def _CalculateSurvivalFunction_pysurvival(self, x, folder):
## Read in models
startTime = time.time()
models = [
load_model('./' + folder + '/DeepSurv' + str(k + 1) + '.zip')
for k in range(50)
]
## Calculate Survival functions
t = np.linspace(12, 120, 240)
s_arr = []
for model in models:
s = model.predict_survival(x)[0]
f = interpolate.interp1d(model.times, s)
s_arr.append(f(t))
s_arr_mean = np.mean(s_arr, axis=0)
s_arr_std = np.std(s_arr, axis=0)
endTime = time.time()
print('Total Time = %f seconds' % (endTime - startTime))
# self._PlotFigure(s_arr_mean, s_arr_std, x, t)
return {
'time': t,
's_mean': s_arr_mean,
's_std': s_arr_std,
'runtime': endTime - startTime
}
def get_predicted_curves(cutaneous_biopsy_ulceration, scenario, cutaneous_biopsy_histological_subtype, cutaneous_biopsy_breslow,
total_count_slnb_ldn, visceral_metastasis_location,total_positives_slnb_ldn,patient_hair_color,
cutaneous_biopsy_lymphatic_invasion,patient_eye_color,cutaneous_biopsy_mitotic_index,age,
patient_phototype, cutaneous_biopsy_satellitosis, MC1R,cutaneous_biopsy_vascular_invasion,
cutaneous_biopsy_regression,LAB2419,T0_date,LAB2406,LAB1307,patient_gender,LAB2469,LAB2544,
neutrofils_per_limfocits,cutaneous_biopsy_neurotropism,LAB2467,LAB1309,primary_tumour_location_coded,
LAB2476,LAB2679,LAB2404, cutaneous_biopsy_predominant_cell_type, LAB2407,LAB1301,LAB2498):
"""
This function returns x and y values to plot the survivorship curve of a test for a patient.
@args:
-cutaneous_biopsy_ulceration...LAB2498: numerical variables from the patient and tests
@returns:
- curve_x: the x values for the survivorship curve. This indicates time in months.
- curve_y: the y values for the survivorship curve. This indicates probability to survive.
"""
#load model
estimator_loaded = load_model('data/ExtraST_model.zip')
#load features used by the model
features = pd.read_csv("data/Features_ExtraST_model.csv").iloc[:,1]
missBIO2=0
test = pd.DataFrame([[cutaneous_biopsy_ulceration, scenario, cutaneous_biopsy_histological_subtype, cutaneous_biopsy_breslow,
total_count_slnb_ldn, visceral_metastasis_location,total_positives_slnb_ldn,patient_hair_color,
cutaneous_biopsy_lymphatic_invasion,patient_eye_color,cutaneous_biopsy_mitotic_index,age,
patient_phototype, cutaneous_biopsy_satellitosis, MC1R,cutaneous_biopsy_vascular_invasion,
cutaneous_biopsy_regression,LAB2419,T0_date,missBIO2,LAB2406,LAB1307,patient_gender,LAB2469,LAB2544,
neutrofils_per_limfocits,cutaneous_biopsy_neurotropism,LAB2467,LAB1309,primary_tumour_location_coded,
LAB2476,LAB2679,LAB2404, cutaneous_biopsy_predominant_cell_type, LAB2407,LAB1301,LAB2498]], columns=features)
n_tests = test.shape[0]
#Fill missing columns (corresponding to missing values)
test.fillna(0, inplace=True)
#Target encoding + normalization
test = preprocess_newdata(test)
#Select only the features used by the classifier
test = test[features.values]
#Predict survival curve
curve_y = estimator_loaded.predict_survival(test.values).flatten()[0:120]
curve_x = np.arange(1,len(curve_y)+1,1)
risk_group = identify_risk_group_df(test, estimator_loaded)
return curve_x, curve_y, risk_group
def identify_risk_group(test_df):
"""
This function returns the group of risk of the given set of patients test_df
given as a dataframe. The variable thr_file gives the vsc file storing the
thresholds information.
For instance, it can be called:
identify_risk_group(pd.read_csv('data/train.csv'))
"""
############# Preprocess data according to the model #############
#load model
estimator_loaded = load_model('trained_models/ExtraST_model.zip')
#load features used by the model
features = pd.read_csv('trained_models/Features_ExtraST_model.csv').iloc[:,1]
#Fill missing columns (corresponding to missing values)
test_df.fillna(0, inplace=True)
#Target encoding + normalization
test_df = preprocess_newdata(test_df)
#Select only the features used by the classifier
test_df = test_df[features.values]
############# Identify risk group #############
#read risk encoding information
thr_info = pd.read_csv('thresholds.csv')
thr_12 = thr_info['Threshold1-2'][0]
thr_23 = thr_info['Threshold2-3'][0]
norm_info = thr_info['Normalization_max'][0]
#get the risk for each patient
risk = estimator_loaded.predict_risk(test_df)
#normalize the risk
risk = np.log(risk)/norm_info
#identify the risk group
risk_group = np.zeros([len(risk)])
for i in range(len(risk)):
if risk[i]<thr_12:
risk_group[i] = int(1)
elif thr_12 < risk[i] < thr_23:
risk_group[i] = int(2)
else:
risk_group[i] = int(3)
return risk_group
def identify_risk_group(
cutaneous_biopsy_ulceration, scenario,
cutaneous_biopsy_histological_subtype, cutaneous_biopsy_breslow,
total_count_slnb_ldn, visceral_metastasis_location,
total_positives_slnb_ldn, patient_hair_color,
cutaneous_biopsy_lymphatic_invasion, patient_eye_color,
cutaneous_biopsy_mitotic_index, age, patient_phototype,
cutaneous_biopsy_satellitosis, MC1R,
cutaneous_biopsy_vascular_invasion, cutaneous_biopsy_regression,
LAB2419, T0_date, LAB2406, LAB1307, patient_gender, LAB2469, LAB2544,
neutrofils_per_limfocits, cutaneous_biopsy_neurotropism, LAB2467,
LAB1309, primary_tumour_location_coded, LAB2476, LAB2679, LAB2404,
cutaneous_biopsy_predominant_cell_type, LAB2407, LAB1301, LAB2498):
"""
This function returns the group of risk of the given set of patients test_df
given as a dataframe. The variable thr_file gives the vsc file storing the
thresholds information.
"""
############# Preprocess data according to the model #############
#load model
estimator_loaded = load_model('data/ExtraST_model.zip')
#load features used by the model
features = pd.read_csv('data/Features_ExtraST_model.csv').iloc[:, 1]
missBIO2 = 0
test_df = pd.DataFrame([[
cutaneous_biopsy_ulceration, scenario,
cutaneous_biopsy_histological_subtype, cutaneous_biopsy_breslow,
total_count_slnb_ldn, visceral_metastasis_location,
total_positives_slnb_ldn, patient_hair_color,
cutaneous_biopsy_lymphatic_invasion, patient_eye_color,
cutaneous_biopsy_mitotic_index, age, patient_phototype,
cutaneous_biopsy_satellitosis, MC1R,
cutaneous_biopsy_vascular_invasion, cutaneous_biopsy_regression,
LAB2419, T0_date, missBIO2, LAB2406, LAB1307, patient_gender, LAB2469,
LAB2544, neutrofils_per_limfocits, cutaneous_biopsy_neurotropism,
LAB2467, LAB1309, primary_tumour_location_coded, LAB2476, LAB2679,
LAB2404, cutaneous_biopsy_predominant_cell_type, LAB2407, LAB1301,
LAB2498
]],
columns=features)
#Fill missing columns (corresponding to missing values)
test_df.fillna(0, inplace=True)
#Select only the features used by the classifier
test_df = test_df[features.values]
#Target encoding + normalization
test_df = preprocess_newdata(test_df)
############# Identify risk group #############
#read risk encoding information
thr_info = pd.read_csv('data/thresholds.csv')
thr_12 = thr_info['Threshold1-2'][0]
thr_23 = thr_info['Threshold2-3'][0]
norm_info = thr_info['Normalization_max'][0]
#get the risk for each patient
risk = estimator_loaded.predict_risk(test_df)
#normalize the risk
risk = np.log(risk) / norm_info
