使用PyMC3中的协变量建立weibullAFT模型？

Question

I'm building a Weibull AFT with covariates model for survival analysis using PyMC3 and theano.tensor. I have been reading all the notebooks associated with survival analysis in PyMC3 to build my model. However, I get the following error:

*File "C:\Users\71516706\Anaconda3\lib\site- 
packages\theano\tensor\basic.py", line 200, in as_tensor_variable
raise AsTensorError("Cannot convert %s to TensorType" % str_x, type(x))

AsTensorError: ('Cannot convert 0     Elemwise{neg,no_inplace}.0\n3     
Elemwise{neg,no_inplace}.0\n4     Elemwise{neg,no_inplace}.0\n6     
Elemwise{neg,no_inplace}.0\n8     Elemwise{neg,no_inplace}.0\n9     
Elemwise{neg,no_inplace}.0\n11    Elemwise{neg,no_inplace}.0\n12    
Elemwise{neg,no_inplace}.0\n14    Elemwise{neg,no_inplace}.0\n16    
Elemwise{neg,no_inplace}.0\n17    Elemwise{neg,no_inplace}.0\n18    
Elemwise{neg,no_inplace}.0\n19    Elemwise{neg,no_inplace}.0\n20    
Elemwise{neg,no_inplace}.0\n21    Elemwise{neg,no_inplace}.0\n23    
Elemwise{neg,no_inplace}.0\n25    Elemwise{neg,no_inplace}.0\n28    
Elemwise{neg,no_inplace}.0\n29    Elemwise{neg,no_inplace}.0\nName: 
runningtime_cens, dtype: object to TensorType', <class 
'pandas.core.series.Series'>)*

The model I want to code is the same as the class weibull_aft_fitter in lifelines ( https://lifelines.readthedocs.io/en/latest/lifelines.fitters.html#module-lifelines.fitters.weibull_aft_fitter )

My code is:

import pandas as pd
import numpy as np
import pymc3 as pm
import theano.tensor as tt

# Defining Log complementary cdf of Weibull distribution
def weibull_lccdf(x, alpha, beta):
    ''' Log complementary cdf of Weibull distribution. '''
    return -(x / beta)**alpha

# Importing Weibull data
data_weibull=pd.read_csv("data_weibull.csv")

# censored list
event = data_weibull['event'].values.astype(bool)

# Specifying the model
with pm.Model() as model:

    # Priors for unknown model parameters
    rho = pm.Gamma('rho', alpha=0.001, beta=0.001, testval=2)
    beta = pm.Normal('beta', mu=0, sd=0.001, shape=2, testval=0)

    # Expected value of lambda parameter
    lambda_obs = pm.Deterministic('lambda_obs', \
         tt.exp(beta[0]+beta[1]*data_weibull['temperature'][event]))
    lambda_cens = pm.Deterministic('lambda_cens', \
         tt.exp(beta[0]+beta[1]*data_weibull['temperature'][~event]))

    # Likelihood (sampling distribution) of observations
    runningtime_obs = pm.Weibull('runningtime_obs', alpha=rho, \
         beta=lambda_obs, observed=data_weibull['runningtime'][event])
    runningtime_cens = pm.Potential('runningtime_cens', \
         weibull_lccdf(data_weibull['runningtime'][~event], rho, \ 
         lambda_cens))

nc=3
ni=10000
nb=1000
nt=5

with model:
    # draw ni posterior samples
    step = pm.NUTS(vars=[rho, beta])
    start = pm.find_MAP()
    trace = pm.sample(draws=ni, tune=nb, chains=nc, step=step, cores=1, 
        start=start)


# Posterior analysis
pm.traceplot(trace, varnames=['rho', 'beta'])
pm.summary(trace, varnames=['rho', 'beta']).round(2)
pm.plot_posterior(trace, varnames=['rho', 'beta'])

Answer 1

Thanks to Cam.Davidson.Pilon, I converted the pandas objects into numpy arrays as follows,

 # Expected value of lambda parameter lambda_obs = pm.Deterministic('lambda_obs', \\ tt.exp(beta[0]+beta[1]*data_weibull['temperature'][event].values)) lambda_cens = pm.Deterministic('lambda_cens', \\ tt.exp(beta[0]+beta[1]*data_weibull['temperature'][~event].values)) # Likelihood (sampling distribution) of observations runningtime_obs = pm.Weibull('runningtime_obs', alpha=rho, \\ beta=lambda_obs, observed=data_weibull['runningtime'][event].values) runningtime_cens = pm.Potential('runningtime_cens', \\ weibull_lccdf(data_weibull['runningtime'][~event].values, rho, \\ lambda_cens))