TimeSequencePredictor

---

TimeSequencePredictor

TimeSequencePredictor can be used to train a pipeline (including feature engineering and model) for automated time series forecasting in a distributed way. AutoML is applied for searching the best set of features as well as model hyper-parameters.

Methods

__init__

tsp = TimeSequencePredictor(name="automl",
                            logs_dir="~/zoo_automl_logs",
                            future_seq_len=1,
                            dt_col="datetime",
                            target_col="value",
                            extra_features_col=None,
                            drop_missing=True,
                            search_alg=None,
                            search_alg_params=None,
                            scheduler=None,
                            scheduler_params=None,)

Arguments

• name: Name of the experiment.

• logs_dir: Where the automl tune logs file located.

• future_seq_len: Integer. The future sequence length to be predicted. The default value is 1.

• dt_col: The name of datetime column of the input data frame.

• target_col: The name of target column to be predicted of the input data frame.

• extra_features_col: The name of extra features column that needs for prediction of the input data frame.

• drop_missing: Boolean. Whether to drop missing values of the input data frame.

• search_alg: Optional(str). The search algorithm to use. We only support "bayesopt" and "skopt" for now. The default search_alg is None and variants will be generated according to the search method in search space.

• search_alg_params: Optional(Dict). params of search_alg.

• scheduler: Optional(str). Scheduler name. Allowed scheduler names are "fifo", "async_hyperband", "asynchyperband", "median_stopping_rule", "medianstopping", "hyperband", "hb_bohb", "pbt". The default scheduler is "fifo".

• scheduler_params: Optional(Dict). Necessary params of scheduler.

fit

Train a pipeline for time series forecasting. It will return a TimeSequencePipeline object.

tsp.fit(self,
        input_df,
        validation_df=None,
        metric="mse",
        recipe=SmokeRecipe(),
        mc=False,
        resources_per_trial={"cpu": 2},
        )

Arguments

• input_df: Input time series data frame. It could look like:

  datetime	value	...
  2019-06-06	1.2	...
  2019-06-07	2.3	...

• validation_df: validation data frame. It should have the same columns with input_df.

• metric: String. Metric used for train and validation. Available values are "mean_squared_error" or "mean_absolute_error".

• recipe: A Recipe object. Various recipes covers different search space and stopping criteria. Default is SmokeRecipe. Available recipes are SmokeRecipe, RandomRecipe, GridRandomRecipe and BayesRecipe.

• resources_per_trial: Machine resources to allocate per trial, e.g. {"cpu": 64, "gpu": 8}.