Day outside of the allowed range for weather prediction.
Correcting Day
Correcting Initial Day to have a five day period.
How to use the Component
The component is divided in two diferent columns: the column of the left hand side has several controls to
visualize data including a menu, two tables and a calendar, and the column of the right hand side has a map and
and area to visualize charts. We describe now these components. The controls on the left column include the following:
Models Table: The models table, located on the top of the left column, show information about the characteristics and
the status of the models that are or have been trainned. In the down side of the table there is one button that allows to create new models.
In addition on the bottom-right side there is a button to reload (update) the data of the table.
This last button is important because inmediately after you submit the creation of a new model this one does NOT
show up in the table, the user should press the reload button to be able to see it.
Calendar: The calendar is used to choose the day to be predicted, or the starting day of the period to be predicted.
This is used once the models have been trainned. By default the selected day corresponds to the current day. Notice that if a prediction
that involves weather is required only a few days (4 days) in the future are susceptible to be predicted. In that particular case the not
allowed days are grey out showing that it cannot be selected.
Matrix Output: Shows the predicted matrix at a future instance of time, actual hour can be chosen by a slider that can be found
inmediately above this matrix.
On the map of the right hand side 3 buttons can be found:
Load Points:Allows to visualize the bike stations on the map.
Create Voronoi: Generate Voronoi areas from the set of points previously loaded, by default these are the locations of the
bike stations.
Load Areas:This allows the user to specify the zonning for the OD Matrix computation by providing a geojson. Notice that the geojson
Features should have an integer "zone_id" property that is used for identify each of the zones. These "zone_id" properties should
start in 1 and should be in consective order (i.e. no gaps are allowed). Right now this buttom only allows the visualization of the areas,
it does not allow to upload new areas.
Create/Train Model
The trainning of a model can be initiated by pressing the New Model
button located inside the model table. This opens up a new window with two menus one for the areas available in the platform another one for specifiying the number
of features to be consider for the model:
Areas: The identification of the zonning to be used for the computation. Right now the only available zonning are the "districts".
Num Features: number of features considered to create the model:
1: Considers only the time of the day, Hour of the Day (HoD). We call it the *slot* of the day, this corresponds to the time interval
of the day, considering the measuring period of the given sensor.
2: In addition to the previous considers the day of the week (DoW).
3: In addition to the previous considers if the given day is a School day and/or a Holiday or not (SaH).
4: In addition to the previous considers the metereological variables (MET).
5: In addition to the previous considers events.
Once this information is properly set, the creation of the predicting model is started by pressing the Submit button. The information related
to the models that have been created is shown in the Models Table.
The information that is shown for each model is:
Model: Unique model Id that identifies each model.
#: Number of features.
Areas:The name of the zonning used for the computation.
Status: Status of the trainning phase:
0: completed
-1: trainning still in process
Score: Score, quality, of model, using the test data set. This value is better the closer to 1.
Perform Prediction
Once the status of a model is equal to zero the model can be used in order to perform predictions.
The prediction can be performed for a full day, for 4 days or the user can choose a custom period.
The result of the prediction is shown both in matrix form below the calendar for a instant of time and in the graph below the
map showing the actual prediction of the confidence
interval associated with the prediction. The prediction is performed for the day selected in the calendar
or starting that day if the period is longer.
The result is shown in matrix form on the left bottom part of the module. The matrix has as many rows and columns
as areas in your zonning geojson, for districts zonning it is 8. Moving the slider on top of the matrix the time
of the prediction can be changed, hour by hour, and by clicking on one of the cells of the matrix the prediction
for the whole time period will be shown on the chart on the bottom right hand side.