The setup.py file reads in the config values from settings.ini. So, absence of settings.ini file in the source distribution (*.tar.gz file), leads to failure in installation of statsforecast.

• [x] Currently (v0.3.0) the PyPI source does not include the settings.ini file. This PR fixes that.
  
• [ ] ~~Changes to README.md file~~:
  • [ ] ~~Fixed some of the formatting errors~~.
  • [ ] ~~Fixed some broken URLs~~.

    The relative URLs do not render properly on PyPI. Converted them from relative to absolute URLs.

Closes #24

Hi guys,

I was playing a little with ETS to see whether we could include it in Darts. For the time being I'm having a hard time to have it outperform statsmodels in terms of runtime (I haven't looked at accuracy). Is there any special trick (or a special regime to be in) in order for the statsforcast version to run faster?

Here is the small benchmark I ran:

import numpy as np
import time

from statsforecast.models import ETS as SF_ETS
import statsmodels.tsa.holtwinters as hw

values = np.random.rand(100,)

# First, we make a dry run for jit:
model_sf = SF_ETS()
model_sf.fit(values)
_ = model_sf.predict(10)

model_sm = hw.ExponentialSmoothing(values)
sm_res = model_sm.fit()
_ = sm_res.forecast(10)

# Now, a small benchmark:
tic = time.time()
for _ in range(100):
    model_sf = SF_ETS()
    model_sf.fit(values)
    _ = model_sf.predict(10)
    
print("Time taken by SF ETS : {:.2f} s.".format(time.time() - tic))

tic = time.time()
for _ in range(100):
    model_sm = hw.ExponentialSmoothing(values)
    sm_res = model_sm.fit()
    _ = sm_res.forecast(10)
    
print("Time taken by SM ETS : {:.2f} s.".format(time.time() - tic))

And I get results like

Time taken by SF ETS : 1.54 s.
Time taken by SM ETS : 0.38 s.

This PR adds "plotly-resampler" as option for the plotting engine of StatsForecast.plot method, see #342

Would love to hear your feedback on this!

Example use of plotly-resampler (in the ElectricityLoadForecasting.ipynb notetbook)

Note that plotly-resampler properties can be passed in the resampler_kwargs argument. :arrow_down: illustrates how for example the number of shown samples can be changed!

StatsForecast.plot(df, engine="plotly-resampler", resampler_kwargs={"default_n_shown_samples": 3000, "show_dash": {"mode": "inline"}})

PS: also fixed a minor typo in the CONTRIBUTING.md file

The core notebook created a logging config when the notebook was ran, but also when the generated core.py file was imported. This prevented anyone who imports statsforecast from setting their own logging config in the most commonly used way.

More details in the linked issue. Fixes Nixtla/statsforecast#275

Describe the bug I'm trying to follow the "Getting started" example with my own data, which happens to be sampled hourly. So I set season_length = 365 * 24, but got the following error:

...

File /opt/conda/lib/python3.9/site-packages/statsforecast/models.py:245, in ETS.forecast(self, y, h, X, X_future, fitted)
    237 def forecast(
    238         self, 
    239         y: np.ndarray, # time series
   (...)
    243         fitted: bool = False, # return fitted values?
    244     ):
--> 245     mod = ets_f(y, m=self.season_length, model=self.model)
    246     fcst = forecast_ets(mod, h)
    247     keys = ['mean']

File /opt/conda/lib/python3.9/site-packages/statsforecast/ets.py:937, in ets_f(y, m, model, damped, alpha, beta, gamma, phi, additive_only, blambda, biasadj, lower, upper, opt_crit, nmse, bounds, ic, restrict, allow_multiplicative_trend, use_initial_values, maxit)
    935 if m > 24:
    936     if seasontype in ['A', 'M']:
--> 937         raise ValueError('Frequency too high')
    938     elif seasontype == 'Z':
    939         warnings.warn(
    940             "I can't handle data with frequency greater than 24. " 
    941             "Seasonality will be ignored."
    942         )

ValueError: Frequency too high

If I resample daily and set season_length = 365, I too get a ValueError: Frequency too high. Same goes with a weekly resample and season_length = 52.

I only have 2 incomplete years of data, so I can't resample monthly: I get various versions of

ValueError: order must be 3 non-negative integers, got (0, 0, 0)

To Reproduce (I can work on a reproducer if it helps, it will take me some time)

Expected behavior The models work with higher frequencies.

