Question

我正在尝试使用TensorFlow概率构建结构时间序列模型，这是我的模型的结构： tf .__ version__ =='2.2.0' tfp .__ version__ =='0.10.0'

import tensorflow as tf
from tensorflow_probability import distributions as tfd
import tensorflow_probability as tfp
import numpy as np
ts = np.random.normal(size=365).astype('float32')
var =  np.random.exponential(3,size=400).astype('float32')
# Local linear trend.
local_linear_trend = tfp.sts.LocalLinearTrend(
    observed_time_series=ts.astype('float32'),
    name='LLT'
)
day_of_week = tfp.sts.Seasonal(
    num_seasons=7,
    num_steps_per_season=1,
    observed_time_series=ts.astype('float32'),
    name='day_of_week',
)
month_of_year = tfp.sts.Seasonal(
    num_seasons=12,
    num_steps_per_season=30,
    observed_time_series=ts.astype('float32'),
    name='month_of_year',
)
yearweek = tfp.sts.Seasonal(
    num_seasons=52,
    num_steps_per_season=7,
    observed_time_series=ts.astype('float32'),
    name='yearweek',
)
external_var = tf.cast(tf.reshape(var,(-1,1)),dtype=tf.float32)
x_var = tfp.sts.LinearRegression(
    design_matrix=external_var,
    weights_prior=tfd.Exponential(rate=2, validate_args=True),
    name='external_factor')

model_components_base = [
    local_linear_trend,
    day_of_week,
    month_of_year,
    x_var
]
model_base = tfp.sts.Sum(
    components=model_components_base,
    observed_time_series=ts
)
optimizer = tf.optimizers.Adam(learning_rate=.1)
variational_posteriors = tfp.sts.build_factored_surrogate_posterior(
    model = model_base,
    seed=1234
)
target_log_prob_fn = model_base.joint_log_prob(
    observed_time_series=ts.astype('float32')
)
elbo_loss_curve = tfp.vi.fit_surrogate_posterior(
    target_log_prob_fn=target_log_prob_fn,
    surrogate_posterior=variational_posteriors,
    optimizer=optimizer,
    num_steps=150,
    seed=1234
)

请注意，对于外部因素，我已经将权重先验参数设置为指数分布，以确保获得正的权重。运行脚本后，出现此错误

tensorflow.python.framework.errors_impl.InvalidArgumentError:  assertion failed: [Sample must be non-negative.] [Condition x >= 0 did not hold element-wise:] [x (monte_carlo_variational_loss/build_factored_surrogate_posterior_JointDistributionNamed/sample/build_factored_surrogate_posterior_external_factor__weights_posterior/sample/Reshape:0) = ] [[-0.00462905131]]
     [[{{node monte_carlo_variational_loss/expectation/external_factor_external_factor_identityExponential/log_prob/Exponential_2/log_prob/assert_non_negative/assert_less_equal/Assert/AssertGuard/else/_1/Assert}}]] [Op:__inference_train_loop_body_17946]

Function call stack:
train_loop_body
ERROR:tensorflow:==================================
Object was never used (type <class 'tensorflow.python.ops.tensor_array_ops.TensorArray'>):
<tensorflow.python.ops.tensor_array_ops.TensorArray object at 0x14365ca90>
If you want to mark it as used call its "mark_used()" method.
It was originally created here:
  File "...venv/lib/python3.7/site-packages/tensorflow/python/ops/control_flow_ops.py", line 2770, in while_loop
    return result  File ".../venv/lib/python3.7/site-packages/tensorflow/python/ops/control_flow_ops.py", line 2718, in <lambda>
    body = lambda i, lv: (i + 1, orig_body(*lv))  File "..../venv/lib/python3.7/site-packages/tensorflow/python/ops/functional_ops.py", line 648, in compute
    return (next_i, flat_a_out, tas)  File "...umm/venv/lib/python3.7/site-packages/tensorflow/python/ops/functional_ops.py", line 643, in <listcomp>
    tas = [ta.write(i, value) for (ta, value) in zip(tas, flat_a_out)]  File "....venv/lib/python3.7/site-packages/tensorflow/python/util/tf_should_use.py", line 237, in wrapped
    error_in_function=error_in_function)
==================================

权重先验tensorflow.python.framework.errors_impl.InvalidArgumentError：断言失败

0 个答案: