Instances¶
This page shows how to load instances of time series. All examples below
expects you to have an initialized instance of the client called eq.
Operations described here are available under eq.instances.*.
Requirements: Use these operations for curves with curve_type set
to any of the following:
INSTANCE
Load instances¶
Method reference: eq.instances.load()
To load multiple instances (typically forecasts), you only need to specify
the curve. By default, it will load 5 instances, but you can increase
(or decrease) this by specifying a limit option. You can also filter
by issue date and tags.
Let’s start by loading the latest instances for the German wind power forecasts:
>>> forecasts = eq.instances.load(
>>> 'DE Wind Power Production MWh/h 15min Forecast'
>>> )
>>> forecasts
[<Timeseries: resolution=<Resolution: frequency=PT15M, timezone=CET>, curve="DE Wind Power Production MWh/h 15min Forecast", instance=<Instance: issued="2020-06-25 12:00:00+00:00", tag="gfs-ens">, begin="2020-06-25 14:00:00+02:00", end="2020-07-11 14:00:00+02:00">,
<Timeseries: resolution=<Resolution: frequency=PT15M, timezone=CET>, curve="DE Wind Power Production MWh/h 15min Forecast", instance=<Instance: issued="2020-06-25 12:00:00+00:00", tag="gfs">, begin="2020-06-25 14:00:00+02:00", end="2020-07-05 14:00:00+02:00">,
<Timeseries: resolution=<Resolution: frequency=PT15M, timezone=CET>, curve="DE Wind Power Production MWh/h 15min Forecast", instance=<Instance: issued="2020-06-25 12:00:00+00:00", tag="ec-ens">, begin="2020-06-25 14:00:00+02:00", end="2020-07-10 14:00:00+02:00">,
<Timeseries: resolution=<Resolution: frequency=PT15M, timezone=CET>, curve="DE Wind Power Production MWh/h 15min Forecast", instance=<Instance: issued="2020-06-25 12:00:00+00:00", tag="ec">, begin="2020-06-25 14:00:00+02:00", end="2020-07-05 14:00:00+02:00">,
<Timeseries: resolution=<Resolution: frequency=PT15M, timezone=CET>, curve="DE Wind Power Production MWh/h 15min Forecast", instance=<Instance: issued="2020-06-25 06:00:00+00:00", tag="gfs-ens">, begin="2020-06-25 08:00:00+02:00", end="2020-07-11 08:00:00+02:00">]
The return type from load() is a
TimeseriesList. This is a subclass of Python’s
built-in list. It has extra validations so that all time series in the list have
the same frequency (hourly, daily, etc.), and it has a method called
to_dataframe which converts the
list of time series to a pandas.DataFrame. See the chapter on
Pandas integration for more details.
Notice that each time series in the list has an instance attribute (with
issued (issue date) and tag). This is what identifies the instances
(or forecasts, if you will):
>>> [f.instance for f in forecasts]
[<Instance: issued="2020-06-25 12:00:00+00:00", tag="gfs-ens">,
<Instance: issued="2020-06-25 12:00:00+00:00", tag="gfs">,
<Instance: issued="2020-06-25 12:00:00+00:00", tag="ec-ens">,
<Instance: issued="2020-06-25 12:00:00+00:00", tag="ec">,
<Instance: issued="2020-06-25 06:00:00+00:00", tag="gfs-ens">]
Energy Quantified has many years of forecasts in the database (non-paying users only get access to instances from the latest 30 days). So, if you would like to load older instances, you can do so! Here is how to load the last instances issued in 2018:
>>> from datetime import datetime
>>> forecasts = eq.instances.load(
>>> 'DE Wind Power Production MWh/h 15min Forecast',
>>> issued_at_latest=datetime(2018, 12, 31, 23, 59, 59) # Last second of 2018!
>>> )
You can also filter by the instance’s tags you would like to load. It is
possible to list more than one tag. Let’s download data for ec and gfs:
>>> forecasts = eq.instances.load(
>>> 'DE Wind Power Production MWh/h 15min Forecast',
>>> tags=['ec', 'gfs'] # Can be a string, too: tags='ec'
>>> )
Combine the parameters issued_at_latest and tags to load the instances
of your liking. There is also an exclude_tags to let you remove certain
tags from the response.
And finally, you can aggregate instances:
>>> from datetime import datetime
>>> from energyquantified.time import Frequency
>>> from energyquantified.metadata import Aggregation, Filter
>>> forecasts = eq.instances.load(
>>> 'DE Wind Power Production MWh/h 15min Forecast',
>>> issued_at_latest=datetime(2020, 6, 1, 0, 0, 0),
>>> tags='ec',
>>> frequency=Frequency.P1D,
>>> aggregation=Aggregation.AVERAGE,
>>> hour_filter=Filter.BASE,
>>> limit=10
>>> )
Get the latest instance¶
Method reference: eq.instances.latest()
You can load the latest instance available like so:
>>> forecast = eq.instances.latest(
>>> 'DE Wind Power Production MWh/h 15min Forecast'
>>> )
>>> forecast
<Timeseries: resolution=<Resolution: frequency=PT15M, timezone=CET>, curve="DE Wind Power Production MWh/h 15min Forecast", instance=<Instance: issued="2020-06-25 18:00:00+00:00", tag="gfs">, begin="2020-06-25 20:00:00+02:00", end="2020-06-26 10:00:00+02:00">
As for the method to load multiple instances, you can put filters on which instance you would like to load:
>>> from datetime import datetime
>>> forecast = eq.instances.latest(
>>> 'DE Wind Power Production MWh/h 15min Forecast',
>>> tags='ec',
>>> issued_at_latest=datetime(2020, 6, 1, 0, 0, 0)
>>> )
Aggregations are supported here, too.
Get a specific instance¶
Method reference: eq.instances.get()
If you know the issue date and tag for an instance, you can load it like seen below. You must always specify the issue date, but you can leave the tag unspecified (which will default to a blank tag).
>>> from datetime import datetime
>>> forecast = eq.instances.get(
>>> 'DE Wind Power Production MWh/h 15min Forecast',
>>> issued=datetime(2020, 6, 1, 0, 0, 0),
>>> tag='ec'
>>> )
>>> forecast.instance
<Instance: issued="2020-06-01 00:00:00+00:00", tag="ec">
Aggregations are supported here, too.
Include ensembles¶
All the above methods — load(), latest() and get() — can also load
scenarios for instances that have these. For instance-based data, we refer to
scenarios as ensembles. The terminology comes from meteorology,
where forecasts with multiple scenarios are called ensemble forecasts.
To load ensembles, add ensembles=True in the parameters.
There is one catch: When loading ensembles, the maximum number of instances you can load at once becomes reduced to 10 due to increased server-side load.
Instances that don’t have ensembles will return a regular, single-valued time series.
