update

app.py

@@ -3,21 +3,20 @@ import json
 import random
 import pandas as pd
 import pickle
+import json
 
 # set page configuration to wide mode
 st.set_page_config(layout="wide")
 
-st.markdown("""
-<style>
-.bounding-box {
-    border: 2px solid #4CAF50;  # Green border
-    border-radius: 5px;         # Rounded corners
-    padding: 10px;              # Padding inside the box
-    margin: 10px;               # Space outside the box
-}
-</style>
-""", unsafe_allow_html=True)
+# section 1
+st.markdown("#### About")
+st.markdown("Differential diagnosis of dementia remains a challenge in neurology due to symptom overlap across etiologies, yet it is crucial for formulating early, personalized management strategies. Here, we present an AI model that harnesses a broad array of data, including demographics, individual and family medical history, medication use, neuropsychological assessments, functional evaluations, and multimodal neuroimaging, to identify the etiologies contributing to dementia in individuals.")
+
+# section 2
+st.markdown("#### Demo")
+st.markdown("Please enter the input features in the textbox below, formatted as a JSON dictionary. Click the \"**RANDOM EXAMPLE**\" button to populate the textbox with a randomly selected case from the NACC testing dataset. Use the \"**PREDICT**\" button to submit your input to the model, which will then provide probability predictions for mental status and all 10 etiologies.")
 
+# load model
 @st.cache_resource
 def load_model():
     import adrd
@@ -25,26 +24,21 @@ def load_model():
         ckpt_path = './ckpt_swinunetr_stripped_MNI.pt'
         model = adrd.model.ADRDModel.from_ckpt(ckpt_path, device='cpu')
     except:
-        ckpt_path = '../adrd_tool_copied_from_sahana/dev/ckpt/ckpt_swinunetr_stripped_MNI.pt'
-        model = adrd.model.ADRDModel.from_ckpt(ckpt_path, device='cpu')
+        # ckpt_path = '../adrd_tool_copied_from_sahana/dev/ckpt/ckpt_swinunetr_stripped_MNI.pt'
+        # model = adrd.model.ADRDModel.from_ckpt(ckpt_path, device='cpu')
+        return None
     return model
 
-@st.cache_resource
-def load_nacc_data():
-    from data.dataset_csv import CSVDataset
-    dat = CSVDataset(
-        dat_file = "./data/test.csv", 
-        cnf_file = "./data/input_meta_info.csv"
-    )
-    return dat
-
 model = load_model()
-dat_tst = load_nacc_data()
 
 def predict_proba(data_dict):
     pred_dict = model.predict_proba([data_dict])[1][0]
     return pred_dict
 
+# load meta data csv
+file_path = './data/input_meta_info.csv'
+input_meta_info = pd.read_csv(file_path)
+
 # load NACC testing data
 from data.dataset_csv import CSVDataset
 dat_tst = CSVDataset(
@@ -52,42 +46,98 @@ dat_tst = CSVDataset(
     cnf_file = "./data/input_meta_info.csv"
 )
 
-# initialize session state for the text input if it's not already set
-if 'input_text' not in st.session_state:
-    st.session_state.input_text = ""
-
-# section 1
-st.markdown("#### About")
-st.markdown("Differential diagnosis of dementia remains a challenge in neurology due to symptom overlap across etiologies, yet it is crucial for formulating early, personalized management strategies. Here, we present an AI model that harnesses a broad array of data, including demographics, individual and family medical history, medication use, neuropsychological assessments, functional evaluations, and multimodal neuroimaging, to identify the etiologies contributing to dementia in individuals.")
-
-# section 2
-st.markdown("#### Demo")
-st.markdown("Please enter the input features in the textbox below, formatted as a JSON dictionary. Click the \"**Random case**\" button to populate the textbox with a randomly selected case from the NACC testing dataset. Use the \"**Predict**\" button to submit your input to the model, which will then provide probability predictions for mental status and all 10 etiologies.")
-
-# layout
-layout_l, layout_r = st.columns([1, 1])
-
-# create a form for user input
-with layout_l:
-    with st.form("json_input_form"):
-        json_input = st.text_area(
-            "Please enter JSON-formatted input features:", 
-            value = st.session_state.input_text,
-            height = 300
+def get_random_example():
+    idx = random.randint(0, len(dat_tst) - 1)
+    random_case = dat_tst[idx][0]
+    return random_case
+
+# Get random example features if the button is clicked
+if 'random_example' not in st.session_state:
+    st.session_state.random_example = None
+
+if st.button("RANDOM EXAMPLE"):
+    st.session_state.random_example = get_random_example()
+    st.rerun()
+
+random_example = st.session_state.random_example
+
+def create_input(i):
+    row = input_meta_info.iloc[i]
+    name = row['Name']
+    description = row['Description']
+
+    # dirty work, inspect keys and values
+    values = row['Values']
+    values = values.replace('\'', '\"')
+    values = values.replace('\"0\": nan, ', '')
+    values = json.loads(values)
+
+    for k, v in list(values.items()):
+        if v == 'Unknown':
+            values.pop(k)
+        elif k in ('9', '99', '999'):
+            values.pop(k)
+
+        # get default value from random example if available
+    default_value = random_example[name] if random_example and name in random_example else None
+    if type(default_value) is float:
+        default_value = int(default_value)
+
+    # Determine the type of widget based on values
+    if 'range' in values:
+        if ' - ' in values['range']:
+            min_value, max_value = map(float, values['range'].split(' - '))
+            min_value, max_value = int(min_value), int(max_value)
+
+            if default_value is not None:
+                if default_value > max_value or default_value < min_value:
+                    default_value = None
+                
+            st.number_input(description, key=name, min_value=min_value, max_value=max_value, value=default_value, placeholder=values['range'])
+        else:
+            min_value = int(values['range'].replace('>= ', ''))
+            if default_value is not None:
+                if default_value < min_value or default_value == 8888:
+                    default_value = None
+
+            st.number_input(description, key=name, min_value=min_value, value=default_value, placeholder=values['range'])
+    else:
+        values = {int(k): v for k, v in values.items()}
+        reverse_mapping = {v: k for k, v in values.items()}
+        if default_value in values:
+            default_index = list(values.keys()).index(default_value)
+        else:
+            default_index = None
+
+        st.selectbox(
+            description, 
+            options = values.keys(), 
+            key = name, 
+            index = default_index,
+            format_func=lambda x: values[x]
         )
 
-        # create three columns
-        left_col, middle_col, right_col = st.columns([3, 4, 1])
+# create form
+with st.form("dynamic_form"):
+    # random_example_button = st.form_submit_button("RANDOM EXAMPLE")
 
-        with left_col:
-            sample_button = st.form_submit_button("Random case")
+    cols = st.columns(3)
+    with cols[0]:
+        for i in range(0, len(input_meta_info), 3):
+            create_input(i)
+    with cols[1]:
+        for i in range(1, len(input_meta_info), 3):
+            create_input(i)
+    with cols[2]:
+        for i in range(2, len(input_meta_info), 3):
+            create_input(i)    
 
-        with right_col:
-            submit_button = st.form_submit_button("Predict")
-    
+    predict_button = st.form_submit_button("PREDICT")
+
+# load mapping
 with open('./data/nacc_variable_mappings.pkl', 'rb') as file:
     nacc_mapping = pickle.load(file)
-    
+
 def convert_dictionary(original_dict, mappings):
     transformed_dict = {}
     
@@ -104,31 +154,17 @@ def convert_dictionary(original_dict, mappings):
             transformed_dict[new_key] = transformed_value
     
     return transformed_dict
- 
-if sample_button:
-    idx = random.randint(0, len(dat_tst) - 1)
-    random_case = dat_tst[idx][0]    
-    st.session_state.input_text = json.dumps(random_case, indent=2)
 
-    # reset input text after form processing to show updated text in the input box
-    if 'input_text' in st.session_state:
-        st.experimental_rerun()
+if predict_button:
+    # get form input
+    names = input_meta_info['Name'].tolist()
+    data_dict = {}
+    for name in names:
+        data_dict[name] = st.session_state[name]
     
-elif submit_button:
-    try:
-        # Parse the JSON input into a Python dictionary
-        data_dict = json.loads(json_input)
-        data_dict = convert_dictionary(data_dict, nacc_mapping)
-        # print(data_dict)
-        pred_dict = predict_proba(data_dict)
-        with layout_r:
-            st.write("Predicted probabilities:")
-            st.code(json.dumps(pred_dict, indent=2))
-    except json.JSONDecodeError as e:
-        # Handle JSON parsing errors
-        st.error(f"An error occurred: {e}")
-
-# section 3
-st.markdown("#### Feature Table")
-df_input_meta_info = pd.read_csv('./data/input_meta_info.csv')
-st.table(df_input_meta_info)
+    # convert
+    data_dict = convert_dictionary(data_dict, nacc_mapping)
+    pred_dict = predict_proba(data_dict)
+
+    st.write("Predicted probabilities:")
+    st.code(json.dumps(pred_dict, indent=2))

