feat: add quantiles to forecast and cross validation methods

[AzulGarza]

this pr adds the quantiles argument to the forecast and cross_validation methods.
this feature will enable users to forecast quantiles directly, instead of transforming the outputs when level is used.
the choice of the column names for the quantile forecasts (TimeGPT-q-{int(100 * q)}) responds to the limitations of distributed engines to handle dots (.) correctly in the columns.
this choice was also included in the docstrings to prevent confusion between quantiles and percentiles.
also a tutorial on how to use the new feature was added.

• feat: add quantiles argument to forecast and cv methods
• feat: add distributed quantile
• feat: add quantile forecasts tutorial
• fix: add correct test pred intervals test

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[github-actions]

Experiment 1: air-passengers

Description:

variable	experiment
h	12
season_length	12
freq	MS
level	None
n_windows	1

Results:

metric	timegpt-1	timegpt-1-long-horizon	SeasonalNaive	Naive
mae	12.6793	11.0623	47.8333	76
mape	0.027	0.0232	0.0999	0.1425
mse	213.936	199.132	2571.33	10604.2
total_time	5.5363	5.4643	0.0093	0.0053

Plot:

Experiment 2: air-passengers

Description:

variable	experiment
h	24
season_length	12
freq	MS
level	None
n_windows	1

Results:

metric	timegpt-1	timegpt-1-long-horizon	SeasonalNaive	Naive
mae	58.1031	58.4587	71.25	115.25
mape	0.1257	0.1267	0.1552	0.2358
mse	4040.21	4110.79	5928.17	18859.2
total_time	6.9567	6.9043	0.0062	0.0053

Plot:

Experiment 3: electricity-multiple-series

Description:

variable	experiment
h	24
season_length	24
freq	H
level	None
n_windows	1

Results:

metric	timegpt-1	timegpt-1-long-horizon	SeasonalNaive	Naive
mae	142.394	196.363	269.23	1331.02
mape	0.0203	0.0234	0.0304	0.1692
mse	63464.7	123119	213677	4.68961e+06
total_time	6.6967	6.522	0.0088	0.009

Plot:

Experiment 4: electricity-multiple-series

Description:

variable	experiment
h	168
season_length	24
freq	H
level	None
n_windows	1

Results:

metric	timegpt-1	timegpt-1-long-horizon	SeasonalNaive	Naive
mae	522.427	353.528	398.956	1119.26
mape	0.069	0.0454	0.0512	0.1583
mse	966294	422332	656723	3.17316e+06
total_time	8.183	5.537	0.0085	0.0078

Plot:

Experiment 5: electricity-multiple-series

Description:

variable	experiment
h	336
season_length	24
freq	H
level	None
n_windows	1

Results:

metric	timegpt-1	timegpt-1-long-horizon	SeasonalNaive	Naive
mae	478.362	361.033	602.926	1340.95
mape	0.0622	0.046	0.0787	0.17
mse	805039	441118	1.61572e+06	6.04619e+06
total_time	10.0357	11.0005	0.0086	0.0078

Plot:

[github-actions]

Experiment 1: air-passengers

Description:

variable	experiment
h	12
season_length	12
freq	MS
level	None
n_windows	1

Results:

metric	timegpt-1	timegpt-1-long-horizon	SeasonalNaive	Naive
mae	12.6793	11.0623	47.8333	76
mape	0.027	0.0232	0.0999	0.1425
mse	213.936	199.132	2571.33	10604.2
total_time	4.4127	4.9495	0.0109	0.0069

Plot:

Experiment 2: air-passengers

Description:

variable	experiment
h	24
season_length	12
freq	MS
level	None
n_windows	1

Results:

metric	timegpt-1	timegpt-1-long-horizon	SeasonalNaive	Naive
mae	58.1031	58.4587	71.25	115.25
mape	0.1257	0.1267	0.1552	0.2358
mse	4040.21	4110.79	5928.17	18859.2
total_time	9.3647	7.5698	0.0061	0.0053

Plot:

Experiment 3: electricity-multiple-series

Description:

variable	experiment
h	24
season_length	24
freq	H
level	None
n_windows	1

Results:

metric	timegpt-1	timegpt-1-long-horizon	SeasonalNaive	Naive
mae	142.394	196.363	269.23	1331.02
mape	0.0203	0.0234	0.0304	0.1692
mse	63464.7	123119	213677	4.68961e+06
total_time	3.9507	5.7009	0.0084	0.0076

Plot:

Experiment 4: electricity-multiple-series

Description:

variable	experiment
h	168
season_length	24
freq	H
level	None
n_windows	1

Results:

metric	timegpt-1	timegpt-1-long-horizon	SeasonalNaive	Naive
mae	522.427	353.528	398.956	1119.26
mape	0.069	0.0454	0.0512	0.1583
mse	966294	422332	656723	3.17316e+06
total_time	5.6801	5.4462	0.0082	0.0077

Plot:

Experiment 5: electricity-multiple-series

Description:

variable	experiment
h	336
season_length	24
freq	H
level	None
n_windows	1

Results:

metric	timegpt-1	timegpt-1-long-horizon	SeasonalNaive	Naive
mae	478.362	361.033	602.926	1340.95
mape	0.0622	0.046	0.0787	0.17
mse	805039	441118	1.61572e+06	6.04619e+06
total_time	6.2763	6.3184	0.0082	0.0078

Plot: