def model_inference(sess, pred, inputs, batch_size, n_his, n_pred, step_idx, min_va_val, min_val):
'''
Model inference function.
:param sess: tf.Session().
:param pred: placeholder.
:param inputs: instance of class Dataset, data source for inference.
:param batch_size: int, the size of batch.
:param n_his: int, the length of historical records for training.
:param n_pred: int, the length of prediction.
:param step_idx: int or list, index for prediction slice.
:param min_va_val: np.ndarray, metric values on validation set.
:param min_val: np.ndarray, metric values on test set.
'''
x_val, x_test, x_stats = inputs.get_data('val'), inputs.get_data('test'), inputs.get_stats()
if n_his + n_pred > x_val.shape[1]:
raise ValueError(f'ERROR: the value of n_pred "{n_pred}" exceeds the length limit.')
y_val,y2, len_val = multi_pred(sess, pred, x_val, batch_size, n_his, n_pred, step_idx)
evl_val,v,v_ = evaluation(x_val[0:len_val, step_idx + n_his, :, :], y_val, x_stats)
# chks: indicator that reflects the relationship of values between evl_val and min_va_val.
chks = evl_val < min_va_val
# update the metric on test set, if model's performance got improved on the validation.
if sum(chks):
min_va_val[chks] = evl_val[chks]
y_pred,y2, len_pred = multi_pred(sess, pred, x_test, batch_size, n_his, n_pred, step_idx)
evl_pred,v,v_ = evaluation(x_test[0:len_pred, step_idx + n_his, :, :], y_pred, x_stats)
min_val = evl_pred
return min_va_val, min_val
def model_inference(exe, gw, gf, program, pred, inputs, args, step_idx,
min_va_val, min_val):
"""inference model"""
x_val, x_test, x_stats = inputs.get_data('val'), inputs.get_data(
'test'), inputs.get_stats()
if args.n_his + args.n_pred > x_val.shape[1]:
raise ValueError(
f'ERROR: the value of n_pred "{args.n_pred}" exceeds the length limit.'
)
# y_val shape: [n_pred, len(x_val), n_route, C_0)
y_val, len_val = multi_pred(exe, gw, gf, program, pred, \
x_val, args.batch_size, args.n_his, args.n_pred, step_idx)
evl_val = evaluation(x_val[0:len_val, step_idx + args.n_his, :, :],
y_val[step_idx], x_stats)
# chks: indicator that reflects the relationship of values between evl_val and min_va_val.
chks = evl_val < min_va_val
# update the metric on test set, if model's performance got improved on the validation.
if sum(chks):
min_va_val[chks] = evl_val[chks]
y_pred, len_pred = multi_pred(exe, gw, gf, program, pred, \
x_test, args.batch_size, args.n_his, args.n_pred, step_idx)
evl_pred = evaluation(x_test[0:len_pred, step_idx + args.n_his, :, :],
y_pred[step_idx], x_stats)
min_val = evl_pred
return min_va_val, min_val
def model_test(inputs, batch_size, n_his, n_pred, inf_mode, load_path='./output/models/'):
'''
Load and test saved model from the checkpoint.
:param inputs: instance of class Dataset, data source for test.
:param batch_size: int, the size of batch.
:param n_his: int, the length of historical records for training.
:param n_pred: int, the length of prediction.
:param inf_mode: str, test mode - 'merge / multi-step test' or 'separate / single-step test'.
:param load_path: str, the path of loaded model.
'''
start_time = time.time()
model_path = tf.train.get_checkpoint_state(load_path).model_checkpoint_path
test_graph = tf.Graph()
with test_graph.as_default():
saver = tf.train.import_meta_graph(pjoin(f'{model_path}.meta'))
with tf.Session(graph=test_graph) as test_sess:
saver.restore(test_sess, tf.train.latest_checkpoint(load_path))
print('>> Loading saved model from {model_path} ...')
pred = test_graph.get_collection('y_pred')
if inf_mode == 'sep':
# for inference mode 'sep', the type of step index is int.
step_idx = n_pred - 1
tmp_idx = [step_idx]
elif inf_mode == 'merge':
# for inference mode 'merge', the type of step index is np.ndarray.
step_idx = tmp_idx = np.arange(3, n_pred + 1, 3)-1
else:
raise ValueError(f'ERROR: test mode "{inf_mode}" is not defined.')
