def predict(self, input):
Predictor.validateInput(input)
points = len(input[0])
if (points < lstmInputPoints):
adapted = np.zeros((1, lstmInputPoints, 2))
for i in range(0, points):
adapted[0, i] = input[0, i]
input = adapted
if (points > lstmInputPoints):
adapted = input[0][points - lstmInputPoints:]
adapted = adapted.reshape(1, lstmInputPoints, 2)
input = adapted
return self.model.predict(input, verbose=0)
def predict(self, input):
Predictor.validateInput(input)
input = self.adapt(input)
return self.model.predict(input, verbose=0)
from argparse import ArgumentParser
import json
parser = ArgumentParser(description='Predict translation')
parser.add_argument('--source', type=str)
parser.add_argument('--config', type=str, required=True)
parser.add_argument('--checkpoint', type=str)
parser.add_argument('--num_candidates', type=int, default=1)
args = parser.parse_args()
with open(args.config) as f:
config = json.load(f)
print('Constructing dictionaries...')
source_dictionary = IndexDictionary.load(config['data_dir'], mode='source', vocabulary_size=config['vocabulary_size'])
target_dictionary = IndexDictionary.load(config['data_dir'], mode='target', vocabulary_size=config['vocabulary_size'])
print('Building model...')
model = build_model(config, source_dictionary.vocabulary_size, target_dictionary.vocabulary_size)
predictor = Predictor(
preprocess=IndexedInputTargetTranslationDataset.preprocess(source_dictionary),
postprocess=lambda x: ' '.join([token for token in target_dictionary.tokenify_indexes(x) if token != '<EndSent>']),
model=model,
checkpoint_filepath=args.checkpoint
)
for index, candidate in enumerate(predictor.predict_one(args.source, num_candidates=args.num_candidates)):
print(f'Candidate {index} : {candidate}')
from predictors import Predictor
from suggestions import Suggestion
import numpy as np
if __name__ == "__main__":
age = 28
sex = 1
height = 180
weight = 75
lean_factor = 0.9
activity_factor = 1.5
food_intake = np.array([[101, 500, 1], [102, 120, 2]])
activity = np.array([[101, height, weight], [105, height, weight]])
p = Predictor()
s = Suggestion()
suggestions, calories_requried = p.suggestions_predictor(
age, sex, height, weight, lean_factor, activity_factor, food_intake,
activity)
# print(len(suggestions))
# print(suggestions)
s1 = s.exercise_suggestion(suggestions[0, 0], suggestions[0, 1])
s2 = s.food_rec(suggestions[0, 2], calories_requried)
print(s1)
print(s2)
# age = input("Age: ")
# sex = input("Sex: ")
# height = input("Height in centimeters: ")
# weight = input("Weight in kilograms: ")
# lean_factor = input("Lean factor: ")
# -------------------------------- predictor training scripts --------------------------------------------------------
if __name__ == '__main__':
step_list = [1, 8, 16, 24, 50, 100]
print('training start now: ')
for n_step in step_list:
i = -1
env = EnvironmentState()
state_set = StateDataset(env,
skip_step=n_step,
size=7000000,
random_torque=True,
remove_torque=False)
state_set_loader = DataLoader(state_set, batch_size=1024)
predictor = Predictor(state_set.output_size, 2)
predictor.stepper.recurrent_step = 0
avm = AverageMeter()
with Timer():
for s in state_set_loader:
i += 1
d = s['s1'][:, 12:14] - s['s0'][:, 12:14] # with torques
r, loss = predictor.optimize(s['s0'], d)
avm.log(loss)
if i % 10 == 0:
print(
f'epoch {i}: trn loss {avm.value:.4f} {avm.std:.4f}, rmse {avm.value**0.5:.4f}, '
f'pred to stat error ratio: {torch.mean(torch.abs(r / d)):.4f} '
f'max_d {torch.max(d):.2f} {torch.min(d):.2f}, max_r {torch.max(r):.2f} {torch.min(r):.2f}'
)
mode='source',
vocabulary_size=config['vocabulary_size'])
target_dictionary = IndexDictionary.load(
config['data_dir'],
mode='target',
vocabulary_size=config['vocabulary_size'])
print('Building model...')
model = build_model(config, source_dictionary.vocabulary_size,
target_dictionary.vocabulary_size)
predictor = Predictor(
preprocess=IndexedInputTargetTranslationDataset.preprocess(
source_dictionary),
postprocess=lambda x: ' '.join([
token for token in target_dictionary.tokenify_indexes(x)
if token != '<EndSent>'
]),
model=model,
checkpoint_filepath=args.checkpoint,
beam_size=3)
timestamp = datetime.now()
if args.save_result is None:
eval_filepath = 'logs/eval-{config}-time={timestamp}.csv'.format(
config=args.config.replace('/', '-'),
timestamp=timestamp.strftime("%Y_%m_%d_%H_%M_%S"))
else:
eval_filepath = args.save_result
evaluator = Evaluator(predictor=predictor, save_filepath=eval_filepath)