Likelycause2

Likelycause is an utility package that uses several functions to attribute causes to variations. Using a combination of arithmetical decompositions and bayesian techniques, this was built to facilitate the workflow of a data-analyst working for the private sector.

What the package contains

This package has everything built under the likelycause2 module, so all the functions should be called using “likelycause2.”. Currently, we have 1 auxiliary function and 1 causal function.

Auxiliary functions

• likelycause2.last_period: The last period function is a utility function that builds variation variables in a dataframe._

Causal functions

• likelycause2.bayes_suspects: The bayes_suspects function calculates the conditional probabilities of the event and each suspicious causes or a combination of those causes. It also suggests an attribution to each individual cause, by adjusting the intersections of causes

Likelycause2.last_period

Description:

The last period function is a utility function that builds variation variables in a dataframe. Variations are defined between moment t and a moment in the past.

Arguments:

• df (pd.DataFrame): the dataframe
• unique_id (string): unique identifier of each line. Must be unique, and can only be 1 column
• interval (string): what is the interval you want to calculate variations for. Accepts days, weeks and hours
• periods (int): number of periods you want to look back on that interval. For last variations, for example, the argument period would be 1
• date_column (string): the date column in your dateframe. Must be a datetime. To convert, use pandas.to_datetime function
• to_past (list): list of columns you want to calculate the variations for

Returns:

Returns the dataframe that was inputed with additinal columns named v+name of the columns in the to_past argument. Those columns represent the variation of that variable between moment t and t-periods. This variation is calculated as (Variable in moment t)/(Variable in moment t-periods).

Likelycause2.bayes_suspects

Description:

The bayes_suspects function calculates the conditional probabilities of the event and each suspicious causes or a combination of those causes. It also suggests an attribution to each individual cause, by adjusting the intersections of causes

Arguments:

• df (pd.DataFrame): the dataframe
• event (string): name of the column that contains the event that we want to explain
• suspects (list): list with name of the columns that contains the potential causes for what we want to explain
• point (dictionary): dictionary with the point for which we want to calculate the probability. Must be a combination of the cause and all the individual points of suspects

Returns:

Returns a dataframe with all the possible probabilities combinations, and the conditional probabilities:

• name: name of that conditional combination. If it has one event, it represents P(event|a). If it has 2 events it represents P(event1 & event2|a)
• prob_ba: P(cause | event)
• prob_a: P(cause)
• prob_b: P(event)
• pbayes: confitional probability
• pbayes_attribution: suggested probability attribution if we want to attribute to individual causes