BioMAT / biomat_app.py
motionsh's picture
Upload biomat_app.py
7ece8c4
raw
history blame
6.19 kB
import streamlit as st
import os
import torch
from torch.utils.data import DataLoader
from config import get_config_universal
from dataset import DataSet
from datasetbuilder import DataSetBuilder
from test import Test
from visualization.steamlit_plot import plot_kinematic_predictions
dataset_name = 'camargo'
config = get_config_universal(dataset_name)
# model_file = 'transformertsai_g1g2rardsasd_g1g2rardsasd.pt'
# model = torch.load(os.path.join('./caches/trained_model/v05', model_file))
sensor_options = {'Thigh & Shank & Foot': ['foot', 'shank', 'thigh'],
'Thigh & Shank': ['thigh', 'shank'],
'Thigh & Foot': ['thigh', 'foot'],
'Shank & Foot': ['shank', 'foot'],
'Thigh': ['thigh'],
'Shank': ['shank'],
'Foot': ['foot']}
@st.cache
def fetch_data(config):
dataset_handler = DataSet(config, load_dataset=True)
kihadataset_train, kihadataset_test = dataset_handler.run_dataset_split_loop()
kihadataset_train['x'], kihadataset_train['y'], kihadataset_train['labels'] = dataset_handler.run_segmentation(
kihadataset_train['x'],
kihadataset_train['y'], kihadataset_train['labels'])
kihadataset_test['x'], kihadataset_test['y'], kihadataset_test['labels'] = dataset_handler.run_segmentation(
kihadataset_test['x'],
kihadataset_test['y'], kihadataset_test['labels'])
train_dataset = DataSetBuilder(kihadataset_train['x'], kihadataset_train['y'], kihadataset_train['labels'],
transform_method=config['data_transformer'], scaler=None, noise=None)
test_dataset = DataSetBuilder(kihadataset_test['x'], kihadataset_test['y'], kihadataset_test['labels'],
transform_method=config['data_transformer'], scaler=train_dataset.scaler,
noise=None)
test_dataloader = DataLoader(dataset=test_dataset, batch_size=config['batch_size'], shuffle=False)
return test_dataloader, kihadataset_test
# @st.cache()
def fetch_model(sensor_name, model_name):
device = torch.device('cpu')
model_names = {'CNNLSTM':'hernandez2021cnnlstm', 'BiLSTM':'bilstm', 'BioMAT': 'transformertsai'}
sensor_names = {'Thigh & Shank & Foot':'thighshankfoot',
'Thigh & Shank':'thighshank',
'Thigh & Foot':'thighfoot',
'Shank & Foot':'shankfoot',
'Thigh':'thigh',
'Shank':'shank',
'Foot':'foot'}
if sensor_names[sensor_name]=='thighshankfoot':
model_file = model_names[model_name] + '_g1g2rardsasd_g1g2rardsasd.pt'
else:
model_file = sensor_names[sensor_name] + '_' + model_names[model_name]+'_g1g2rardsasd_g1g2rardsasd.pt'
# model = torch.load(os.path.join('./caches/trained_model/v05', model_file), map_location=device)
st.write(model_file)
model = torch.load(os.path.join('./caches/trained_model/v05', model_file))
return model
# @st.cache
def fetch_predictions(model):
test_handler = Test()
y_pred, y_true, loss = test_handler.run_testing(config, model, test_dataloader=test_dataloader)
y_true = y_true.detach().cpu().clone().numpy()
y_pred = y_pred.detach().cpu().clone().numpy()
return y_pred, y_true, loss
# sensor_name = 'Thigh & Shank & Foot'
# config['sensor_sensor'] = sensor_options[sensor_name]
# test_dataloader, kihadataset_test = fetch_data(config)
# model = fetch_model(sensor_name, 'BioMAT')
# y_pred, y_true, loss = fetch_predictions(model)
# fig = plot_kinematic_predictions(y_true, y_pred, kihadataset_test['labels'], 'AB24',
# selected_activities= ['LevelGround Walking', 'Ramp Ascent', 'Ramp Descent', 'Stair Ascent', 'Stair Descent'],
# selected_index_to_plot=1)
st.set_page_config(layout="wide")
# col1, col2, col3 = st.columns(3)
# with col2:
st.title('BioMAT:Biomechanical Multi-Activity Transformer Model for Joint Kinematic Prediction From IMUs')
# st.info('If you change the sensor configeration, press rerun', icon="ℹ️")
st.sidebar.title('Sensor Configuration')
selected_sensor = st.sidebar.selectbox('Pick one', ['Thigh & Shank & Foot',
'Thigh & Shank',
'Thigh & Foot',
'Shank & Foot',
'Thigh',
'Shank',
'Foot'])
config['selected_sensors'] = sensor_options[selected_sensor]
print(config)
st.sidebar.title('Model Configuration')
selected_model = st.sidebar.selectbox('Pick one', ['CNNLSTM',
'BiLSTM',
'BioMAT'])
st.sidebar.title('Subject')
selected_subject = st.sidebar.slider('Pick a Subject Number', min_value=23, max_value=25, step=1)
st.sidebar.title('Activity')
selected_activities = st.sidebar.multiselect('Pick Output Activities',
['LevelGround Walking', 'Ramp Ascent', 'Ramp Descent', 'Stair Ascent', 'Stair Descent'])
index_to_plot = st.sidebar.number_input('Enter a number between 0 and 5', min_value=0, max_value=5)
if st.sidebar.button('Predict'):
with st.spinner('Data is loading...'):
test_dataloader, kihadataset_test = fetch_data(config)
st.success('Data is loaded!')
with st.spinner('Model is loading...'):
model = fetch_model(selected_sensor, selected_model)
st.success('Model is loaded!')
with st.spinner('Prediction ...'):
y_pred, y_true, loss = fetch_predictions(model)
st.success('Prediction is Completed!')
st.write('plot ...')
subject = 'AB' + str(selected_subject)
fig = plot_kinematic_predictions(y_true, y_pred, kihadataset_test['labels'], subject,
selected_activities=selected_activities, selected_index_to_plot=index_to_plot)
st.plotly_chart(fig, use_container_width=True)
#