def main():
"""Create a trained feature set.
1. Extract features from a cropped image set
2. Evaluate features on labelled training data
3. Remove features with high false positive rate
4. Evaluate features again on same data
"""
trainer = Trainer()
detector = Detector(750)
trainer.set_detector(detector)
# feat = load_features('data/gazebo1_932.dat')
# trainer.set_features(feat)
trainer.load_data('train/gazebo/pos_info_PreyEmptyWorld.dat')
trainer.load_data('train/gazebo/pos_info_PurplePrey1.dat')
trainer.load_data('train/gazebo/pos_info_PurplePrey2.dat')
trainer.load_data('train/gazebo/sim2.dat')
# trainer.load_data('train/quadbox/white_courtyard1.dat')
# trainer.load_data('train/quadbox/white_courtyard2.dat')
# trainer.load_data('train/quadbox/black_drone_court.dat')
# trainer.load_data('train/quadbox/pos_info.txt')
# trainer.load_data('train/idea2/pos_info_red-inclass.dat')
# trainer.load_data('train/idea2/red2m.dat')
# trainer.load_data('train/idea2/courtyard323.dat')
# trainer.load_data('train/idea2/orange10am.dat')
# trainer.load_data('train/idea2/orange7_30am.dat')
# trainer.load_data('train/pos_info_Courtyard_multi.dat')
# trainer.load_data('train/idea2/orange9am.dat')
# trainer.subsample_data(2)
trainer.train_and_test(.8)
trainer.evaluate(1)
trainer.feature_selection()
# trainer.evaluate(subsample=0.4)
# trainer.feature_selection()
# trainer.evaluate(subsample=0.7)
trainer.save_features('sim2')
return
trainer.train_and_test(show=False)
trainer.feature_selection()
trainer.evaluate(show=False)
trainer.save_features('ir_2')
return
def run_evaluation(datasets, model, verbose=True):
results = []
if verbose:
print("Training Set Results")
train_acc, train_loss = Trainer.evaluate(
model,
datasets.get_loader(DatasetType.Train),
verbose=verbose,
)
results.append(train_acc)
results.append(train_loss)
time.sleep(0.1)
if verbose:
print("")
print("Validation Set Results")
val_acc, val_loss = Trainer.evaluate(
model,
datasets.get_loader(DatasetType.Validation),
verbose=verbose,
)
results.append(val_acc)
results.append(val_loss)
time.sleep(0.1)
if verbose:
print("")
print("Test Set Results")
test_acc, test_loss = Trainer.evaluate(
model,
datasets.get_loader(DatasetType.Test),
verbose=verbose,
)
results.append(test_acc)
results.append(test_loss)
time.sleep(0.1)
if verbose:
print("")
print("Output for results.md")
print((" {:.3f} |" * len(results)).format(*results))
return results
model.config.length_token)
dl = DataLoader(args, tokenizer)
tr_dataset, val_dataset, test_dataset = dl()
trainer = Trainer(model, tokenizer, args)
if args.load_training_state:
trainer.load_training_state_dict(args.base_dir)
trainer.fit(tr_dataset, val_dataset)
if args.save_pretrained_path:
trainer.model.save_pretrained(
os.path.join(args.base_dir, args.save_pretrained_path))
tst_loss = trainer.evaluate(test_dataset)
wandb.log({"tst_loss": tst_loss})
for mode, dataset in zip(["tr", "val", "tst"],
[tr_dataset, val_dataset, test_dataset]):
out = fetch_translations_and_bleu(model, dataset, tokenizer,
args.iterations, args.B,
args.bleu_num_samples)
data = list(zip(out["src"], out["tgt"], out["pred"]))
wandb.log({
mode + '_bleu':
out["bleu"],
mode + '_predictions':
wandb.Table(data=data, columns=['src', 'tgt', 'tgt_pred'])
})