Desktop (please complete the following information):

• OS: Ubuntu 22.04
• Browser: Firefox
• Version: 1.0.0

Hi, great work on statsforecasts! This package looks very nice. I'm considering integrating a couple of the models in Darts (https://github.com/unit8co/darts). I'm wondering about your future plans - do you intend to maintain this package on the long term? How likely can we expect API changes in the future releases?

Also as a side note - I took a quick look at the Croston method, and it looks like the method accepts h and future_xreg, which I'm not sure is intended as those are not used.

In general I think slightly more documentations on your different models could be helpful for users :)

I'm currently trying to perform some forecastings on a set of daily time series and I was wondering whether is there a way to get the predictions on the training data, that are used to compute the residuals (difference between actual and predictions in the train). In StatsForecast class there is no possibility for doing that. I'm mainly interested to obtain them with auto_arima approach, but it could be extended also for the remaining approaches.

Is it possible to add a method or attribute to get them?

Thank you

Hi! I was wondering if you have implemented (or planning to implement) a model summary table for the ARIMA model that contain the coefficients, their p-values, etc.?

Like https://www.statsmodels.org/dev/generated/statsmodels.tsa.arima.model.ARIMAResults.summary.html

Many thanks!

This is how the docs render locally for me with the latest changes to the readme. I believe this is due to the embedded html, maybe we can try to achieve that formatting without using html.

I get ValueError: math domain error cause by tmp['bic'] = tmp['aic'] + npar*(math.log(nstar) - 2) from statsforecast/arima.py", line 1225,

I guess nstar is not > 0

I am trying to solve a timeseries problem with intermittent zero demand in the timeframe(Monthly data). I am getting this warning/error.

/opt/conda/lib/python3.7/site-packages/statsforecast/arima.py:866: RuntimeWarning: divide by zero encountered in log return 0.5 * np.log(res) /opt/conda/lib/python3.7/site-packages/statsforecast/ets.py:443: RuntimeWarning: divide by zero encountered in double_scalars l0 = l0 / b0 /opt/conda/lib/python3.7/site-packages/statsforecast/ets.py:443: RuntimeWarning: divide by zero encountered in double_scalars l0 = l0 / b0 /opt/conda/lib/python3.7/site-packages/statsforecast/ets.py:448: RuntimeWarning: invalid value encountered in float_scalars b0 = max(y_sa[1] / y_sa[0], 1e-3) /opt/conda/lib/python3.7/site-packages/statsforecast/ets.py:448: RuntimeWarning: invalid value encountered in float_scalars b0 = max(y_sa[1] / y_sa[0], 1e-3) /opt/conda/lib/python3.7/site-packages/statsforecast/ets.py:443: RuntimeWarning: divide by zero encountered in double_scalars l0 = l0 / b0 /opt/conda/lib/python3.7/site-packages/statsforecast/ets.py:448: RuntimeWarning: invalid value encountered in float_scalars b0 = max(y_sa[1] / y_sa[0], 1e-3) /opt/conda/lib/python3.7/site-packages/statsforecast/ets.py:443: RuntimeWarning: divide by zero encountered in double_scalars l0 = l0 / b0 /opt/conda/lib/python3.7/site-packages/statsforecast/ets.py:448: RuntimeWarning: invalid value encountered in float_scalars b0 = max(y_sa[1] / y_sa[0], 1e-3) /opt/conda/lib/python3.7/site-packages/statsforecast/ets.py:448: RuntimeWarning: invalid value encountered in double_scalars b0 = max(y_sa[1] / y_sa[0], 1e-3)

and throws an error showing

Exception: no model able to be fitted

Any thoughts on how this can be resolved? How can I use your package for this?

Regards Shravan

What happened + What you expected to happen

I wanted to try distributed computation using dask. I followed the docs https://nixtla.github.io/statsforecast/distributed.fugue.html and it's working fine using dask local client.

