def error_plot(dt, prestime, t, ystar, learners, tmin=None, tmax=None, ax=None):
ax = plt.gca() if ax is None else ax
es = [learner['e'] for learner in learners]
if tmin is not None or tmax is not None:
tmin = t[0] if tmin is None else tmin
tmax = t[-1] if tmax is None else tmax
tmask = (t >= tmin) & (t <= tmax)
t = t[tmask]
es = [e[tmask] for e in es]
vsynapse = Alpha(0.01, default_dt=dt)
# esynapse = Alpha(1*prestime, default_dt=dt)
# esynapse = Alpha(1*prestime, default_dt=dt)
esynapse = Alpha(5*prestime, default_dt=dt)
# esynapse = Alpha(20*prestime, default_dt=dt)
# yrms = rms(ystar, axis=1).mean()
yrms = rms(vsynapse.filtfilt(ystar), axis=1).mean()
# yrms = rms(esynapse.filtfilt(ystar), axis=1).mean()
# print(yrms)
for e in es:
# erms = esynapse.filtfilt(rms(e, axis=1) / yrms)
# erms = rms(esynapse.filtfilt(e), axis=1) / yrms
# erms = rms(vsynapse.filtfilt(e), axis=1) / yrms
erms = esynapse.filtfilt(rms(vsynapse.filtfilt(e), axis=1) / yrms)
plt.plot(t, erms)
def output1_plot(dt, t, ystar, learner, tmin=None, tmax=None, ax=None):
ax = plt.gca() if ax is None else ax
y = learner['y']
if tmin is not None or tmax is not None:
tmin = t[0] if tmin is None else tmin
tmax = t[-1] if tmax is None else tmax
tmask = (t >= tmin) & (t <= tmax)
t = t[tmask]
ystar = ystar[tmask]
y = y[tmask]
# dinds = slice(0, 2)
dinds = list(range(2))
vsynapse = Alpha(0.01, default_dt=dt)
ystar = ystar[:, dinds]
# y = y[:, dinds]
y = vsynapse.filtfilt(y[:, dinds])
ax.plot(t, y[:, dinds])
ax.set_color_cycle(None)
ax.plot(t, ystar[:, dinds], ':')
plt.legend(['dim %d' % (i+1) for i in range(len(dinds))], loc='best')
plt.xlim((tmin, tmax))
plt.ylim((-1.5, 1.5))
# plt.xticks((tmin, tmax))
# ax.set_xticks(ax.get_xticks()[::2])
ax.get_xaxis().get_major_formatter().set_useOffset(False)
ax.get_xaxis().get_major_formatter().set_scientific(False)
def output_plot(dt, t, ystar, learners, tmin=None, tmax=None, ax=None):
ax = plt.gca() if ax is None else ax
ys = [learner['y'] for learner in learners]
if tmin is not None or tmax is not None:
tmin = t[0] if tmin is None else tmin
tmax = t[-1] if tmax is None else tmax
tmask = (t >= tmin) & (t <= tmax)
t = t[tmask]
ystar = ystar[tmask]
ys = [y[tmask] for y in ys]
# dinds = slice(0, 2)
dinds = list(range(2))
dstyles = ['-', ':']
# dstyles = ['-', '-.']
vsynapse = Alpha(0.01, default_dt=dt)
ystar = ystar[:, dinds]
ys = [vsynapse.filtfilt(y[:, dinds]) for y in ys]
for k, (dind, dstyle) in enumerate(zip(dinds, dstyles)):
ax.set_color_cycle(None)
ax.plot(t, ystar[:, k], 'k', linestyle=dstyle)
for y in ys:
ax.plot(t, y[:, k], linestyle=dstyle)
plt.ylim((-1.5, 1.5))
def error_layers_plots(dt, t, learners):
vsynapse = Alpha(0.01, default_dt=dt)
for learner in [l for l in learners if 'els' in l]:
plt.figure()
plt.subplot(211)
dind = 0
e = vsynapse.filtfilt(learner['e'])
els = [vsynapse.filtfilt(el) for el in learner['els']]
plt.plot(t, e[:, dind])
[plt.plot(t, el[:, dind]) for el in els]
plt.subplot(212)
plt.plot(t, norm(e, axis=1))
[plt.plot(t, norm(el, axis=1)) for el in els]
plt.show()
def trials_error_plot(prestime, t, ystar, learners):
pdt = 0.01
vsynapse = Alpha(0.02, default_dt=pdt)
plt.figure()
dinds = slice(0, 2)
plt.subplot(211)
plt.plot(t, ystar[:, dinds])
for learner in learners:
y = vsynapse.filtfilt(learner['y'][:, dinds])
plt.plot(t, y)
plt.ylabel('outputs')
plt.subplot(212)
esynapse = Alpha(5 * prestime, default_dt=pdt)
for learner in learners:
e = norm(esynapse.filtfilt(learner['e']), axis=1)
plt.plot(t, e)
