search

ctlearn-project / ctlearn

Deep Learning for IACT Event Reconstruction
BSD 3-Clause "New" or "Revised" License
53 stars 43 forks source link

Update cnn_rnn model to load data from multiple telescopes #81

Open Jsevillamol opened 5 years ago

Jsevillamol commented 5 years ago

In the current commit the cnn_rnn model is not usable ('telescope_data' is no longer among the features and causes a KeyError).

aribrill commented 5 years ago

The current model does work if you set merge_tel_types to True, which has the (side) effect of renaming the data array to telescope_data for compatibility. However, this is a convoluted way to do it and is not very clear.

aribrill commented 5 years ago

This is a separate issue from updating the model to load data from multiple telescopes. In that case the data arrays are named _data, _triggers, etc. with merge_tel_types==False. If the data are of different tel_types, in general you have to treat them differently because they have different shapes.

bryankim96 commented 5 years ago

I made the exact same mistake when testing my changes (not setting merge_tel_types=True when using a default model). The error message made it reasonably clear what was happening, but maybe we could consider simplifying things in the example config file? For example, we could make merge_tel_types=True by default in the example config, as it only includes one telescope anyways? Or maybe it should automatically behave like merge_tel_types=True (renaming the inputs to "telescope..." when there is only one telescope type? It seems like the fact that you can name single telescope data in two ways "telescope..." and "[telescope_name]..." could cause some confusion.

lightningedge007a commented 5 years ago

Hi @aribrill @qi-feng @nietootein , I am 3rd year, Computer Science, Bachelor of Technology student from Jaypee Institute of Information Technology. I found ROOT Input to be an amazing project and I want to pick it up for GSOC 2019. I am currently exploring the scientific methods, gathering research papers and exploring CTLearn which I will update in my proposal. How can I contact you for feedback? I am skilled in the following : Machine Learning, Deep Learning, Artificial Intelligence, Data Analysis, Data Science,Natural Language Processing (NLP), Artificial Intelligence (AI), Computer Vision, PyTorch, Keras, TensorFlow, Caffe, Python, Java, C/C++ Docker, Git, Google Colab, Amazon Web Services (AWS), Cloud Computing PL/SQL, Object-Oriented Programming (OOP), Linux

my email: abhishekverma1610@gmail.com LinkedIn : https://www.linkedin.com/in/abhishekverma1610/