However, when I attempted to use a remote dask cluster, I got empty result:

from dask.distributed import Client
from fugue_dask import DaskExecutionEngine
from statsforecast import StatsForecast
from statsforecast.models import Naive
from statsforecast.utils import generate_series
from statsforecast.distributed.fugue import FugueBackend
import pandas as pd

# Instantiate FugueBackend with DaskExecutionEngine
dask_client = Client('tcp://***:***')
engine = DaskExecutionEngine(dask_client=dask_client)
remote_backend = FugueBackend(engine=engine, as_local=True)
Y_df = pd.read_parquet('https://datasets-nixtla.s3.amazonaws.com/m4-hourly.parquet')


from statsforecast.models import (
    AutoARIMA,
    HoltWinters,
    CrostonClassic as Croston, 
    HistoricAverage,
    DynamicOptimizedTheta as DOT,
    SeasonalNaive
)


# Create a list of models and instantiation parameters
models = [
    AutoARIMA(season_length=24),
    HoltWinters(),
    Croston(),
    SeasonalNaive(season_length=24),
    HistoricAverage(),
    DOT(season_length=24)
]

# Instantiate StatsForecast class as sf
sf = StatsForecast(
    df=Y_df, 
    models=models,
    freq='H', 
    n_jobs=-1,
    fallback_model = SeasonalNaive(season_length=7),
    backend=remote_backend
)

forecasts_df = sf.forecast(h=48, level=[90])
print(forecasts_df.size) # returns 0

I didn't get any errors, but the result is empty

Versions / Dependencies

Local setup:

OS: mac Python: 3.10 py packages:

• dask==2022.12.1
• dask-cloudprovider==2022.10.0
• datasetsforecast==0.0.7
• distributed==2022.12.1
• fugue==0.7.3
• fugue-sql-antlr==0.1.1
• s3fs==2022.11.0
• statsforecast==1.4.0
• pandas==1.5.2

Dask cluster:

• scheduer & worker: dask, version 2022.12.1
• Python: 3.8

Reproduction script

• Create a remote dask cluster
• Use the code above to connect to dask scheduler, and run the sample

Issue Severity

High: It blocks me from completing my task.

Description

When using StatsForecast.forecast_fitted_values() while the backend parameter in the StatsForecast object is set to my Fugue backend, I get the following error: NotImplementedError: Execution with a distributed backend only supports forecast and cross_validation methods. Try setting the backend parameter to None.

Use case

I would like to use StatsForecast.forecast_fitted_values() with my Fugue backend so I can use the fitted values in reconciliation approaches provided by the hierarchicalforecast package, while training the models for the hierarchy quickly using Spark.

Description

Currently AutoARIMA does support exogenous variables. Which is great because it allows for including some external information into your model without losing yourself in complexity or giving up on the effectiveness of 'simple' statistical techniques.

In addition to this it might prove worth while to add this feature to AutoETS.

There's this paper in which they implement it: https://www.monash.edu/business/econometrics-and-business-statistics/research/publications/ebs/wp02-15.pdf

Additionally, in this post Stephen Kolassa suggests a couple of ways of implementing it, among which doing it similarly to AutoARIMA (regression with ARMA errors): https://stats.stackexchange.com/questions/220830/holt-winters-with-exogenous-regressors-in-r

Hyndman also shares some of his thought on the problem (old post though): https://robjhyndman.com/hyndsight/ets-regressors/

Maybe, we can get Osman to share his code with us.

Could be that there's other people implementing it too. But I haven't come across it yet.

Use case

No response

Description

Models in statsforecast, mlforecast, and neuraforecast currently can only receive train_y but not train_ds.

Other models such as uber/orbit can process arbitrary sparse training data (e.g. train_ds=[2010, 2011, 2014, 2016]).

Use case

I am trying to build an adapter for the orbit API to be integrated into the nixtla model ecosystem (implement custom adapter class with the fit, predict)

What happened + What you expected to happen

When trying to use the FugueBackend class to distribute my exogenous forecasts, I encounter the following error

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<command-2016733390302825> in <cell line: 13>()
     11 spark = SparkSession.builder.getOrCreate()
     12 backend = FugueBackend(spark, {"fugue.spark.use_pandas_udf":True})
---> 13 forecasts = forecast(
     14     spark.createDataFrame(df),
     15     models=[ETS()],

/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.9/site-packages/statsforecast/distributed/utils.py in forecast(df, models, freq, h, X_df, level, parallel)
     20 ):
     21     backend = parallel if parallel is not None else ParallelBackend()
---> 22     return backend.forecast(df, models, freq, h=h, X_df=X_df, level=level)
     23 
     24 # %% ../../nbs/distributed.utils.ipynb 6

/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.9/site-packages/statsforecast/distributed/fugue.py in forecast(self, df, models, freq, **kwargs)
     81         schema = schema + ",AutoARIMA_lo_99:float, AutoARIMA_hi_99:float"
     82         print(schema)
---> 83         return transform(
     84             df,
     85             self._forecast_series,

/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.9/site-packages/fugue/interfaceless.py in transform(df, using, schema, params, partition, callback, ignore_errors, engine, engine_conf, force_output_fugue_dataframe, persist, as_local, save_path, checkpoint)
    134     else:
    135         src = dag.df(df)
--> 136     tdf = src.transform(
    137         using=using,
    138         schema=schema,

/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.9/site-packages/fugue/workflow/workflow.py in transform(self, using, schema, params, pre_partition, ignore_errors, callback)
    539         if pre_partition is None:
    540             pre_partition = self.partition_spec
--> 541         df = self.workflow.transform(
    542             self,
    543             using=using,

/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.9/site-packages/fugue/workflow/workflow.py in transform(self, using, schema, params, pre_partition, ignore_errors, callback, *dfs)
   1954         tf._has_rpc_client = not isinstance(callback, EmptyRPCHandler)  # type: ignore
   1955         tf.validate_on_compile()
-> 1956         return self.process(
   1957             *dfs,
   1958             using=RunTransformer,

/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.9/site-packages/fugue/workflow/workflow.py in process(self, using, schema, params, pre_partition, *dfs)
   1615             using = _PROCESSOR_REGISTRY.get(using)
   1616         _dfs = self._to_dfs(*dfs)
-> 1617         task = Process(
   1618             len(_dfs),
   1619             processor=using,

/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.9/site-packages/fugue/workflow/_tasks.py in __init__(self, input_n, processor, schema, params, pre_partition, deterministic, lazy, input_names)
    314     ):
    315         self._processor = _to_processor(processor, schema)
--> 316         self._processor._params = ParamDict(params)
    317         self._processor._partition_spec = PartitionSpec(pre_partition)
    318         self._processor.validate_on_compile()

/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.9/site-packages/triad/collections/dict.py in __init__(self, data, deep)
    175     def __init__(self, data: Any = None, deep: bool = True):
    176         super().__init__()
--> 177         self.update(data, deep=deep)
    178 
    179     def __setitem__(  # type: ignore

/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.9/site-packages/triad/collections/dict.py in update(self, other, on_dup, deep)
    262         for k, v in to_kv_iterable(other):
    263             if on_dup == ParamDict.OVERWRITE or k not in self:
--> 264                 self[k] = copy.deepcopy(v) if deep else v
    265             elif on_dup == ParamDict.THROW:
    266                 raise KeyError(f"{k} exists in dict")

/usr/lib/python3.9/copy.py in deepcopy(x, memo, _nil)
    144     copier = _deepcopy_dispatch.get(cls)
    145     if copier is not None:
--> 146         y = copier(x, memo)
    147     else:
    148         if issubclass(cls, type):

/usr/lib/python3.9/copy.py in _deepcopy_dict(x, memo, deepcopy)
    228     memo[id(x)] = y
    229     for key, value in x.items():
--> 230         y[deepcopy(key, memo)] = deepcopy(value, memo)
    231     return y
    232 d[dict] = _deepcopy_dict

/usr/lib/python3.9/copy.py in deepcopy(x, memo, _nil)
    144     copier = _deepcopy_dispatch.get(cls)
    145     if copier is not None:
--> 146         y = copier(x, memo)
    147     else:
    148         if issubclass(cls, type):

/usr/lib/python3.9/copy.py in _deepcopy_dict(x, memo, deepcopy)
    228     memo[id(x)] = y
    229     for key, value in x.items():
--> 230         y[deepcopy(key, memo)] = deepcopy(value, memo)
    231     return y
    232 d[dict] = _deepcopy_dict

/usr/lib/python3.9/copy.py in deepcopy(x, memo, _nil)
    170                     y = x
    171                 else:
--> 172                     y = _reconstruct(x, memo, *rv)
    173 
    174     # If is its own copy, don't memoize.

/usr/lib/python3.9/copy.py in _reconstruct(x, memo, func, args, state, listiter, dictiter, deepcopy)
    268     if state is not None:
    269         if deep:
--> 270             state = deepcopy(state, memo)
    271         if hasattr(y, '__setstate__'):
    272             y.__setstate__(state)

/usr/lib/python3.9/copy.py in deepcopy(x, memo, _nil)
    144     copier = _deepcopy_dispatch.get(cls)
    145     if copier is not None:
--> 146         y = copier(x, memo)
    147     else:
    148         if issubclass(cls, type):

/usr/lib/python3.9/copy.py in _deepcopy_dict(x, memo, deepcopy)
    228     memo[id(x)] = y
    229     for key, value in x.items():
--> 230         y[deepcopy(key, memo)] = deepcopy(value, memo)
    231     return y
    232 d[dict] = _deepcopy_dict

/usr/lib/python3.9/copy.py in deepcopy(x, memo, _nil)
    170                     y = x
    171                 else:
--> 172                     y = _reconstruct(x, memo, *rv)
    173 
    174     # If is its own copy, don't memoize.

/usr/lib/python3.9/copy.py in _reconstruct(x, memo, func, args, state, listiter, dictiter, deepcopy)
    268     if state is not None:
    269         if deep:
--> 270             state = deepcopy(state, memo)
    271         if hasattr(y, '__setstate__'):
    272             y.__setstate__(state)

/usr/lib/python3.9/copy.py in deepcopy(x, memo, _nil)
    144     copier = _deepcopy_dispatch.get(cls)
    145     if copier is not None:
--> 146         y = copier(x, memo)
    147     else:
    148         if issubclass(cls, type):

/usr/lib/python3.9/copy.py in _deepcopy_dict(x, memo, deepcopy)
    228     memo[id(x)] = y
    229     for key, value in x.items():
--> 230         y[deepcopy(key, memo)] = deepcopy(value, memo)
    231     return y
    232 d[dict] = _deepcopy_dict

/usr/lib/python3.9/copy.py in deepcopy(x, memo, _nil)
    159                     reductor = getattr(x, "__reduce_ex__", None)
    160                     if reductor is not None:
--> 161                         rv = reductor(4)
    162                     else:
    163                         reductor = getattr(x, "__reduce__", None)

/databricks/spark/python/pyspark/context.py in __getnewargs__(self)
    493     def __getnewargs__(self) -> NoReturn:
    494         # This method is called when attempting to pickle SparkContext, which is always an error:
--> 495         raise RuntimeError(
    496             "It appears that you are attempting to reference SparkContext from a broadcast "
    497             "variable, action, or transformation. SparkContext can only be used on the driver, "

RuntimeError: It appears that you are attempting to reference SparkContext from a broadcast variable, action, or transformation. SparkContext can only be used on the driver, not in code that it run on workers. For more information, see SPARK-5063.

Versions / Dependencies

statsforecast: 1.0.0 fugue: 0.7.3 python: 3.9 OS: Unbuntu

Reproduction script

from statsforecast.distributed.utils import forecast
from statsforecast.distributed.fugue import FugueBackend
from statsforecast.models import ETS
from statsforecast.core import StatsForecast

from pyspark.sql import SparkSession
df = pd.DataFrame({"ds": [1, 2, 3,4,5,6,7,8,9], "y": [1,2,3,4,5,6,7,8,9], "x":[1,2,3,4,5,6,7,8,9]})
df["unique_id"] = 1

X_df = pd.DataFrame({"ds":[4], "x":[4], "unique_id":1})
spark = SparkSession.builder.getOrCreate()
backend = FugueBackend(spark, {"fugue.spark.use_pandas_udf":True})
forecasts = forecast(
    spark.createDataFrame(df),
    models=[ETS()],
    X_df=spark.createDataFrame(X_df),
    h=1,
    freq="D",
    parallel=backend)

Issue Severity

High: It blocks me from completing my task.

Description

Hi! I'm new to this library. I cannot find the possible values for each parameter. For example, in https://nixtla.github.io/statsforecast/models.html#autoarima, I can see the defaults, but if I wanted to modify the ic parameter from 'aicc' to some other value, I don't know what that other value that would be. Is it possible to expose all the possible values for each parameter. Thanks, Victor

Link

No response