plt.ylabel('errors')
def cosyne_plot(dt, prestime, t, ystar, learners, n_test_pre, n_train,
offline_data=None):
vsynapse = Alpha(0.01, default_dt=dt)
learner = learners[-1]
n_show_pre = 10
n_show_post = 10
assert n_show_pre <= n_test_pre
t0 = n_test_pre*prestime
t1 = t0 + n_train*prestime
# t2 = t1 + n_test_post*prestime
tpre0 = 0
tpre1 = tpre0 + n_show_pre*prestime
tpost0 = t1
# tpost0 = t1 + 10*prestime
tpost1 = tpost0 + n_show_post*prestime
plt.figure(figsize=(6.4, 7))
# subplot_shape = (2, 3)
subplot_shape = (3, 3)
ax = plt.subplot2grid(subplot_shape, (0, 2))
# output_plot(dt, t, ystar, learners, tmin=tpre0, tmax=tpre1, ax=ax)
output1_plot(dt, t, ystar, learner, tmin=tpre0, tmax=tpre1, ax=ax)
plt.title('Pre-learning output')
ax = plt.subplot2grid(subplot_shape, (1, 2))
# output_plot(dt, t, ystar, learners, tmin=tpost0, tmax=tpost1, ax=ax)
output1_plot(dt, t, ystar, learner, tmin=tpost0, tmax=tpost1, ax=ax)
plt.xlabel('simulation time [s]')
plt.title('Post-learning output')
hid_spikes0 = learner['hs'][0][:, :40]
hid_spikes1 = learner['hs'][-1][:, :40]
ax = plt.subplot2grid(subplot_shape, (0, 0))
spike_plot(dt, t, hid_spikes0, tmin=tpre0, tmax=tpre1, ax=ax)
plt.title('Pre-learning spikes 1')
ax = plt.subplot2grid(subplot_shape, (1, 0))
spike_plot(dt, t, hid_spikes0, tmin=tpost0, tmax=tpost1, ax=ax)
plt.xlabel('simulation time [s]')
plt.title('Post-learning spikes 1')
ax = plt.subplot2grid(subplot_shape, (0, 1))
spike_plot(dt, t, hid_spikes1, tmin=tpre0, tmax=tpre1, ax=ax)
plt.title('Pre-learning spikes 2')
ax = plt.subplot2grid(subplot_shape, (1, 1))
spike_plot(dt, t, hid_spikes1, tmin=tpost0, tmax=tpost1, ax=ax)
plt.xlabel('simulation time [s]')
plt.title('Post-learning spikes 2')
ax = plt.subplot2grid(subplot_shape, (2, 0), colspan=3)
error_plot(dt, prestime, t, ystar, [learner], tmin=t0, tmax=t1, ax=ax)
if offline_data:
Yrms = rms(offline_data['Y'], axis=1).mean()
eo = offline_data['learners'][-1]['batch_errors'] / Yrms
dto = prestime*offline_data['n_per_batch']
to = t0 + dto*np.arange(len(eo))
tmask = (to >= t0) & (to <= t1)
to = to[tmask]
eo = eo[tmask]
esynapse = Alpha(5*prestime, default_dt=dto)
eo = esynapse.filtfilt(eo)
plt.plot(to, eo, 'k:')
plt.xlim((t0, t1))
plt.legend(['spiking', 'non-spiking'], loc='best')
plt.xlabel('simulation time [s]')
plt.ylabel('normalized RMS error')
plt.title('Error')
plt.tight_layout()
def plot_batches(x, label=None):
filt = Alpha(200, default_dt=n_per_batch)
y = filt.filtfilt(x) if len(x) > 0 else []
batch_inds = n_per_batch * np.arange(len(x))
plt.semilogy(batch_inds, y, label=label)
# --- plot results (rows=[train, test], cols=learners)
rows = 2
cols = len(results[0])
plt.figure(figsize=(7, 6))
# filt = Alpha(3000, default_dt=n_per_batch)
filt = Alpha(10000, default_dt=n_per_batch)
for col in range(cols):
ax = plt.subplot(rows, cols, col + 1)
for i, label in enumerate(f_df_labels):
learner = results[i][col]
x = learner.batch_errors
# y = x
y = filt.filtfilt(x) if len(x) > 0 else []
batch_inds = n_per_batch * np.arange(len(x))
ax.semilogy(batch_inds, y, label=label)
plt.ylim([5e-3, 2e-1])
ax.set_xticklabels([])
if col == 0:
plt.ylabel('train error')
# if col == cols - 1:
# plt.legend(loc=3)
plt.title(learner.name)
filt = Alpha(1, default_dt=1)
for col in range(cols):
plt.subplot(rows, cols, cols + col + 1)
learner.train(epochs, batch_fn, test_set=test_set)
for learner in learners:
print(", ".join("||W%d|| = %0.3f" % (i, norm(w))
for i, w in enumerate(learner.network.weights)))
