WeatherTables

import numpy as np
import pandas as pd

# load the data
bike = pd.read_csv("bike.csv")
df = pd.read_csv("df_weather.csv")
df = df.iloc[:, 1:]
df.rename(columns={'cty2': 'city'}, inplace=True)
bikeW = pd.merge(bike, df, how = 'left', on = 'city')

# divide a brand's distribution between nbS different intervals for the weather feature var
def sub_region(array, row = '0', var = 'avg_days_precip', nbS = 4, split = "default"):
    bikeV = bikeW[bikeW[var] > -1]
    minA = min(bikeV[var])
    maxA = max(bikeV[var])
    
    if split == "default":
        splitL = [minA + k*float(maxA - minA)/nbS for k in range(nbS)]
        splitU = [minA + k*float(maxA - minA)/nbS for k in range(1, nbS+1)]
    else:
        splitL = [minA] + [splt for splt in split]
        splitU = [splt for splt in split] + [maxA]
        nbS = len(split)+1
    
    count = np.zeros(nbS)
    
    for elt in array:
        for lo, up in zip(splitL, splitU):
            if elt >= lo and elt < up:
                pos = [k for k, x in zip(range(nbS), splitL) if x == lo][0]
                count[pos] += 1
    colInt = []
    prop = []
    for i in range(len(count)):
        prop.append(count[i]/len(array))
        colInt.append(var + ": " + str(splitL[i]) + "-" + str(splitU[i]))
    
    sub_dist = pd.DataFrame(columns = colInt, index = [row])
    sub_dist.loc[row] = prop
                
    return sub_dist

# compute the proportion among a brand for a given weather feature: P(feature split|brand)
def prop_var(var, nbS = 4, size = 10, split="defaut"):
    bikeV = bikeW[bikeW[var] > -1]

    bikeVinfo = bikeV.groupby('make')[var].agg(['size' ,'min', 'mean', 'max'])\
                                          .sort(['size'], ascending=[0])
    makesize = []
    for mk in bikeV.make:
        makesize.append(bikeVinfo.ix[mk]['size'])
    bikeV['makesize'] = makesize
    bikeV.head()

    prop = pd.DataFrame({})
    mksize = []
    for mk in np.unique(bikeV.make):
        mksize.append(np.unique(bikeV[bikeV.make == mk]['makesize'])[0])
        temp = sub_region(bikeV[bikeV.make == mk][var], mk, var, nbS, split)
        prop = pd.concat([prop, temp], axis = 0)

    prop = pd.concat([prop, pd.DataFrame({'makesize': mksize}, index = prop.index)], axis=1)
    prop = prop[prop.makesize > size]
    
    Hi, Hi_R, Hi_Int = [], [], []
    Lo, Lo_R, Lo_Int = [], [], []
    for mk in prop.index:
        m, M = min(prop.ix[mk][:-1]), max(prop.ix[mk][:-1])
        indm = [i for i in range(len(prop.columns)) if prop.ix[mk][i] == m][0]
        indM = [i for i in range(len(prop.columns)) if prop.ix[mk][i] == M]
        if len(indM) == 1:
            indM = indM[0]
        else:
            indM = indM[1]
        Intm, IntM = prop.columns[indm].split(":")[1], prop.columns[indM].split(":")[1]
        Lo.append(m)
        Lo_R.append(indm)
        Lo_Int.append(Intm)
        Hi.append(M)
        Hi_R.append(indM)
        Hi_Int.append(IntM)
    prop = pd.concat([prop, pd.DataFrame({'Highest': Hi, 'Lowest': Lo, \
                                           'High_Region': Hi_R, 'Low_Region': Lo_R, \
                                           'High_Int': Hi_Int, 'Low_Int': Lo_Int}, \
                                index = prop.index)], axis=1)

    return prop

# compute the probability of getting a brand given that the observation lies in a
# particular split of the previous weather feature: P(brand|feature split)
def proba_var(var, nbS = 4, size = 10, split="default"):
    bikeV = bikeW[bikeW[var] > -1]
    prop = prop_var(var, nbS, size, split)
    mkind = prop.index
    proba = pd.DataFrame({})

    for mk in mkind:
        probaCol = ['proba_tot']
        probaVal = []
        probaVal.append(prop.ix[mk]['makesize']/len(bikeV))
        for i in range(nbS):
            coli = prop.columns[i]
            sizei = round(np.dot(prop[coli], prop['makesize']))#+1
            mk_nbi = round(np.dot(prop.ix[mk][coli], prop.ix[mk]['makesize']))
            probaCol.append('proba_' + str(i+1) + ": " + coli)
            probaVal.append(mk_nbi/sizei)
        temp = pd.DataFrame(columns = probaCol, index = [mk])
        temp.loc[mk] = probaVal
        proba = pd.concat([proba, temp], axis=0)
    return proba