17/replication_package/code/lib/data_helpers/builder_utils.py

import pandas as pd
import os
import shutil
import re
from functools import reduce
from datetime import datetime, timedelta

from lib.experiment_specs import study_config
from lib.data_helpers import data_utils

"""loads the phone data config from the provided config path"""

class BuilderUtils():

    def get_config(self, config_path):
        if os.path.isfile(config_path):
            pd_config_df = pd.read_csv(config_path,index_col= "index")
            pd_config_dict = pd_config_df.to_dict(orient = 'index')
            return pd_config_dict
        else:
            return {}

    """
    - Purpose: transports zipped files from PhoneDashboardPort and PCPort to the PhoneAddictionDropbox to the specified directory
    - Inputs:
        - port: specifies location of the port 
        - keyword: specifies the kind of inport from the source (e.g. budget, use, etc). the keyword must be in the file name for the function to work
        - new_directory: the directory where the files will be transported
    - """
    def transport_new_zip_files(self,port,keyword,new_directory):
        new_adds = []
        added_files = os.listdir(new_directory)
        empty_files_dir = os.listdir(os.path.join("data","external","input","PhoneDashboard","BuggyFiles","Empty"))
        for zipfile in os.listdir(port):

            if ".zip" not in zipfile:
                continue

            # if "UseIndiv" nearly exactly do process as "Use"
            if keyword == "UseIndiv":
                keyword = "Use"

            # change zipfile name for pd use data
            if ("full" in zipfile) & (keyword == "Use"):
                new_zipfile = zipfile.replace("full","use")
                os.rename(os.path.join(port, zipfile), os.path.join(port, new_zipfile))
                zipfile = new_zipfile

            # change zipfile name for pd custom delay data, as soon as possible
            if ("snooze_delays" in zipfile):
                new_zipfile = zipfile.replace("snooze_","")
                os.rename(os.path.join(port, zipfile), os.path.join(port, new_zipfile))
                zipfile = new_zipfile

            if (keyword.lower() not in zipfile) and (keyword.upper() not in zipfile):
                continue

            #if it already exists, skip
            if zipfile in added_files:
                continue

            #if in the empty or corrupt directory in PA dropbox, also place it in empty or corrupt dir in port
            if zipfile in empty_files_dir:
                try:
                    old_file = os.path.join(port, zipfile)
                    new_file = os.path.join(port, "Empty", zipfile)
                    os.rename(old_file, new_file)
                except:
                    print(f"{zipfile}couldn't move zipfile to PDPort/Empty")
                continue


            #if out of date range, skip
            match = re.search(r'\d{4}-\d{2}-\d{2}', zipfile)
            zip_date = datetime.strptime(match.group(), '%Y-%m-%d')
            if zip_date <= study_config.first_pull or zip_date >= study_config.last_pull:
                continue

            #else, copy and transfer it
            else:
                old_file_path = os.path.join(port,zipfile)
                new_file_path = os.path.join(new_directory,zipfile)
                new_adds.append(zipfile)
                shutil.copy(old_file_path,new_file_path)
        print(new_adds)
        return new_adds

    """ updates the existing config by adding the new config entries, and saves the updated config"""
    def update_config(self,existing,new,config_path):
        existing.update(new)
        pd_config_df = pd.DataFrame.from_dict(existing, orient='index').reset_index()
        pd_config_df.to_csv(config_path, index=False)


    """Default raw data processor invoked by event_puller.py"""
    @staticmethod
    def default_puller_process(df: pd.DataFrame, zip_file: str, event_puller):
        for time_col in event_puller.time_cols:
            df = data_utils.clean_iso_dates(df, time_col, keep_nan=False, orig_tz=event_puller.raw_timezone)
            df = df.drop(columns=[time_col + "Date", time_col + "DatetimeHour", time_col + "EasternDatetimeHour"])
            df = df.rename(columns={time_col + "Datetime": time_col})

        if "TimeZone" in df.columns:
            df = df.drop(columns=["TimeZone"])

        match = re.search(r'\d{4}-\d{2}-\d{2}', zip_file)
        df["AsOf"] = datetime.strptime(match.group(), '%Y-%m-%d')
        df["AsOf"] = df["AsOf"].apply(lambda x: x.date())
        return df

