Prescient

Prescient is a automated forecasting tool. Prescient is based on Facebook Prophet API and configurable tool with several parameters related with Facebook Prophet API.

How it works

python Prescient.py [config_filename.yml]

Example:

python Prescient.py config.yml

Prescient Config

# Prescient Example Config
# Created by Mutlu Polatcan
# 01.02.2018

################################### REQUIRED ATTRIBUTES #####################################
prescient.dataset.filepath: "../datasets/energy-demand-forecasting/zone_1_energy_demand.csv"

prescient.training.data.percent.min: 70
prescient.training.data.percent.max: 80
prescient.training.data.percent.increment.by: 5

prescient.interval.width.min: 0.05
prescient.interval.width.max: 0.85
prescient.interval.width.increment.by: 0.05

prescient.changepoint.prior.scale.min: 0.001
prescient.changepoint.prior.scale.max: 0.5
prescient.changepoint.prior.scale.increment.by: 0.001

prescient.predict.next: 7
prescient.predict.freq: "D"

prescient.convergence.detection.acr.threshold: -0.5 # must be negative
prescient.convergence.detection.measure.number: 3

prescient.paralellism: 3

prescient.holiday.weekends.enabled: true

################################### OPTIONAL ATTRIBUTES #######################################
prescient.holiday.special.days:
  - "2017-01-30"
  - "2017-02-28"

Config Attributes and Details

---

Training Data Percentage

Prescient is automated forecasting tool and so that, it wants training data percentage range from you. Prescient automatically creates training files according to this percentage range and use these files for training models.

NOTE !!! Valid values for this attribute is between 0 and 100

• prescient.training.data.percent.min (Type: double): Minimum percentage of training data.
• prescient.training.data.percent.max (Type: double): Maximum percentage of training data.
• prescient.training.data.percent.increment.by (Type: double): Incrementation amount of percentage of training data.

---

Interval Width

The biggest source of uncertainty in the forecast is the potential for future trend changes. Again, these interval assume that the future will see the same frequency and magnitude of rate changes as the past. we assume that the future will see similar trend changes as the history. In particular, we assume that the average frequency and magnitude of trend changes in the future will be the same as that which we observe in the history. We project these trend changes forward and by computing their distribution we obtain uncertainty intervals. This assumption is probably not true, so you should not expect to get accurate coverage on these uncertainty intervals. One property of this way of measuring uncertainty is that allowing higher flexibility in the rate, by increasing prescient.changepoint.prior.scale, will increase the forecast uncertainty. This is because if we model more rate changes in the history then we will expect more in the future, and makes the uncertainty intervals a useful indicator of overfitting.

• prescient.interval.width.min (Type: double): Minimum uncertainty interval width
• prescient.interval.width.max (Type: double): Maximum uncertainty interval width
• prescient.interval.width.increment.by (Type: double): Incrementation amount of uncertainty interval width

---

Changepoint Prior Scale

Facebook Prophet API

• prescient.changepoint.prior.scale.min: (Type: double):
• prescient.changepoint.prior.scale.max: (Type: double):
• prescient.changepoint.prior.scale.increment.by: (Type: double):

---

Prediction Amount and Frequency

• prescient.predict.next (Type: integer): How far do you want to see future ?

• prescient.predict.freq (Type: string): Prediction frequency for Prescient.

    NOTE !!! Valid frequencies are 'H', 'D', 'M', 'Y'
  

---

Parallelism

Prescient trains models and make predictions on these models concurrently. So that, there is attribute for how many models train and make predictions on these models concurrently.

• prescient.paralellism (Type: integer): Parallelism amount of your CPU. If you have a quad core CPU and assign 4 to this parameter in config file, Prescient trains 4 models concurrently. If you increase this parameter, your total training and prediction time will be reduced.

---

Measure Number and Accuracy Change Rate Threshold for Convergence Detection

Prescient is automated forecasting tool and so that it must be find best model for us. Prescient uses simple optimization for this. Prescient looks last n (given by config) model's accuracies and compare it with current best model. If average accuracy change rate, belows user defined threshold, Prescient doesn't try any other parameters.

NOTE !!! Average accuracy change rate threshold must be NEGATIVE

• prescient.convergence.detection.acr.threshold (Type: double): Threshold of average accuracy rate change.
• prescient.convergence.detection.measure.number (Type: integer): Model training and predictions numbers for calculate average accuracy change rate by last n model's accuracies. (Window size)

---

Special Days and Holidays

Prescient also models holiday effects via Facebook Prophet API. Prescient automatically finds weekends in time series data and gives it to the Prophet to train models to modelling weekend effects. Also users can add special holidays for modelling holiday effects of these days. For example, "Christmas", "Ramadan" etc.

• prescient.holiday.weekends.enabled (Type: boolean): Enable or disable Prescient's weekend effect modelling.
• prescient.holiday.special.days (Type: list of date strings): User defined special days for holiday effect modelling.

---