Hi everyone, I faced this errors when ruining the main.py by sample data files [train 00-04 and test], so, please anyone can suggest to me the way of how to fix this issues?. Thanks for all advance.

Traceback (most recent call last):
File "C:/....../DeepTTE-master-2/main.py", line 172, in
run()
File "C:/........./DeepTTE-master-2/main.py", line 168, in run
evaluate(model, elogger, config['test_set'], save_result=True)
File "C:/......./DeepTTE-master-2/main.py", line 119, in evaluate
data_iter = data_loader.get_loader(input_file, args.batch_size)
File "C:.........\DeepTTE-master-2\data_loader.py", line 105, in get_loader
batch_sampler = BatchSampler(dataset, batch_size)
File "C:.........\DeepTTE-master-2\data_loader.py", line 70, in init
self.count = len(dataset)
File "C:..........\DeepTTE-master-2\data_loader.py", line 29, in len
return len(self.content)
TypeError: object of type 'map' has no len()

It is a very Nice job!

Two question:

1. You said that: "The GPS devices usually generate one record for every fixed length time gap ....... To avoid such case, we resample each historical trajectory such that the distance gap between two consecutive points are around 200 to 400 meters."

Additionally, you said that the distance gaps in the testing data are equal.
Are the distance gaps between two consecutive points in the training data equal? or each distance gap in the training data is different and each one is randomly selected from 200 to 400 meters?

2. You said that: "During the test phase, to make the testing data consistent
   with the training data, we convert a path P to a sequence of
   location points with equal distance gaps."

How do you convert the path P to a sequence of location points with equal distance gaps? Could you generously publish the code that achieves the function? You encourage users to User Your Own Data, but we can not use our own data if we don't know how to convert the path P to a sequence of location points with equal distance gaps.

Thank you for your answer.

I downloaded the data from the website which the author offers us, each record shows us driver_id, latitude, longitude, state and time tap, but I can`t get a trajectory from the records. Can anybody give me some tips?

Thank you for the great work!

I would like to ask why we need to do normalisations of local_dist at models.base.GeoConv.Net.forward() and of dict at models.base.Attr.Net.forward(). Haven't we done so in main.py when we are calling data_loader.get_loader?

I found that the 'time_gap' & 'dist_gap' of the sample data are the values compare with the first values. But the static value of these feature in the config.json are calculate by the interval of these values. And you normalize these feature by the values in the config.json.... So why the 'time_gap' and 'dist_gap' generate by the first values not just by the interval of the sampled trajectory point?

Hello everyone, I have an issue while running the code: first, it seems that everything is fine and the training is running, loss values are printed. But, after some time, I get this error:

Progress 0.07%, average loss 0.07827574014663696 Traceback (most recent call last):
File "/home/s6haoeze/DeepTTE/main.py", line 329, in
run()
File "/home/s6haoeze/DeepTTE/main.py", line 319, in run
train(model, elogger, train_set = config['train_set'], eval_set = config['eval_set'])
File "/home/s6haoeze/DeepTTE/main.py", line 241, in train
_, loss = model.eval_on_batch(attr, traj, config)
File "/home/s6haoeze/DeepTTE/models/DeepTTE.py", line 137, in eval_on_batch
entire_out, (local_out, local_length) = self(attr, traj, config)
File "/home/s6haoeze/anaconda3/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "/home/s6haoeze/DeepTTE/models/DeepTTE.py", line 124, in forward
sptm_s, sptm_l, sptm_t = self.spatio_temporal(traj, attr_t, config)
File "/home/s6haoeze/anaconda3/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "/home/s6haoeze/DeepTTE/models/base/SpatioTemporal.py", line 92, in forward
packed_inputs = nn.utils.rnn.pack_padded_sequence(conv_locs, lens, batch_first = True)
File "/home/s6haoeze/anaconda3/lib/python3.9/site-packages/torch/nn/utils/rnn.py", line 262, in pack_padded_sequence
_VF._pack_padded_sequence(input, lengths, batch_first)
RuntimeError: Length of all samples has to be greater than 0, but found an element in 'lengths' that is <= 0

Although in my dataset, the length of the trajectories are greater than 18, it seems that somewhere in the code some length is interpreted as shorter than 2 (the line lens = [x - self.kernel_size + 1 for x in traj['lens']] in SpatioTemporal.py). What can be done? Thanks a lot!

If my dataset doesn't contain the state attribute, how can I still train my model? Attr.py doesn't seem to have any clues. My data doesn't have a 'states' attribute, do I need to add one and set it to blanks? Right now when I try to train the model, I get a KeyError in the data loader. Removing 'states' from the keys to use then yields an index out of range error in Attr.py.

Would love to use this in a notebook environment via !pip install. Would you be able to a add a setup.py?

I‘m appreciate if you could provide the baselines of DeepTTE you referred to in the paper, such as MLPTTE, GBDT.
email address:[email protected]
Thank you anyway!

I‘m appreciate if you could provide the work of the baselines you refered in paper.
email address:[email protected]
thankyou anyway!

When I run train.py, it reported a mistake about no run._log.log file? Can you help me solve this problem, thanks.
Error details：
(torch2.0) nlg@ubuntu:/PycharmProj2/DeepTTE-master$ python main.py --task train --batch_size 10 --result_file ./result/deeptte.res --pooling_method attention --kernel_size 3 --alpha 0.1 --log_file run_log
Traceback (most recent call last):
File "main.py", line 161, in
run()
File "main.py", line 148, in run
elogger = logger.Logger(args.log_file)
File "/home/nlg/PycharmProj2/DeepTTE-master/logger.py", line 5, in init
self.file = open('./logs/{}.log'.format(exp_name), 'w')
IOError: [Errno 2] No such file or directory: './logs/run_log.log'
(torch2.0) nlg@ubuntu:/PycharmProj2/DeepTTE-master$

I found dist is normalized in data_loader and Attr_layer
Attr layer:
dist = utils.normalize(attr['dist'], 'dist')
em_list.append(dist.view(-1, 1))

Data loader:
for key in stat_attrs:
x = torch.FloatTensor([item[key] for item in data])
attr[key] = utils.normalize(x, key)

which may lead some problem

Hello guys and good job for the great work. I was reading the GeoConv.py and I am wondering why a Linear(4, 16) is used before the Conv1D?

Why would someone want to span more dimensions and transform the given vector(loc_i) to a 16x4 one? Is there any intuitive explanation for this?

Thanks for your time

Hi, thank you for the wonderful job. I think this is the first work I have seen that brings deep learning into the area of travel time estimation. I just have a question and I wonder if I could ask you about it. It is as follows:

Regarding the resampling of the data, I am curious how is the distance gap between two points calculated? Is it the on-road distance between the two points, or great circle distance based on the GPS coordinates of the two points?

Thank you.

You lose model.cuda() in main.py when training.

What is the weight file?

FileNotFoundError: [Errno 2] No such file or directory: '/saved_weights'

Hi Team,

First off, great work. This has been really helpful.
Would be great to get your inputs on a couple of points.

1. While trying to estimate travel time between a start point and an endpoint, is it necessary to provide
   the expected trajectory to be followed by the vehicle.
   Any workaround for that is possible?

2. I am trying this out for long distances.
   Say we have A- startpoint ,B- endpoint , total distance 500kms
   At the starting of the trip we provide the expected trajectory between A and B to estimate the travel time.
   Then if the vehicle covers says 150km and is at an intermediate point C .
   Now is it recommended to find the new ETA that we use C as the starting point?

Is there any way that we also incorporate the current driving pattern from A to C to predict the driving pattern from C to B ?

If the distance between two consecutive track points must be around 200 and 400 meters, points that move very slow or stay points are thrown away. So does it mean that the model may lose the ability to predict roads that suffer from traffic jam? Because many points are abandoned.

Thanks for yours paper work!
I'm a college student who is studying the traffic time prediction. The complete data link mentioned in the 'read_me.md' is failed, it is a pity~~
So can you tell me how to download the complete data? Thanks a lot

I ran the code on the pytorch-cpu-windows-py37 platform successfully, but when tring to use GPU, I have some problem. While training, the first epoch was normal, but an error occured when the second epoch started.

I dont understand why "cudnn RNN backward can only be called in training mode", Im sure I was training the model, because the "task" argument was "train". Error occurs only on GPU with the same code.

Hi,

Can you please release some test data?

Thanks.

i have tried to run it at win10/cuda9.1.85/anaconda4.5.4/python3.6.5 for some time, but failed. After dealing with some little bugs because of python version from 2 to 3, i'm struggled with the problem:

E:\workbench\DeepTTE-master>python main.py --task train --batch_size 10 --result_file ./result/deeptte.res --pooling_method attention --kernel_size 3 --alpha 0.1 --log_file run_log
Training on epoch 0
Train on file train_00
Traceback (most recent call last):
File "main.py", line 161, in
run()
File "main.py", line 151, in run
train(model, elogger, train_set = config['train_set'], eval_set = config['eval_set'])
File "main.py", line 69, in train
for idx, (attr, traj) in enumerate(data_iter):
File "D:\Anaconda3\lib\site-packages\torch\utils\data\dataloader.py", line 451, in iter
return _DataLoaderIter(self)
File "D:\Anaconda3\lib\site-packages\torch\utils\data\dataloader.py", line 239, in init
w.start()
File "D:\Anaconda3\lib\multiprocessing\process.py", line 105, in start
self._popen = self._Popen(self)
File "D:\Anaconda3\lib\multiprocessing\context.py", line 223, in _Popen
return _default_context.get_context().Process._Popen(process_obj)
File "D:\Anaconda3\lib\multiprocessing\context.py", line 322, in _Popen
return Popen(process_obj)
File "D:\Anaconda3\lib\multiprocessing\popen_spawn_win32.py", line 65, in init
reduction.dump(process_obj, to_child)
File "D:\Anaconda3\lib\multiprocessing\reduction.py", line 60, in dump
ForkingPickler(file, protocol).dump(obj)
AttributeError: Can't pickle local object 'MySet.init..'

can anybody give some suggestions??

Hi, I run the command in readme.
And I get following error, can you help me?

Traceback (most recent call last):
File "main.py", line 161, in
run()
File "main.py", line 151, in run
train(model, elogger, train_set = config['train_set'], eval_set = config['eval_set'])
File "main.py", line 77, in train
loss.backward()
File "/home/dsc/.local/lib/python2.7/site-packages/torch/tensor.py", line 93, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph)
File "/home/dsc/.local/lib/python2.7/site-packages/torch/autograd/init.py", line 89, in backward
allow_unreachable=True) # allow_unreachable flag
RuntimeError: backward_input can only be called in training mode

Hello All!

I am a complete beginner here and a Windows user. When I try to run the code, either by running the main file or as
python main.py --task train --batch_size 10 --result_file ./result/deeptte.res --pooling_method attention --kernel_size 3 --alpha 0.1 --log_file run_log
provided in the "README.md" I get this error: "No module named 'DeepTTE'"

Can someone please comment what I could be doing wrong.