    # add phase column to each obs based study_config survey start times
    # start_buffer =1 means that days will be counted the day after the survey start
    # end_buffer = -1 means that the days will be counted the day before the survey start
    @staticmethod
    def add_phase_label(raw_df, raw_df_date, start_buffer=1, end_buffer=-1):
        df = raw_df.copy()
        if "Phase" in df.columns.values:
            df = df.drop(columns="Phase")

        for phase, specs in study_config.phases.items():
            # label use with phases if we're a day into a phase
            if datetime.now() > specs["StartSurvey"]["Start"] + timedelta(1):
                start_date = (study_config.phases[phase]["StartSurvey"]["Start"] + timedelta(start_buffer)).date()
                end_date = (study_config.phases[phase]["EndSurvey"]["Start"] + timedelta(end_buffer)).date()
                df.loc[(df[raw_df_date] >= start_date) & (df[raw_df_date] <= end_date), "Phase"] = phase

        df["Phase"] = df["Phase"].astype('category')
        return df

    """
    Purpose: Iterates through a subsets dict and creates new avg daily use columns
    
    One key-value pair of a subset dict:
    
    "PCSC" : {
                "Filters": {"SCBool":[True]},
                "DenomCol": "DaysWithUse"},
    
    """
    @staticmethod
    def get_subsets_avg_use(df_p, subsets: dict):
        subset_dfs = []
        for label, specs in subsets.items():
            filters = specs["Filters"]
            denom_col = specs["DenomCol"]
            num_cols = specs["NumCols"]
            subset_df = BuilderUtils.subset_avg_use(df_p, label, filters, denom_col,num_cols)
            subset_dfs.append(subset_df)
        df_merged = reduce(lambda x, y: pd.merge(x, y, on='AppCode', how = 'outer'), subset_dfs)

        # If they are in this df, then they recorded some use in the phase, so we convert all of their nan's
        # (i.e. for a specfic subset) in the df to 0
        df_merged = df_merged.fillna(0)
        return df_merged

    """
    Input:
     - df: the event level df in the given phase
     - label: the variable label
     - specs: {variables to subset on: values of variables to keep}
     - denom_col: the column name of the variable in the df which contains the denomenator value
        - if == "NAN", the function will create it's own denomenator equal to days for which there is non-zero use for
        the given subset
     - num_cols: list of columns to sum over (often it's just [Use], but it can be [Checks,Pickups,Use]
    """
    @staticmethod
    def subset_avg_use(df: pd.DataFrame, label: str, filters: dict, denom_col: str, num_cols: list):
        # if we don't want to subset the phase data at all
        if len(filters) == 0:
            pass

        # go through each filter (note that at all filters for each variable must be met)
        else:
            for var, keep_vals in filters.items():
                df = df.loc[df[var].isin(keep_vals),:]

        for col in [denom_col]+[num_cols]:
            df[col] = df[col].fillna(0)

        sum_df = df.groupby(by=['AppCode',denom_col], as_index=False)[num_cols].sum()

        sum_dfs = []
        for num_col in num_cols:
            sum_df = sum_df.rename(columns={num_col: f"{label}{num_col}Total"})
            sum_df[f"{label}{num_col}Total"] = sum_df[f"{label}{num_col}Total"].round(0)
            sum_df[f"{label}{num_col}"] = (sum_df[f"{label}{num_col}Total"] / (sum_df[denom_col])).round(0)
            sum_dfs.append(sum_df[["AppCode", f"{label}{num_col}", f"{label}{num_col}Total"]])
        final = reduce(lambda df1, df2: pd.merge(df1, df2, on='AppCode', how = 'outer'), sum_dfs)
        return final

    # add phase column to each obs based on time they completed the survey, indicating what phase they are in at the timestamp
    # start_buffer =1 means that days will be counted the day after the survey start
    # end_buffer = -1 means that the days will be counted the day before the survey start
    @staticmethod
    def add_personal_phase_label(raw_df, raw_master, raw_df_date, start_buffer=1, end_buffer=-1, drop_bool=True):
        df = raw_df.copy()
        if "Phase" in df.columns.values:
            df = df.drop(columns="Phase")

        for phase, specs in study_config.phases.items():

            # label use with phases if we're a day into a phase
            if datetime.now() > specs["StartSurvey"]["Start"] + timedelta(1):

                raw_master = data_utils.inpute_missing_survey_datetimes(raw_master, phase)
                old_code = study_config.phases[phase]["StartSurvey"]["Code"]
                new_code = study_config.phases[phase]["EndSurvey"]["Code"]
                start_col = f"{old_code}_SurveyEndDatetime"
                end_col = f"{new_code}_SurveyStartDatetime"

                df = df.merge(raw_master[["AppCode", start_col, end_col]], on="AppCode", how="inner")
                for col in [start_col, end_col]:
                    df[col] = pd.to_datetime(df[col], infer_datetime_format=True).apply(lambda x: x.date())

                df.loc[(df[raw_df_date] >= df[start_col].apply(lambda x: x + timedelta(start_buffer)))
                       & (df[raw_df_date] <= df[end_col].apply(lambda x: x + timedelta(end_buffer))), "Phase"] = phase

                if drop_bool:
                    df = df.drop(columns=[start_col, end_col])
        df["Phase"] = df["Phase"].astype('category')
        return df