Update README.md

README.md

@@ -2,32 +2,50 @@
 license: apache-2.0
 tags:
 - generated_from_trainer
+- medical
 model-index:
-- name: stop_reasons_classificator_multilabel_pt
+- name: stop_reasons_classificator_multilabel
   results: []
+datasets:
+- opentargets/clinical_trial_reason_to_stop
+language:
+- en
+metrics:
+- accuracy
+library_name: transformers
 ---
 
 <!-- This model card has been generated automatically according to the information the Trainer had access to. You
 should probably proofread and complete it, then remove this comment. -->
 
-# stop_reasons_classificator_multilabel_pt
+# stop_reasons_classificator_multilabel
+
+This model is a fine-tuned version of [bert-base-uncased](https://huggingface.co/bert-base-uncased) on the task of classification of why a clinical trial has stopped early in 17 different classes. The datased used for fine tuning was manually curated by a group of experts in Open Targets and is also available for download at the [Hub](https://huggingface.co/datasets/opentargets/clinical_trial_reason_to_stop).
+
 
-This model is a fine-tuned version of [bert-base-uncased](https://huggingface.co/bert-base-uncased) on an unknown dataset.
 It achieves the following results on the evaluation set:
 - Loss: 0.0899
 - Accuracy Thresh: 0.9760
 
 ## Model description
 
-More information needed
+This research has been done by Olesya Razuvayevskaya (@LesyaR).
+
+We fine-tuned BERT model for the task of predicting the stop reasons on the training set of 3,571
+human-annotated stopped clinical trials (Devlin et al., 2018). We used a BERT uncased pre-trained
+model with a one-layer feed-forward classifier. The fine-tuning was performed by using the
+Hugging Face transformer library (Wolf et al., 2019). The classifier uses 50 hidden units and the
+ReLu activation function.
 
 ## Intended uses & limitations
 
-More information needed
+This model is intended to be used by the whole scientific community. It is Apache 2.0 licensed.
 
 ## Training and evaluation data
 
-More information needed
+An expert-curated data set of >5000 reasons why a clinical trials have stopped. These data have been extracted from clinicaltrials.gov.
+
+A set of experts from the Open Targets Consortium assigned these free text labels to a set of 17 different classes after receiving training.
 
 ## Training procedure
 
@@ -60,4 +78,4 @@ The following hyperparameters were used during training:
 - Transformers 4.26.0
 - Pytorch 1.12.1+cu102
 - Datasets 2.9.0
-- Tokenizers 0.13.2
+- Tokenizers 0.13.2