TimeSequencePipeline

TimeSequencePipeline integrates feature engineering and time sequence model into a data analysis pipeline. You can get an TimeSequencePipeline object after calling TimeSequencePredictor.fit() or load_ts_pipeline.

Methods

load_ts_pipeline

ts_pipeline = load_ts_pipeline(file)

Arguments

file: saved ts pipeline file.

describe

Since you may get your ts_pipeline from a saved file which is a result pipeline of TimeSeqencePredictor. You can use describe method to get the initialization info for the TimeSeqencePredictor, including future_seq_len, dt_col, target_col, extra_features_col, drop_missing.

ts_pipeline.describe()

fit

Used for incremental fitting. Note that fit in TimeSequencePipeline doesn`t run in distributed mode.

ts_pipeline.fit(input_df, validation_df=None, mc=False, epoch_num=20)

Arguments

input_df: Input time series data frame. It should have the same datetime column, target column, extra features columns (if there are) name with the names you got from ts_pipeline.describe().
validation_df: The validation data frame. It should have the same columns with input_df.
mc: Boolean. Whether to use Monte Carlo Dropout to predict with uncertainty. Default is False. You can refer to the paper here for more details about Monte Carlo Dropout.
epoch_num: Integer. Number of epochs to run incremental fitting.

evaluate

ts_pipeline.evaluate(input_df,
                     metrics=["mse"],
                     multioutput=`raw_values`)

Arguments

input_df: Input data frame. It should have the same column names as the data frame you used to train your TimeSequencePredictor. And you can get the information with pipeline.describe()
metrics: The metrics you want to use for evaluation. The available metrics are me, mae, mse, rmse, msle, r2, mpe, mape, mspe, smape, mdape and smdape.

multioutput: string in [raw_values, uniform_average]

        - `raw_values` : Returns a full set of errors.
        - `uniform_average` : Errors of all outputs are averaged with uniform weight.

predict

ts_pipeline.predict(input_df)

Arguments

input_df: Input data frame to do prediction.It should have the same column names as the data frame you used to train your TimeSequencePredictor. And you can get the information with pipeline.describe()
```
    - a TFDataset object
    - A Numpy array (or array-like), or a list of arrays
       (in case the model has multiple inputs).
    - A dict mapping input names to the corresponding array/tensors,
```

save

ts_pipeline.save(ppl_file="my.ppl")

Arguments

ppl_file: The file name you want to save your pipeline.

config_save

ts_pipeline.config_save(config_file=="my.json")

Arguments

config_file: The config file name that you want to save all the pipeline configs in.