app_backup_1.py

@@ -0,0 +1,135 @@
+import streamlit as st
+import json
+import random
+import pandas as pd
+import pickle
+
+# set page configuration to wide mode
+st.set_page_config(layout="wide")
+
+st.markdown("""
+<style>
+.bounding-box {
+    border: 2px solid #4CAF50;  # Green border
+    border-radius: 5px;         # Rounded corners
+    padding: 10px;              # Padding inside the box
+    margin: 10px;               # Space outside the box
+}
+</style>
+""", unsafe_allow_html=True)
+
+@st.cache_resource
+def load_model():
+    import adrd
+    try:   
+        ckpt_path = './ckpt_swinunetr_stripped_MNI.pt'
+        model = adrd.model.ADRDModel.from_ckpt(ckpt_path, device='cpu')
+    except:
+        # ckpt_path = '../adrd_tool_copied_from_sahana/dev/ckpt/ckpt_swinunetr_stripped_MNI.pt'
+        # model = adrd.model.ADRDModel.from_ckpt(ckpt_path, device='cpu')
+        return None
+    return model
+
+@st.cache_resource
+def load_nacc_data():
+    from data.dataset_csv import CSVDataset
+    dat = CSVDataset(
+        dat_file = "./data/test.csv", 
+        cnf_file = "./data/input_meta_info.csv"
+    )
+    return dat
+
+model = load_model()
+dat_tst = load_nacc_data()
+
+def predict_proba(data_dict):
+    pred_dict = model.predict_proba([data_dict])[1][0]
+    return pred_dict
+
+# load NACC testing data
+from data.dataset_csv import CSVDataset
+dat_tst = CSVDataset(
+    dat_file = "./data/test.csv", 
+    cnf_file = "./data/input_meta_info.csv"
+)
+
+# initialize session state for the text input if it's not already set
+if 'input_text' not in st.session_state:
+    st.session_state.input_text = ""
+
+# section 1
+st.markdown("#### About")
+st.markdown("Differential diagnosis of dementia remains a challenge in neurology due to symptom overlap across etiologies, yet it is crucial for formulating early, personalized management strategies. Here, we present an AI model that harnesses a broad array of data, including demographics, individual and family medical history, medication use, neuropsychological assessments, functional evaluations, and multimodal neuroimaging, to identify the etiologies contributing to dementia in individuals.")
+
+# section 2
+st.markdown("#### Demo")
+st.markdown("Please enter the input features in the textbox below, formatted as a JSON dictionary. Click the \"**Random case**\" button to populate the textbox with a randomly selected case from the NACC testing dataset. Use the \"**Predict**\" button to submit your input to the model, which will then provide probability predictions for mental status and all 10 etiologies.")
+
+# layout
+layout_l, layout_r = st.columns([1, 1])
+
+# create a form for user input
+with layout_l:
+    with st.form("json_input_form"):
+        json_input = st.text_area(
+            "Please enter JSON-formatted input features:", 
+            value = st.session_state.input_text,
+            height = 300
+        )
+
+        # create three columns
+        left_col, middle_col, right_col = st.columns([3, 4, 1])
+
+        with left_col:
+            sample_button = st.form_submit_button("Random case")
+
+        with right_col:
+            submit_button = st.form_submit_button("Predict")
+    
+with open('./data/nacc_variable_mappings.pkl', 'rb') as file:
+    nacc_mapping = pickle.load(file)
+    
+def convert_dictionary(original_dict, mappings):
+    transformed_dict = {}
+    
+    for key, value in original_dict.items():
+        if key in mappings:
+            new_key, transform_map = mappings[key]
+            
+            # If the value needs to be transformed
+            if value in transform_map:
+                transformed_value = transform_map[value]
+            else:
+                transformed_value = value  # Keep the original value if no transformation is needed
+            
+            transformed_dict[new_key] = transformed_value
+    
+    return transformed_dict
+ 
+if sample_button:
+    idx = random.randint(0, len(dat_tst) - 1)
+    random_case = dat_tst[idx][0]    
+    st.session_state.input_text = json.dumps(random_case, indent=2)
+
+    # reset input text after form processing to show updated text in the input box
+    if 'input_text' in st.session_state:
+        st.experimental_rerun()
+    
+elif submit_button:
+    try:
+        # Parse the JSON input into a Python dictionary
+        data_dict = json.loads(json_input)
+        data_dict = convert_dictionary(data_dict, nacc_mapping)
+        # print(data_dict)
+        pred_dict = predict_proba(data_dict)
+        with layout_r:
+            st.write("Predicted probabilities:")
+            st.code(json.dumps(pred_dict, indent=2))
+    except json.JSONDecodeError as e:
+        # Handle JSON parsing errors
+        st.error(f"An error occurred: {e}")
+
+# section 3
+st.markdown("#### Feature Table")
+df_input_meta_info = pd.read_csv('./data/input_meta_info.csv')
+st.table(df_input_meta_info)