x_test, x_stats = inputs.get_data('test'), inputs.get_stats()
io.savemat('xtest.mat', {'xtest': x_test})
y_test, y2,len_test = multi_pred(test_sess, pred, x_test, batch_size, n_his, n_pred, step_idx)
io.savemat('ytest.mat', {'ytest': y_test})
io.savemat('y2.mat', {'y2': y2})
evl1,v1,v1_ = evaluation(x_test[0:len_test, 0 + n_his, :, :], y2[0, :, :, :], x_stats)
evl2,v2,v2_ = evaluation(x_test[0:len_test, 1 + n_his, :, :], y2[2, :, :, :], x_stats)
evl3,v3,v3_= evaluation(x_test[0:len_test, 2 + n_his, :, :], y2[2, :, :, :], x_stats)
evl,v,v_ = evaluation(x_test[0:len_test, step_idx + n_his, :, :], y_test, x_stats)
m=x_test[0:len_test, step_idx + n_his, :, :]
io.savemat('xtest1.mat',{'m':m})
io.savemat('evl1.mat', {'evl1': evl1})
io.savemat('evl2.mat', {'evl2': evl2})
io.savemat('evl3.mat', {'evl3': evl3})
io.savemat('v1.mat', {'v1': v1})
io.savemat('v2.mat', {'v2': v2})
io.savemat('v3.mat', {'v3': v3})
io.savemat('v1_.mat', {'v1_': v1_})
io.savemat('v2_.mat', {'v2_': v2_})
io.savemat('v3_.mat', {'v3_': v3_})
for ix in [2]:
te = evl[ix - 2:ix + 1]
print(f'Time Step {ix + 1}: MAPE {te[0]:7.3%}; MAE {te[1]:4.3f}; RMSE {te[2]:6.3f}.')
print(f'Model Test Time {time.time() - start_time:.3f}s')
print('Testing model finished!')
def model_test(inputs,
batch_size,
n_his,
n_pred,
inf_mode,
load_path='./output/models/'):
'''
Load and test saved model from the checkpoint.
:param inputs: instance of class Dataset, data source for test.
:param batch_size: int, the size of batch.
:param n_his: int, the length of historical records for training.
:param n_pred: int, the length of prediction.
:param inf_mode: str, test mode - 'merge / multi-step test' or 'separate / single-step test'.
:param load_path: str, the path of loaded model.
'''
start_time = time.time()
model_path = tf.train.get_checkpoint_state(load_path).model_checkpoint_path
test_graph = tf.Graph()
with test_graph.as_default():
saver = tf.train.import_meta_graph(pjoin(
'{0}.meta'.format(model_path)))
with tf.Session(graph=test_graph) as test_sess:
saver.restore(test_sess, tf.train.latest_checkpoint(load_path))
print('>> Loading saved model from {0} ...'.format(model_path))
pred = test_graph.get_collection('y_pred')
if inf_mode == 'sep':
# for inference mode 'sep', the type of step index is int.
step_idx = n_pred - 1
tmp_idx = [step_idx]
elif inf_mode == 'merge':
# for inference mode 'merge', the type of step index is np.ndarray.
step_idx = tmp_idx = np.arange(3, n_pred + 1, 3) - 1
else:
raise ValueError(
'ERROR: test mode "{0}" is not defined.'.format(inf_mode))
x_test, x_stats = inputs.get_data('test'), inputs.get_stats()
y_test, len_test = multi_pred(test_sess, pred, x_test, batch_size,
n_his, n_pred, step_idx)
evl = evaluation(x_test[0:len_test, step_idx + n_his, :, :], y_test,
x_stats)
for ix in tmp_idx:
#te = evl[ix - 2:ix + 1]
te = evl
print(
f'Time Step {ix + 1}: MAPE {te[0]:7.3%}; MAE {te[1]:4.3f}; RMSE {te[2]:6.3f}.'
)
print(f'Model Test Time {time.time() - start_time:.3f}s')
print('Testing model finished!')
def model_test(inputs, batch_size, n_his, n_pred, inf_mode, load_path='./output/models/'):
'''
Load and test saved model from the checkpoint.
:param inputs: instance of class Dataset, data source for test.
:param batch_size: int, the size of batch.
:param n_his: int, the length of historical records for training.
:param n_pred: int, the length of prediction.
:param inf_mode: str, test mode - 'merge / multi-step test' or 'separate / single-step test'.
:param load_path: str, the path of loaded model.
'''
start_time = time.time()
model_path = tf.train.get_checkpoint_state(load_path).model_checkpoint_path
test_graph = tf.Graph()
with test_graph.as_default():
saver = tf.train.import_meta_graph(pjoin(f'{model_path}.meta'))
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.25 # 占用GPU25%的显存
config.gpu_options.allow_growth = True
with tf.Session(graph=test_graph,config=config) as test_sess:
saver.restore(test_sess, tf.train.latest_checkpoint(load_path))
print(f'>> Loading saved model from {model_path} ...')
pred = test_graph.get_collection('y_pred')
if inf_mode == 'sep':
# for inference mode 'sep', the type of step index is int.
step_idx = n_pred - 1
tmp_idx = [step_idx]
elif inf_mode == 'merge':
# for inference mode 'merge', the type of step index is np.ndarray.
# step_idx = tmp_idx = np.arange(3, n_pred + 1, 3) - 1
step_idx = tmp_idx = np.array([1,2,3,4,5,6,7,8])-1
else:
raise ValueError(f'ERROR: test mode "{inf_mode}" is not defined.')
