reichlab / cdcforecastutils Goto Github PK

Again, similar to FluSight package. This should have the following checks in place for a file, specific to both national/regional and state forecasts:

• check that all locations are valid locations
• check that all targets are valid targets
• check that any loc-target probabilistic predictions sum to 1. Warning if sum is between .9 and 1.1. Error if they are <0.9 or >1.1.
• generate a summary (maybe a table?) of what target/location forecasts (perhaps separate for bin/point forecast types) are present for a given forecast.

• date_start_and_end_to_date_seq
• download_and_preprocess_flu_data
• download_and_preprocess_hosp_data
• download_and_preprocess_state_flu_data
• fetch_delphi_data_multi_issue
• generate_csv_from_submission_df
• get_current_time_in_date_seq
• get_time_from_start_of_season
• get_time_left_in_season
• rowMax
• rowMaxWeek
• trajectories_to_short_term_and_seasonal_binned_distributions

• download_and_preprocess_flu_data
• download_and_preprocess_hosp_data
• download_and_preprocess_state_flu_data
• fetch_delphi_data_multi_issue
• get_time_left_in_season
• get_current_time_in_date_seq
• get_time_from_start_of_season
• date_start_and_end_to_date_seq
• numeric_samples_to_binned_distribution
• categorical_samples_to_binned_distribution
• trajectories_to_binned_distributions

Do it

At the moment verify_entry() will immediately stop at the test where the entry fail, without running the subsequent tests.

We would like to have the ability that verify_entry() can perform all test on this entry, create messages for any tests that the entry does not pass, and then stop the process.

preds <- simulate(
          object = sarimaFit,
          nsim = 1000,
          seed = 1,
          newdata = state_data$unweighted_ili,
          h = time_left_in_season
        )

Should be transformed

preds <- simulate(
          object = sarimaFit,
          nsim = 1000,
          seed = 1,
          newdata = state_data$unweighted_ili,
          h = time_left_in_season + NUM_NAs
        )

similar to in FluSight. Currently, when I read a template using read_csv() it tries to convert "bin" and "value" columns to double. for now, they should be left as character, I think.

currently in vignette