Oufattole/meds-torch

Issues

• Allow Predict Script to Include Custom Prefixed Keys in Output and Batch Indexes

  #112 opened in 12 hours
  0

• The `Module` ABC class is confusing and should be re-done as a decorator or Mixin

  #111 opened in 12 hours
  0

• Add a predict script that generates embeddings

  #109 opened 12 hours ago
  0

• Update the Triplet autoregressive model to perform conditional sampling of values conditioned on the code

  #107 opened 3 days ago
  0

• Unconstrained Generative evaluation

  #106 opened 3 days ago
  0

• Zero-shot evaluation

  #105 opened 3 days ago
  0

• Add API support for generation that allow generating data after a prediction time and caching both the predictions and the real data.

  #104 opened 3 days ago
  1

• Static data is not being used in cases where it is not prepended.

  #100 opened 6 days ago
  0

• The EOS max seq length adjustment should happen in the dataset, not in the config, to avoid violating the user's inputs.

  #99 opened 6 days ago
  0

• Delete the unused `dynamic_only_collate`

  #96 opened 6 days ago
  0

• Is the max seq len constraint not respected if postpend EOS or prepend static is enabled?

  #95 opened 6 days ago
  1

• All collate functions should be the same and static data pre-pending should happen in getitem.

  #94 opened 6 days ago
  6

• Simplify the dataset index for now to just have a subject ID and an end event index.

  #92 opened 12 days ago
  1

• Doctests are not being run by github workflows

  #91 opened 8 days ago
  3

• Out labels is never used and should be deleted.

  #89 opened 7 days ago
  0

• Clean up docstrings and add doctests where appropriate

  #87 opened 13 days ago
  1

• Clean up and improve the getitem code path for subject data

  #86 opened 13 days ago
  0

• This line is wrong if collate type is not event stream because start is reset to zero, when that isn't what the target for this should be.

  #84 opened 6 days ago
  1

• Remove the event stream specific code as nothing is using that for now.

  #81 opened 6 days ago
  0

• Tests take too long

  #80 opened 14 days ago
  0

• General MEDS dataset may have patient splits stored separately

  #79 opened 14 days ago
  0

• Benchmarking

  #78 opened 14 days ago
  0

• Add processing of meds-transform data to the integration tests

  #77 opened 14 days ago
  1

• Simplify the task processing logic and task dataframe expectations to reflect a purely single-task structure with the meds label schema guaranteed.

  #75 opened 8 days ago
  0

• Improve and simplify the logic around reading the schema parquet files.

  #74 opened 14 days ago
  0

• Should look into using 1D tensors with offsets instead of padding at all.

  #72 opened 14 days ago
  0

• We need to handle some way of not densifying the 4th dimension (text or ecgs for example) as that will be extremely sparse.

  #71 opened 14 days ago
  0

• Use JNRTs with their built in logic (e.g., slicing, flattening, vstacking, appending, and to-dense)

  #70 opened 6 days ago
  0

• remove `get_task_info`, as we assume we are always doing binary classification. You should remove task_info related logic and errors from `load_subject` too.

  #68 opened 14 days ago
  0

• Remove `filter_to_min_seq_len`, we assume meds-transform handles this

  #63 opened 14 days ago
  0

• remove `filter_to_subset`, as this is supported in pytorch lightning

  #62 opened 14 days ago
  0

• Delete `set_inter_event_timestamps`

  #61 opened 14 days ago
  0

• Syntax should be updated to support NRT > 0.0.6

  #60 opened 8 days ago
  0

• Tests are failing because NRT syntax is for an older version

  #58 opened 15 days ago
  0

• Remove all aspects of text tokenization from the dataset class

  #56 opened 14 days ago
  0

• Dataset API and Configuration

  #55 opened 15 days ago
  7

• Complete synthetic testing data

  #54 opened 19 days ago
  0

• Frozen reps kept in memory

  #51 opened 23 days ago
  0

• Update the Project description

  #50 opened a month ago
  0

• Fairness Evaluation

  #48 opened a month ago
  6

• Improve Pretrained Model Loading for Fine-tuning

  #47 opened a month ago
  0

• Implement time delta normalization in set_inter_event_time_stats

  #45 opened a month ago
  0

• Investigate and resolve performance bottleneck

  #44 opened a month ago
  1

• Refine input validation for predefined workflows

  #43 opened a month ago
  0

• Remove duplicate Ray Tune checkpoints in common directory

  #42 opened a month ago
  0

• Resolving hydra ${now:...} syntax when tuning not working.

  #41 opened a month ago
  0

• Model Performance is Suboptimal

  #39 opened 2 months ago
  1

• Autoregressive Modeling

  #30 opened 3 months ago
  8

• Merge the Text Tokenization Branch

  #29 opened 3 months ago
  0

• Adding dummy data

  #28 opened 3 months ago
  0