npy")
data_time=np.arange(0.,16.,1.)*400./60.#core.get_data("fig3/lc_time.npy")
core.show_axis(ax)
core.make_spines(ax)
ax.semilogy(data_time,data_euc[0:16],color=core.label_to_color["euclidean"],linewidth=core.linewidth["normal"])
ax.semilogy(data_time,data_nat[0:16],color=core.label_to_color["natural"],linewidth=core.linewidth["normal"])
#ax.plot(data_time,data_euc+data_euc_std,color=core.label_to_color["euclidean"],linewidth=core.linewidth["normal"])
#ax.plot(data_time,data_nat+data_nat_std,color=core.label_to_color["natural"],linewidth=core.linewidth["normal"])
#ax.plot(data_time,data_euc-data_euc_std,color=core.label_to_color["euclidean"],linewidth=core.linewidth["normal"])
#ax.plot(data_time,data_nat-data_nat_std,color=core.label_to_color["natural"],linewidth=core.linewidth["normal"])
ax.set_xlabel(core.x_labels["lc"])
ax.set_ylabel(core.y_labels["lc"])
ax.legend(("Euclidean gradient","Natural gradient"),loc="upper_right",fontsize=6.,labelspacing=0.05,borderaxespad=0.,borderpad=0.,handlelength=0.2,frameon=False)
def plot_spikes_after_teach(ax):
data_teach=core.get_data("fig3/2019-02-02_output_spikes_psth_teacher.npy")
ax.scatter(np.nonzero(data_teach)[1],np.nonzero(data_teach)[0],color=core.label_to_color["teacher"],s=0.01)
core.show_axis(ax)
core.make_spines(ax)
ax.set_xlabel(core.x_labels["spikes_teach"])
ax.set_ylabel(core.y_labels["spikes_teach"])
pass
def plot_attenuation_summary_perc(ax):
ax.set_xlim([0.,11.])
data_wchange = core.get_data('fig6/att_weight_change_perc.npy')
data_dist = core.get_data('fig6/distances.npy')
color=[core.label_to_color["attenuation_im"]]*np.shape(data_dist)[0]
color[2]=core.label_to_color["attenuation_ldist"]
color[-3]=core.label_to_color["attenuation_hdist"]
core.show_axis(ax)
core.make_spines(ax)
ax.scatter(data_dist,data_wchange,color=color,s=core.linewidth["scatter"])
ax.get_yaxis().get_major_formatter().set_scientific(False)#prevent exponent notation on axis
ax.yaxis.set_major_formatter(ticker.FormatStrFormatter('%.2f'))#cut y axis ticks 2 digits behind floating point
ticks = np.linspace(0.,50.,2.)
ax.yaxis.set_ticks(ticks)
ax.set_xlabel(r"$d[\mu m]$")
ax.set_ylabel(core.y_labels["weightchange_dend"])
pass
self.model = model_from_json(open(m_name + '.json').read())
self.model.load_weights(m_name + '.h5')
self.model.compile(loss=self.loss,
optimizer=self.opt(lr=self.lr),
metrics=['accuracy'])
if __name__ == "__main__":
nb_epoch = 10
batch_size = 256
k_folds = 5
# get some data
X_train, Y_train, _, _ = core.get_data(60000, 10000, 10)
# split training into folds
X_folds = np.array_split(X_train, k_folds)
Y_folds = np.array_split(Y_train, k_folds)
# create the model
Model = ModelDM(lr=.001,
img_rows=28,
img_cols=28,
img_channels=1,
n_l1=16,
n_l2=None,
dout=0,
n_filt=8,
opt='Adam',
