17/replication_package/code/lib/utilities/figures.py

import matplotlib.mlab as mlab
import matplotlib.pyplot as plt
from datetime import datetime
from functools import reduce
import pandas as pd
import numpy as np
import os
import math
import sys
import pickle
from datetime import datetime
from IPython.display import display

#crosstabs a dictionary of dfs, by var1, var2, and keeps the bottom and top 'rows' rows
def crosstab_dic(dic, var1, var2, rows):
    for name, df in dic.items():
        df[var1] = df[var1].fillna(-1)
        # df[var2]= df[var2].fillna("Missing")

        ct = pd.crosstab(df[var1], df[var2], dropna=False, margins=True)
        # ct["Overall"] = ct.sum(axis = 1)
        # ct.loc['Overall']= ct.sum()
        if len(ct) > rows * 2 + 2:
            print(f"{name}: first {rows} and last {rows} rows")
            display(ct.iloc[np.r_[0:rows, -rows:0]])
        else:
            print(f"{name}: all rows")
            display(ct)

# Plots a cdf for a list of val_cols, that are identified by an obs_col
def cdf(raw_df, obs_col, val_cols,x_label,x_upper,x_lower = 0):
    for val_col in val_cols:
        df = raw_df[[obs_col, val_col]]
        df = df.loc[df[val_col].notnull()]
        assert len(raw_df[obs_col]) == len(raw_df[obs_col].unique())
        df = df.sort_values(val_col).reset_index(drop=True)
        df = df.reset_index()
        df["Perc"] = df["index"] / len(df)
        plt.plot(df[val_col], df["Perc"])
        tot = len(df)
        sub = len(df.loc[(df[val_col]<x_upper)&(df[val_col]>x_lower)])
        print(f"frac of {val_col}: {sub/tot}")

    plt.xlabel(x_label)
    plt.xlim(x_lower, x_upper)
    plt.legend(val_cols)
    plt.show()

# Graphs a Phone Dashboard Variable Over 'CreatedDate'
def varXtime(raw_df,var, max_yval ,time_var , remove_max = False, label_freq = 3):
    df = raw_df.groupby(by=[time_var,'AppCode'],as_index = False)[var].sum()
    #display(df.head(40))
    df = df.groupby([time_var]).describe().reset_index()
    df.columns = [''.join(col).strip().replace(var,"") for col in df.columns.values]
    days = list(df[time_var].unique())
    print(f"Graph over {len(days)} days")

    count_df = df[[time_var,"count"]]
    dist_df = df.drop(columns = ["count",'std'])
    if remove_max == True:
        dist_df = df.drop(columns=["max"])

    #display(dist_df.head(40))
    for metric,df in {var:dist_df,"Count":count_df}.items():

        plt.xlabel(time_var,)
        plt.ylabel(metric)

        for col in df.columns:
            if col == time_var:
                continue
            plt.plot(df[time_var],df[col])
        labels_dt = list(df[time_var].unique())[::label_freq]
        labels = [x.strftime("%Y-%m-%d") for x in labels_dt]
        #print(labels)
        plt.xticks(labels_dt,labels, rotation=90 )

        if metric == var:
            plt.ylim(df['min'].min(), max_yval)

        plt.legend()
        plt.show()

## Assumes df is on the date level
def varsXtime(df:pd.DataFrame, vars: list, date_var: str):
    for var in vars:
        plt.plot(df[date_var], df[var])
    labels_dt = list(df[date_var].unique())[::2]
    labels = [x.strftime("%Y-%m-%d") for x in labels_dt]
    plt.xticks(labels_dt, labels, rotation=90)
    plt.legend()
    plt.show()