In the below example, we are loading the GFS ensemble forecast issued 1 June 2020 at 00:00. And aggregations are supported here, too:
>>> from datetime import datetime
>>> forecast = eq.instances.get(
>>> 'DE Wind Power Production MWh/h 15min Forecast',
>>> issued=datetime(2020, 6, 1, 0, 0, 0),
>>> tag='gfs-ens', # GFS ensemble forecast
>>> frequency=Frequency.P1D,
>>> ensembles=True # Include ensembles
>>> )
>>> forecast.data[:3]
[<MeanScenariosValue: date=2020-06-02 00:00:00+02:00, value=4249.56, scenarios=(4230.24, 4200.12, 3958.99, 4803.86, 5132.65, 4467.72, 5137.52, 4272.63, 3883.69, 3667.21, 4463.02, 4183.24, 4166.79, 4374.41, 3916.84, 3866.79, 3837.91, 4055.36, 3977.33, 4376.41, 4267.8)>,
<MeanScenariosValue: date=2020-06-03 00:00:00+02:00, value=5150.15, scenarios=(5438.17, 5270.41, 4628.31, 4947.27, 5635.71, 5177.4, 4583.76, 5898.94, 5563.79, 4547.67, 5143.17, 5709.71, 5038.66, 4519.17, 4647.19, 4686.25, 5193.25, 5323.04, 5720.27, 5247.36, 5233.52)>,
<MeanScenariosValue: date=2020-06-04 00:00:00+02:00, value=12355.81, scenarios=(11182.13, 11389.47, 9822.78, 10551.62, 12745.04, 10715.13, 15139.99, 11685.89, 11184.46, 10147.47, 12218.74, 14013.28, 13878.11, 11320.92, 17547.07, 10672.34, 13702.91, 9896.63, 13989.7, 15525.05, 12143.3)>]
Relative queries (day-ahead forecasts)¶
Method reference: eq.instances.relative()
When benchmarking models (forecasts), one often would like to know what a forecast was for the day ahead. And you would like to do this over a date interval. For example, we would like to know Monday’s forecast for Tuesday, and Tuesday’s forecast for Wednesday, and so on.
Energy Quantified’s API has solved this by via an operation we call relative forecasts.
The relative forecasts work for 0 or more days ahead:
days_ahead=0means forecasts for intraday
days_ahead=1means forecasts for day ahead
days_ahead=2means forecasts for day after day ahead… and so on
You must filter on the tag, and you can filter on the time-of-day the forecast was issued. When there isn’t any forecast issued for a specific day, then that day will have no values.
>>> from datetime import datetime, time
>>> day_ahead_forecast = eq.instances.relative(
>>> 'DE Wind Power Production MWh/h 15min Forecast',
>>> begin=datetime(2020, 6, 1, 0, 0, 0),
>>> end=datetime(2020, 6, 5, 0, 0, 0),
>>> tag='ec',
>>> days_ahead=1, # The day-ahead forecast (0 or higher allowed)
>>> time_of_day=time(0, 0), # Issued at exactly 00:00
>>> frequency=Frequency.P1D
>>> )
>>> day_ahead_forecast.data
[<Value: date=2020-06-01 00:00:00+02:00, value=10720.75>,
<Value: date=2020-06-02 00:00:00+02:00, value=4144.67>,
<Value: date=2020-06-03 00:00:00+02:00, value=6397.83>,
<Value: date=2020-06-04 00:00:00+02:00, value=12686.8>]
If you don’t know precisely when the forecast was issued, or you would like only to get forecasts issued before a particular time of the day, use the before_time_of_day instead. You can also decide whether to select the earliest or latest issued instance by specifying the issued parameter.
There is also a parameter for after_time_of_day.
Here we select the latest day ahead wind power forecasts issued before 12:00 every day from 1 June to 5 June:
>>> from datetime import datetime, time
>>> day_ahead_forecast = eq.instances.relative(
>>> 'DE Wind Power Production MWh/h 15min Forecast',
>>> begin=datetime(2020, 6, 1, 0, 0, 0),
>>> end=datetime(2020, 6, 5, 0, 0, 0),
>>> tag='ec',
>>> days_ahead=1,
>>> before_time_of_day=time(12, 0), # Issued before 12 o'clock
>>> issued='latest', # Set to "earliest" or "latest"
>>> frequency=Frequency.P1D
>>> )
>>> day_ahead_forecast.data
[<Value: date=2020-06-01 00:00:00+02:00, value=10720.75>,
<Value: date=2020-06-02 00:00:00+02:00, value=4144.67>,
<Value: date=2020-06-03 00:00:00+02:00, value=6397.83>,
<Value: date=2020-06-04 00:00:00+02:00, value=12686.8>]
Aggregations are also supported, as you can see from the examples above.
Next steps¶
Learn how to load time series, period-based series, and period-based series instances.
