def run(chosen_model, hp=None):
fingers = hparams.pop('fingers')
columns = sum(
list(map(lambda f: list(range(6 * f, 6 * (f + 1))), fingers)), [])
x_train = np.array(x_traing[:, :, columns])
x_test = np.array(x_testg[:, :, columns])
x_train = chosen_model.feature_extraction(x_train)
x_test = chosen_model.feature_extraction(x_test)
m = chosen_model.train(x_train, y_traing, class_count, **hp)
fig, ax = plt.subplots(1,
1,
figsize=(1.3 * class_count, 1.3 * class_count))
# Actual prediction results
y_hat = chosen_model.test(m, x_test)
cm = u_metrics.create_confusion_matrix(class_count, y_hat, y_testg)
u_plot.plot_confusion_matrix(cm, ax=ax, cbar=False)
class_names = [
'Null', 'Hand sw. left', 'Hand sw. right', 'Pinch in', 'Pinch out',
'Thumb d. tap', 'Grab', 'Ungrab', 'Page flip', 'Peace', 'Metal'
]
class_namese = [''] * 10
ax.set_xticklabels(class_names, rotation=90)
ax.invert_yaxis()
ax.set_yticklabels(class_names, rotation=0)
fig.show()
'dropout': dropout,
})
model.apply(init_weights)
trainer = Trainer(
gpus=1,
min_epochs=12,
max_epochs=100,
progress_bar_refresh_rate=0,
)
trainer.fit(model)
xs = model.tds[:][0]
ys = model.tds[:][1]
y_hat = common_test(model, xs.cpu().numpy())
fig, ax = plt.subplots(1, 1)
cm_train = u_metrics.create_confusion_matrix(
classes, y_hat, ys)
u_plot.plot_confusion_matrix(cm_train, title=f'All', ax=ax)
plt.show()
print('Dist: ',
np.histogram(ys.cpu().numpy(), bins=classes)[0])
acc_train, recall_train, f1_train = u_metrics.get_metrics(
y_hat,
ys.cpu().numpy())
print(
f'layers: {lstm_layers:6} hidden: {hidden:6} dout: {dropout:6} LR: {lr:6} Acc: {acc_train:6.3f}, Rec: {recall_train:6.3f}, F1: {f1_train:6.3f}'
)
def create_results(model,
test_f,
x_train,
x_test,
y_train,
y_test,
class_count,
dataset,
model_name,
show=False,
plot_y_dist=False):
logger = get_logger()
fig, (ax1, ax2) = plt.subplots(1,
2,
figsize=(1.6 * class_count,
0.8 * class_count))
# First re-test on the training data
y_hat_train = test_f(model, x_train)
cm_train = u_metrics.create_confusion_matrix(class_count, y_hat_train,
y_train)
u_plot.plot_confusion_matrix(cm_train, title=f'Training', ax=ax1)
acc_train, recall_train, f1_train = u_metrics.get_metrics(
y_hat_train, y_train)
add_log(
model_name,
f'TAcc: {acc_train:.3f} | TRec: {recall_train:.3f} | TF1: {f1_train:.3f} || ',
newline=False)
# Actual prediction results
y_hat = test_f(model, x_test)
cm = u_metrics.create_confusion_matrix(class_count, y_hat, y_test)
u_plot.plot_confusion_matrix(cm, title=f'Test', ax=ax2)
acc, recall, f1 = u_metrics.get_metrics(y_hat, y_test)
add_log(model_name, f'Acc: {acc:.3f} | Rec: {recall:.3f} | F1: {f1:.3f}')
fig.suptitle(f'Model: {model_name}, Dataset: {dataset.value}', fontsize=16)
fig.subplots_adjust(top=.9)
logger.experiment.add_figure('confusion_matrix', fig)
if show:
fig.show()
if plot_y_dist:
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(14, 7))
u_plot.plot_class_histogram(y_train,
ax=ax1,
title='Class distribution on train')
u_plot.plot_class_histogram(y_test,
ax=ax2,
title='Class distribution on test')
fig.suptitle(f'Model: {model_name}, dataset: {dataset.value}',
fontsize=16)
fig.subplots_adjust(top=.9)
logger.experiment.add_figure('y_dist', fig)
if show:
fig.show()
logger.save()