#identify the risk group
if risk < thr_12:
risk_group = int(1)
elif thr_12 < risk < thr_23:
risk_group = int(2)
else:
risk_group = int(3)
return risk_group
import numpy as np
import pandas as pd
from pysurvival.utils import load_model
import matplotlib.pyplot as plt
# Import selected features from finalPrediction csv
predictionData = pd.read_csv('./finalPredictionData.csv')
# Remove unnecessary columns
onlyPredictionData = predictionData.drop(columns = [ 'AliveStatus0Dead1Alive','NumDays'])
survivalModel = load_model('./survival_model.zip')
preds = survivalModel.predict_survival(onlyPredictionData)
preds_df = pd.DataFrame(preds).T
preds_df.to_excel('preds.xlsx')
plt.plot(preds_df, label = "Survival Data")
plt.legend()
plt.show()
import dash
import dash_core_components as dcc
import dash_html_components as html
import dash_bootstrap_components as dbc
import plotly.express as px
from pysurvival.models.semi_parametric import NonLinearCoxPHModel
import pandas as pd
from dash.dependencies import Input, Output, State
from pysurvival.utils import load_model
from toolbox import *
from dash_table import DataTable
#read torch models
pfsMod = load_model('modelData/final_pfs.zip')
rfsMod = load_model('modelData/final_rfs.zip')
pfsModMMC = load_model('modelData/MMCPFS.zip')
rfsModMMC = load_model('modelData/MMCRFS.zip')
external_stylesheets = [
'https://codepen.io/chriddyp/pen/bWLwgP.css', dbc.themes.GRID
]
app = dash.Dash(__name__, external_stylesheets=external_stylesheets)
#app.css.config.serve_locally = True
#app.scripts.config.serve_locally = True
app.layout = html.P(
id='page_content',
className='app_body',
def retrieving_MaxMin_riskGroup(risk_group, data_path = 'data/train.csv'):
"""
This function retrieves the survival curve of the patient with maximum and minimum
risk for the risk_group specified in the input.
risk_group can be 1, 2 or 3
The output is: times (domain of the curve), survival_curve_min (curve of the
minimum risk patient) and survival_curve_max (curve of the maximum risk patient)
"""
############# Preprocess data according to the model #############
#load model
estimator_loaded = load_model('data/ExtraST_model.zip')
#load features used by the model
features = pd.read_csv('data/Features_ExtraST_model.csv').iloc[:,1]
features.drop(features[features == 'missBIO2'].index, inplace = True)
#read the dataset
test_df = pd.read_csv(data_path)
#Select only the features used by the classifier
test_df = test_df[features]
#Fill missing columns (corresponding to missing values)
test_df.fillna(0, inplace=True)
#Target encoding + normalization
test_df = preprocess_newdata(test_df)
#read risk encoding information
thr_info = pd.read_csv('data/thresholds.csv')
thr_12 = thr_info['Threshold1-2'][0]
thr_23 = thr_info['Threshold2-3'][0]
norm_info = thr_info['Normalization_max'][0]
#get the risk of the patients
risk = estimator_loaded.predict_risk(test_df)
risk = np.log(risk)/norm_info
#cluster the patients according to their risk group
if risk_group == 1:
patient_index = np.where(risk < thr_12)[0]
elif risk_group == 2:
patient_index = np.where((thr_12 < risk) & (risk < thr_23))[0]
else:
patient_index = np.where(risk>thr_23)[0]
#get the index of the patient with maximum and minimum risk within each group
aux = np.argmax(risk[patient_index])
maximum_patient = patient_index[aux]
aux = np.argmin(risk[patient_index])
minimum_patient = patient_index[aux]
survival_curve_min = estimator_loaded.predict_survival(test_df.iloc[minimum_patient]).flatten()
survival_curve_max = estimator_loaded.predict_survival(test_df.iloc[maximum_patient]).flatten()
times = np.arange(survival_curve_max.shape[0])
return times, survival_curve_min, survival_curve_max
"""