# --- plots
fig = plt.figure()
rows, cols = 4, 1
filt = Alpha(100, default_dt=n_per_batch)
ax = fig.add_subplot(rows, cols, 1)
ax.set_yscale('log')
for learner in learners:
ax.plot(filt.filtfilt(learner.batch_errors), label=learner.name)
ax = fig.add_subplot(rows, cols, 2)
ax.set_yscale('log')
for learner in learners:
if learner.delta_norms is not None:
ax.plot(filt.filtfilt(learner.delta_norms), label=learner.name)
ax = fig.add_subplot(rows, cols, 3)
for learner in learners:
if getattr(learner, 'bp_angles', None) is not None:
angles = np.array(learner.bp_angles)
ax.plot(filt.filtfilt(angles) * (180 / np.pi), label=learner.name)
ax.set_ylim([0, 90])
ax = fig.add_subplot(rows, cols, 4)
rows = 2
cols = 2
n_batches = len(learners[0].batch_errors)
batch_inds = (n_per_batch / 1000.) * np.arange(n_batches)
epoch_inds = (trainX.shape[0] / 1000.) * np.arange(1, epochs+1)
# - train subplot
ax = fig.add_subplot(rows, cols, 1)
# filt = Alpha(1000, default_dt=n_per_batch)
# filt = Alpha(3000, default_dt=n_per_batch)
filt = Alpha(10000, default_dt=n_per_batch)
for learner in learners:
y = filt.filtfilt(learner.batch_errors)
ax.semilogy(batch_inds, y, label=learner.name)
# plt.ylim([1e-4, 5e-1])
plt.ylim([None, 2e-1])
plt.xlabel('thousands of examples')
plt.ylabel('train error')
plt.legend(loc='best')
# plt.title("Train error")
# - test subplot
ax = fig.add_subplot(rows, cols, 2)
filt = Alpha(0.5, default_dt=1)
# filt = Alpha(1, default_dt=1)
def plot_batches(x, label=None, color=None):
filt = Alpha(10, default_dt=n_per_batch)
y = filt.filtfilt(x) / Yrms
batch_inds = n_per_batch * np.arange(len(x))
plt.plot(batch_inds, y, label=label, color=color)
# epoch_inds = (trainX.shape[0] / 1000.) * np.arange(1, epochs+1)
fig = plt.figure()
rows, cols = 5, 1
# layer_styles = ('-', '-.', ':', '--')
layer_styles = ('-', '--', ':')
# filt = Alpha(10, default_dt=n_per_batch)
# filt = Alpha(20, default_dt=n_per_batch)
# filt = Alpha(30, default_dt=n_per_batch)
filt = Alpha(100, default_dt=n_per_batch)
ax = fig.add_subplot(rows, cols, 1)
ax.set_yscale('log')
batch_errors = np.array([learner.batch_errors for learner in learners])
ax.plot(batch_inds, filt.filtfilt(batch_errors, axis=1).T)
ax = fig.add_subplot(rows, cols, 2)
ax.set_yscale('log')
delta_norms = np.array([
x.delta_norms if x.delta_norms else np.nan * np.ones((n_batches, n_hids))
for x in learners
])
delta_norms = filt.filtfilt(delta_norms, axis=1)
for i in range(n_hids):
ax.set_color_cycle(None)
ax.plot(batch_inds, delta_norms[..., i].T, linestyle=layer_styles[i])
ax = fig.add_subplot(rows, cols, 3)
bp_angles = (180 / np.pi) * np.array([
getattr(learner, 'bp_angles', np.nan * np.ones((n_batches, n_hids)))
n_fdfs = batch_errors.shape[1]
n_learners = batch_errors.shape[2]
assert n_learners == len(learner_names)
rows = 2
cols = n_learners
plt.figure(figsize=(7, 6))
filt = Alpha(3000, default_dt=n_per_batch)
# filt = Alpha(10000, default_dt=n_per_batch)
for col in range(cols):
ax = plt.subplot(rows, cols, col + 1)
# error = filt.filt(batch_errors[:, :, col, :], axis=-1)
error = filt.filtfilt(batch_errors[:, :, col, :], axis=-1)
# batch_inds = n_per_batch * np.arange(error.shape[-1])
batch_inds = (n_per_batch / 1000.) * np.arange(error.shape[-1])
error, batch_inds = error[..., ::10], batch_inds[::10]
sns.tsplot(
data=np.transpose(error, (0, 2, 1)),
time=batch_inds,
condition=f_df_labels,
# err_style='unit_traces',
legend=(col == 0))
ax.set_ylim([5e-3, 2e-1])
ax.set_yscale('log')
ax.set_xticklabels([])
if col == 0: