Ritvik Vij

icon indicating a website link [email protected]

icon company Amazon icon location New Delhi, India icon location Applied Scientist @ Amazon. Dual Degree(B.Tech + M.Tech) in Computer science, IIT Delhi'22. Highly interested in ML and Vision

Results 10 comments of Ritvik Vij

some question about calculating probalitility

In practice, using the intensity function doesn't work very well. Utilising a non-linearity of the model output and then a regression layer works well for event type prediction.

CUDA memory error while training with real data

Try reducing the batch size in run.sh

Error in Utils.time_loss

I agree with you. These calculations impact training as well because the RMSE loss is one of the loss functions being minimized overall. This raises serious doubts over the code...

I wonder why the paper and code results are so different.

The scale of the reported results is different from the scale of timestamp in the dataset.

Prediction per timestamp

I also believe what you have stated is true (Reference lines 69 and 70 in model.py). For a sequence of length L, the model should train (and predict) the next...

Prediction per timestamp

If you go through utils.py, you will notice that the dataloader consists of length seq_len. If there is a sequence of length L, the author utilizes all contiguous seq_len length...

Ambiguity in calculating log likelihood

In addition to the Non-event LL term, the same issue persists in the event LL as well (eq. 8 in paper). Wherever lambda is written in the paper, it should...

failure to use Prange with @njit(parallel=True) on loop which can be parallelised

It seems that issue #5601 and this are similar, please add duplicate label

failure to use Prange with @njit(parallel=True) on loop which can be parallelised

I am now able to run my implementation after building from source using the master branch. But setting parallel=True increases my time of implementation as compared to not setting it(serial...

failure to use Prange with @njit(parallel=True) on loop which can be parallelised

Have you tried running the code by building Numba from source using the master branch? There was a recent change regarding the parallel accelerator. Numba is unable to handle slicing...