def ident_plant(_input, _output, expe_name, epochs=50, force_train=False, display_training_history=False):
filename = "/tmp/plant_id__planar_mip.h5"
if force_train or not os.path.isfile(filename):
plant_i = keras.layers.Input((5,), name ="plant_i") # x1_k, x2_k, x3_k, x4_k, u_k
if 1:
plant_l = keras.layers.Dense(4, activation='linear', kernel_initializer='uniform', use_bias=False, name="plant")
plant_o = plant_l(plant_i)
else:
plant_l1 = keras.layers.Dense(8, activation='relu', kernel_initializer='uniform', use_bias=True, name="plant1")
plant_l2 = keras.layers.Dense(12, activation='relu', kernel_initializer='uniform', use_bias=True, name="plant2")
plant_l3 = keras.layers.Dense(4, activation='linear', kernel_initializer='uniform', use_bias=True, name="plant3")
plant_o = plant_l3(plant_l2(plant_l1(plant_i)))
plant_ann = keras.models.Model(inputs=plant_i, outputs=plant_o)
plant_ann.compile(loss='mean_squared_error', optimizer='adam', metrics=['accuracy'])
early_stopping = keras.callbacks.EarlyStopping(monitor='val_loss', patience=8)
history = plant_ann.fit(_input, _output, epochs=epochs, batch_size=32, verbose=1, shuffle=True, validation_split=0.1, callbacks=[early_stopping])
if display_training_history:
margins = (0.04, 0.07, 0.98, 0.93, 0.27, 0.2)
figure = plu.prepare_fig(figsize=(20.48, 7.68), margins=margins)
ax = plt.subplot(1,2,1); plu.decorate(ax, title='loss'); plt.plot(history.history['loss'])
ax = plt.subplot(1,2,2); plu.decorate(ax, title='accuracy'); plt.plot(history.history['acc'])
#plu.save_if('../docs/plots/plant_id__mip_simple__{}_training_history.png'.format(expe_name))
plant_ann.save(filename)
else:
plant_ann = keras.models.load_model(filename)
return plant_ann
def plot_training_dataset(_ann_in, _ann_out):
fig = pu.prepare_fig(window_title='Training dataset')
names = '$y_k$', '$y_{k-1}$', '$u_{k-1}$', '$u_k$', '$y_{k+1}$'
for i in range(4):
ax = plt.subplot(1, 5, i + 1)
plt.hist(_ann_in[:, i])
pu.decorate(ax, title=names[i])
ax = plt.subplot(1, 5, 5)
plt.hist(_ann_out)
pu.decorate(ax, title=names[4])
def plot_dataset(time, X, U, exp_name):
margins = (0.04, 0.07, 0.98, 0.93, 0.27, 0.2)
figure = plu.prepare_fig(figsize=(20.48, 7.68), margins=margins)
plots = [('$x$', 'm', X[:, 0]), ('$\\theta$', 'deg', np.rad2deg(X[:, 1])),
('$\dot{x}$', 'm/s', X[:, 2]),
('$\dot{\\theta}$', 'deg/s', np.rad2deg(X[:, 3])),
('$\\tau$', 'N', U[:, 0])]
for i, (_ti, _un, _d) in enumerate(plots):
ax = plt.subplot(1, 5, i + 1)
plt.hist(_d, bins=100)
plu.decorate(ax, title=_ti, xlab=_un)
def simulate_plant_and_ann(plant, ann_plant):
time = np.arange(0., 25.05, plant.dt)
sp, yp, ya = scipy.signal.square(0.25 * time), np.zeros(
len(time)), np.zeros(len(time))
for k in range(1, len(time) - 1):
yp[k + 1] = plant.io_dyn(yp[k], yp[k - 1], sp[k], sp[k - 1])
ya[k + 1] = ann_plant.predict(ya[k], ya[k - 1], sp[k - 1], sp[k])
fig = pu.prepare_fig(window_title='Time simulation')
ax = plt.gca()
plt.plot(time, sp, label='sp')
plt.plot(time, yp, label='plant')
plt.plot(time, ya, label='ann')
pu.decorate(ax, title='$y$', xlab='time in s', ylab='m', legend=True)
plt.savefig(ut.cs_asset('docs/plots/so_lti__ident__io__keras.png'))
def plot_training(ann):
fig = pu.prepare_fig(window_title='Training history')
_h = ann.history.history
plt.plot(_h['loss'], label='loss')
plt.plot(_h['val_loss'], label='val_loss')
pu.decorate(plt.gca(), 'loss', xlab='epochs', legend=True)