Question

要学习PyMC，我正在尝试做一个简单的隐马尔可夫模型，如下所示：

with pymc3.Model() as hmm:
    # Transition "matrix"
    a_t = np.ones(num_states)
    T = [pymc3.Dirichlet('T{0}'.format(i), a = a_t,  shape = num_states) for i in xrange(num_states)]
    # Emission "matrix"
    a_e = np.ones(num_emissions)
    E = [pymc3.Dirichlet('E{0}'.format(i), a = a_e,  shape = num_emissions) for i in xrange(num_states)]
    # State models
    p0 = np.ones(num_states) / num_states
    # No shape, so each state is a scalar tensor
    states = [pymc3.Categorical('s0', p = p0)]
    emissions = [pymc3.Categorical('z0',
                               p = ifelse(eq(states[0], 0), E[0], ifelse(eq(states[0], 1), E[1], E[2])),
                               observed = example_observation[0])]
    for i in xrange(1, num_times):
        states.append(pymc3.Categorical('s{0}'.format(i),
                                    p = ifelse(eq(states[i-1], 0), T[0], ifelse(eq(states[i-1], 1), T[1], T[2]))))
        emissions.append(pymc3.Categorical('z{0}'.format(i),
                                    p = ifelse(eq(states[i], 0), E[0], ifelse(eq(states[i], 1), E[1], E[2])),
                                    observed = example_observation[i]))

我认为这个模型应该是正确的，但是当我尝试从这个模型中采样时，我得到一个非常奇怪的InvalidValueError：

InvalidValueError: InvalidValueError
        type(variable) = TensorType(float32, scalar)
        variable       = TensorConstant{-inf}
        type(value)    = <type 'numpy.ndarray'>
        dtype(value)   = float32
        shape(value)   = ()
        value          = -inf
        min(value)     = -inf
        max(value)     = -inf
        isfinite       = False
        client_node    = None
        hint           = Graph Input 'TensorConstant{-inf}' has invalid value -inf
        specific_hint  = none
        context        = ...
  TensorConstant{-inf} [@A]

我上传了ipython notebook with the complete code。关于我做错了什么的暗示？

PyMC3中隐马尔可夫模型的问题

0 个答案: