forecasting

make_future_dataframe doesn't support regressors currently. So code like:

m = Prophet()
m.add_regressor('var')
m.fit(df)
forecasts = m.predict(m.make_future_dataframe(periods=7))

gives an error like:

ValueError: Regressor 'var' missing from dataframe when attempting to generate forecasts

I know prophet may not know what exact values to put for var in each of the rows a

Collection of follow-ups to #5827. These can/should be broken out into individual PRs. Many are relatively straightforward and would make a good first PR.

General

• Documentation (none was added in original PR).
• Release notes.
• Example notebook.
• Double-check how sm.tsa.arima.ARIMA works with fix_params (it should fail except when the fit method is statespace

Is your feature request related to a problem? Please describe.

It is useful to record the time estimators take to fit to help with algorithm comparison. We do this in clustering and classification base class, it would be nice to store transform time in the BaseTransformer class

Describe the solution you'd like
a class variable, e.g. transform_time, that is initialised in reset and c

The parameters of the likelihood model assume invalid values during training.

See also fixes needed in #1893 because of this, and issue referenced therein.

To Reproduce

import mxnet as mx
from gluonts.mx.distribution.distribution import softplus
softplus(mx.nd, mx.nd.array([-20]))

Error message or code output

[-1.9073486e-06]
<NDArray 1 @cpu(0)>

Environment

• Operating system: macos
• Python version: 3.7.8
• GluonTS ve

How to get result? y cloumn and yhat are NAN

Is your feature request related to a problem? Please describe.

It would be nice to directly support simulating from a fitted ARIMA model, e.g. to have a simulate method to call that would delegate to statsmodels.tsa.arima.model.ARIMA.simulate. Right now, the only way I found is to use arima_res_ member of the fitted object.

Describe the solution you'd like

Class `pmdarima.arima.ar

We have a lot of antiquated docstrings that don't render well into ReadTheDocs. A kind of grunge (but incredibly useful) task would be to refactor these docstrings into proper ReadTheDocs format. This would allow us to render them effectively...

right now the mase (numpy version) returns different values than MASELoss (pytorch version)


the numpy version averages both delta_y and scale before dividing, while the pyto

Is your feature request related to a problem? Please describe.
The cross_validation method has the argument input_size; forecast should also have it.

🚀 Feature Request

Change "zero" strategy in _OneSegmentTimeSeriesImputerTransform so it would allow to choose value that will be used to fill in NaNs.

Proposal

1. In class ImputerMode add new strategy called "constant"
   https://github.com/tinkoff-ai/etna/blob/687a1a9f955d87fe8fbc8c379d678e53aef94c24/etna/transforms/missing_values/imputation.py#L12-L19

2. Add new strategy "cons