impute_missing_values is not working

[hidesoon]

I just run the example from the official website:

I have encountered one warning and one error:

# Build model using observed time series to set heuristic priors.
linear_trend_model = tfp.sts.LocalLinearTrend(
  observed_time_series=observed_time_series)
model = tfp.sts.Sum([linear_trend_model],
                    observed_time_series=observed_time_series)

# Fit model to data
parameter_samples, _ = tfp.sts.fit_with_hmc(model, observed_time_series)

WARNING:tensorflow:From D:\xxx\xxx\Anaconda3\envs\pdm\lib\site-packages\tensorflow_probability\python\sts\fitting.py:561: SeedStream.init (from tensorflow_probability.python.util.seed_stream) is deprecated and will be removed after 2019-10-01. Instructions for updating: SeedStream has moved to tfp.util.SeedStream. WARNING:tensorflow:From D:\xxx\xxx\Anaconda3\envs\pdm\lib\site-packages\tensorflow_core\python\ops\linalg\linear_operator_diag.py:166: calling LinearOperator.init (from tensorflow.python.ops.linalg.linear_operator) with graph_parents is deprecated and will be removed in a future version. Instructions for updating: Do not pass graph_parents. They will no longer be used. WARNING:tensorflow:From D:\xxx\xxx\Anaconda3\envs\pdm\lib\site-packages\tensorflow_core\python\ops\linalg\linear_operator_block_diag.py:199: LinearOperator.graph_parents (from tensorflow.python.ops.linalg.linear_operator) is deprecated and will be removed in a future version. Instructions for updating: Do not call graph_parents.

# Impute missing values
imputed_series_distribution = tfp.sts.impute_missing_values(
  model, observed_time_series)
print('imputed means and stddevs: ',
      imputed_series_distribution.mean(),
      imputed_series_distribution.stddev())

---

TypeError Traceback (most recent call last) in 1 # Impute missing values 2 imputed_series_distribution = tfp.sts.impute_missing_values( ----> 3 model, observed_time_series) 4 print('imputed means and stddevs: ', 5 imputed_series_distribution.mean(),

TypeError: impute_missing_values() missing 1 required positional argument: 'parameter_samples'

Then I tried to fill the missing argument:

# Impute missing values
imputed_series_distribution = tfp.sts.impute_missing_values(
  model, observed_time_series, parameter_samples=None)
print('imputed means and stddevs: ',
      imputed_series_distribution.mean(),
      imputed_series_distribution.stddev())

---

TypeError Traceback (most recent call last) in 1 # Impute missing values 2 imputed_series_distribution = tfp.sts.impute_missing_values( ----> 3 model, observed_time_series, parameter_samples=None) 4 print('imputed means and stddevs: ', 5 imputed_series_distribution.mean(),

D:\xxx\xxx\Anaconda3\envs\pdm\lib\site-packages\tensorflow_probability\python\sts\forecast.py in impute_missing_values(model, observed_time_series, parameter_samples, include_observation_noise) 449 tf.shape(input=observed_time_series))[-2] 450 lgssm = model.make_state_space_model( --> 451 num_timesteps=num_timesteps, param_vals=parameter_samples) 452 posterior_means, posterior_covs = lgssm.posterior_marginals( 453 observed_time_series, mask=mask)

D:\xxx\xxx\Anaconda3\envs\pdm\lib\site-packages\tensorflow_probability\python\sts\structural_time_series.py in make_state_space_model(self, num_timesteps, param_vals, initial_state_prior, initial_step) 156 return self._make_state_space_model( 157 num_timesteps=num_timesteps, --> 158 param_map=self._canonicalize_param_vals_as_map(param_vals), 159 initial_state_prior=initial_state_prior, 160 initial_step=initial_step)

D:\xxx\xxx\Anaconda3\envs\pdm\lib\site-packages\tensorflow_probability\python\sts\structural_time_series.py in _canonicalize_param_vals_as_map(self, param_vals) 129 param_map = param_vals 130 else: --> 131 param_map = {p.name: v for (p, v) in zip(self.parameters, param_vals)} 132 133 return param_map

TypeError: zip argument #2 must support iteration

Hope the example code from the official instruction page can be fixed.

Many thanks

[davmre]

Ah, good catch; the example should read

imputed_series_distribution = tfp.sts.impute_missing_values(
  model,
  observed_time_series,
  parameter_samples=parameter_samples)

where parameter_samples is as returned from fit_with_hmc.

We'll have a fix pushed soon. Thanks for reporting this!

[hidesoon]

thanks @davmre to point out the solution, it should be a simple one,