Code used to run the experiments for Result2

demo_promptehr.py

@@ -0,0 +1,201 @@
+#!/usr/bin/env python
+# coding: utf-8
+
+# In[1]:
+
+
+import transformers
+
+
+# In[2]:
+
+
+transformers.__version__
+
+
+# In[3]:
+
+
+from pytrial.data.demo_data import load_synthetic_ehr_sequence
+from pytrial.tasks.trial_simulation.data import SequencePatient
+
+demo = load_synthetic_ehr_sequence(n_sample=100)
+
+
+# In[4]:
+
+
+demo
+
+
+# In[5]:
+
+
+len(demo['visit'])
+
+
+# In[6]:
+
+
+demo.keys()
+
+
+# In[7]:
+
+
+# build sequence dataset
+seqdata = SequencePatient(data={'v':demo['visit'], 'y':demo['y'], 'x':demo['feature'],},
+ metadata={
+ 'visit':{'mode':'dense'},
+ 'label':{'mode':'tensor'}, 
+ 'voc':demo['voc'],
+ 'max_visit':20,
+ }
+ )
+
+print('visit', demo['visit'][0]) # a list of visit events
+print('mortality', demo['y'][0]) # array of labels
+print('feature', demo['feature'][0]) # array of patient baseline features
+print('voc', demo['voc']) # dict of dicts containing the mapping from index to the original event names
+print('order', demo['order']) # a list of three types of code
+print('n_num_feature', demo['n_num_feature']) # int: a number of patient's numerical features
+print('cat_cardinalities', demo['cat_cardinalities']) # list: a list of cardinalities of patient's categorical features
+
+
+# In[9]:
+
+
+demo['voc']
+
+
+# In[14]:
+
+
+demo['voc']['med'].idx2word
+
+
+# In[ ]:
+
+
+
+
+
+# In[18]:
+
+
+from promptehr import PromptEHR
+
+# fit the model
+model = PromptEHR(
+ code_type=demo['order'],
+ n_num_feature=demo['n_num_feature'],
+ cat_cardinalities=demo['cat_cardinalities'],
+ num_worker=0,
+ eval_step=1,
+ epoch=5,
+ device=[0],
+)
+model.fit(
+ train_data=seqdata,
+ val_data=seqdata,
+)
+
+
+# In[15]:
+
+
+model.evaluate(seqdata)
+
+
+# In[17]:
+
+
+model
+
+
+# In[ ]:
+
+
+
+
+
+# In[ ]:
+
+
+# save the model
+model.save_model('./simulation/promptEHR')
+
+
+# In[12]:
+
+
+# generate fake records
+res = model.predict(seqdata, n_per_sample=10, n=10, verbose=True)
+
+
+# In[13]:
+
+
+print(res)
+
+
+# In[ ]:
+
+
+
+
+
+# In[1]:
+
+
+import os
+os.chdir('../')
+
+
+# In[16]:
+
+
+# if you want pretrained model downloaded
+from promptehr import PromptEHR
+model = PromptEHR()
+model.from_pretrained()
+
+
+# In[17]:
+
+
+model.training_args
+
+
+# In[19]:
+
+
+model.evaluate(seqdata)
+
+
+# In[ ]:
+
+
+
+
+
+# In[ ]:
+
+
+
+
+
+# In[15]:
+
+
+model.fit(
+ train_data=seqdata,
+ val_data=seqdata,
+)
+
+
+# In[ ]:
+
+
+
+