x_test, x_stats = inputs.get_data('test'), inputs.get_stats()
y_test, len_test = multi_pred(test_sess, pred, x_test, batch_size, n_his, n_pred, step_idx)
evl = evaluation(x_test[0:len_test, step_idx + n_his, :, :], y_test, x_stats)
print(f'evl.shape={evl.shape}')
cnt=0;
for ix in tmp_idx:
# te = evl[ix - 2:ix + 1]
te = evl[cnt:cnt+3]
cnt+=3;
print(f'Time {(ix + 1)*15}: MAPE {te[0]:7.3%}; MAE {te[1]:4.3f}; RMSE {te[2]:6.3f}.')
print(f'Model Test Time {time.time() - start_time:.3f}s')
print('Testing model finished!')
def model_test(exe, gw, gf, program, pred, inputs, args):
"""test model"""
if args.inf_mode == 'sep':
# for inference mode 'sep', the type of step index is int.
step_idx = args.n_pred - 1
tmp_idx = [step_idx]
elif args.inf_mode == 'merge':
# for inference mode 'merge', the type of step index is np.ndarray.
step_idx = tmp_idx = np.arange(3, args.n_pred + 1, 3) - 1
print(step_idx)
else:
raise ValueError(f'ERROR: test mode "{args.inf_mode}" is not defined.')
x_test, x_stats = inputs.get_data('test'), inputs.get_stats()
y_test, len_test = multi_pred(exe, gw, gf, program, pred, \
x_test, args.batch_size, args.n_his, args.n_pred, step_idx)
# save result
gt = x_test[0:len_test, args.n_his:, :, :].reshape(-1, args.n_route)
y_pred = y_test.reshape(-1, args.n_route)
city_df = pd.read_csv(args.city_file)
city_df = city_df.drop(0)
np.savetxt(os.path.join(args.output_path, "groundtruth.csv"),
gt.astype(np.int32),
fmt='%d',
delimiter=',',
header=",".join(city_df['city']))
np.savetxt(os.path.join(args.output_path, "prediction.csv"),
y_pred.astype(np.int32),
fmt='%d',
delimiter=",",
header=",".join(city_df['city']))
for i in range(step_idx + 1):
evl = evaluation(x_test[0:len_test, step_idx + args.n_his, :, :],
y_test[i], x_stats)
for ix in tmp_idx:
te = evl[ix - 2:ix + 1]
print(
f'Time Step {i + 1}: MAPE {te[0]:7.3%}; MAE {te[1]:4.3f}; RMSE {te[2]:6.3f}.'
)
def model_test(inputs,
batch_size,
n_his,
n_pred,
inf_mode,
load_path='./output/models/'):
'''
Load and test saved model from the checkpoint.
:param inputs: instance of class Dataset, data source for test.
:param batch_size: int, the size of batch.
:param n_his: int, the length of historical records for training.
:param n_pred: int, the length of prediction.
:param inf_mode: str, test mode - 'merge / multi-step test' or 'separate / single-step test'.
:param load_path: str, the path of loaded model.
'''
start_time = time.time()
model_path = tf.train.get_checkpoint_state(load_path).model_checkpoint_path
test_graph = tf.Graph()
with test_graph.as_default():
saver = tf.train.import_meta_graph(pjoin(f'{model_path}.meta'))
with tf.Session(graph=test_graph) as test_sess:
saver.restore(test_sess, tf.train.latest_checkpoint(load_path))
print(f'>> Loading saved model from {model_path} ...')
pred = test_graph.get_collection('y_pred')
if inf_mode == 'sep':
# for inference mode 'sep', the type of step index is int.
step_idx = n_pred - 1
tmp_idx = [step_idx]
elif inf_mode == 'merge':
# for inference mode 'merge', the type of step index is np.ndarray.
step_idx = tmp_idx = np.arange(3, n_pred + 1, 3) - 1
else:
raise ValueError(f'ERROR: test mode "{inf_mode}" is not defined.')
x_test, x_stats = inputs.get_data('test'), inputs.get_stats()
y_test, len_test = multi_pred(test_sess, pred, x_test, batch_size,
n_his, n_pred, step_idx)
v_ = z_inverse(y_test, x_stats['mean'], x_stats['std'])
print("v_ shape", v_.shape)
cnt = 0
with open('result.csv', 'w') as csvFile:
writer = csv.writer(csvFile)
writer.writerow(["Id", "Expected"])
for i in range(7):
for j in range(12):
idx = 268 * i + 24 * j
r = v_[:, idx, :, :]
#print(idx, r[0].flatten())
for k in range(3):
for p in range(228):
_r = r[k].flatten()
name = str(cnt - 4) + "_" + str(15 * k +
15) + "_" + str(p)
if (cnt >= 4):
writer.writerow([name, _r[p]])
cnt += 1
print("cnt", cnt)
#writer.writerow(v_)
csvFile.close()
evl = evaluation(x_test[0:len_test, step_idx + n_his, :, :], y_test,
x_stats)
for ix in tmp_idx:
te = evl[ix - 2:ix + 1]
print(
f'Time Step {ix + 1}: MAPE {te[0]:7.3%}; MAE {te[1]:4.3f}; RMSE {te[2]:6.3f}.'
)
print(f'Model Test Time {time.time() - start_time:.3f}s')
print('Testing model finished!')
