github.py

#!/usr/bin/env python3

import pandas as pd
import numpy as np
import scipy
import scipy.stats
import csv

epsilon = 0.1
maxiter = 1000
S_lim = 3

inputdata = {}

def rows2pd(key, rows):
    if rows[0][0] != '':
        raise Exception('data error');
    colnames = rows[0][1:]
    rownames = list(map(lambda x: x[0], rows[1:]))
    inputdata[key] = pd.DataFrame(map(lambda x: x[1:], rows[1:]), dtype=float);
    inputdata[key].index = rownames;
    inputdata[key].columns = colnames;
    return

with open('in.csv', 'r', encoding='utf-8') as csvfile:
    reader = csv.reader(csvfile)
    inkey = ''
    rows = []
    for row in reader:
        if len(row):
            strrowlen = len(''.join(row))
            if strrowlen:
                if len(row[0]) == strrowlen:
                    if inkey and rows:
                        rows2pd(inkey, rows)
                    inkey = row[0]
                    rows = []
                    print('reading', row[0])
                else:
                    rows.append(row)
    if inkey and rows:
        rows2pd(inkey, rows)

print('input:')
for key, data in inputdata.items():
    print(key)
    print(data)
    print()


print()
print('working...')
print()



df_group_by_sphere_pzk = {}
for tenskey, tens in inputdata.items():
    df_group_by_sphere_pzk[tenskey] = {}
    for row in tens.index:
        for col in tens.columns:
            df_group_by_sphere_pzk[tenskey][row + '_' + col] = tens.loc[row,col]
df_group_by_sphere_pzk = pd.DataFrame(df_group_by_sphere_pzk).T



df_sphere_by_group_pzk = {}
df_pzk_by_group_sphere = {}
for col in df_group_by_sphere_pzk.columns:
    scol = col.split('_')[0]
    pcol = col.split('_')[1]
    if scol not in df_sphere_by_group_pzk:
        df_sphere_by_group_pzk[scol] = {}
    if pcol not in df_pzk_by_group_sphere:
        df_pzk_by_group_sphere[pcol] = {}
    for row in df_group_by_sphere_pzk.index:
        df_sphere_by_group_pzk[scol][row + '_' + pcol] = df_group_by_sphere_pzk.loc[row,col]
        df_pzk_by_group_sphere[pcol][row + '_' + scol] = df_group_by_sphere_pzk.loc[row,col]
df_sphere_by_group_pzk = pd.DataFrame(df_sphere_by_group_pzk).T
df_pzk_by_group_sphere = pd.DataFrame(df_pzk_by_group_sphere).T



Y1_K = []
n1 = len(df_group_by_sphere_pzk.columns)
n2 = len(df_group_by_sphere_pzk.index)
for k in range(n2):
    for row in range(n1):
        msrow = [0] * n2 * n1
        msrow[row*n2 + k] = 1
        Y1_K.append(msrow)
Y1_K = np.matrix(Y1_K)
Y1_K2 = Y1_K @ Y1_K
Y1_X = np.matrix(df_group_by_sphere_pzk)
Y1_X = Y1_X.reshape((Y1_X.size, 1))
Y1_COL = Y1_K2 @ Y1_X
Y1 = Y1_COL.reshape(len(df_group_by_sphere_pzk.index), len(df_group_by_sphere_pzk.columns))



Y2_K = []
n1 = len(df_group_by_sphere_pzk.index)
n3 = len(df_pzk_by_group_sphere.index) #100
n2 = len(df_sphere_by_group_pzk.index) #5
for k in range(n2):
    for row in range(n3):
        msrow = [0] * n2 * n3
        msrow[row*n2 + k] = 1
        Y2_K.append(msrow)
Y2_K = np.matrix(Y2_K)
Y2_I = np.identity(n1)
Y2_K2 = np.kron(Y2_I, Y2_K @ Y2_K)
Y2_X = np.matrix(df_sphere_by_group_pzk)
Y2_X = Y2_X.reshape((Y2_X.size, 1))
Y2_COL = Y2_K2 @ Y2_X
Y2 = Y2_COL.reshape(len(df_sphere_by_group_pzk.index), len(df_sphere_by_group_pzk.columns))
Y2.shape


Y3 = np.matrix(df_pzk_by_group_sphere)
Y3.shape


n1 = len(df_group_by_sphere_pzk.index)
n2 = len(df_sphere_by_group_pzk.index)
n3 = len(df_pzk_by_group_sphere.index)
A1 = np.identity(n1)
A2 = np.identity(n2)
A3 = np.identity(n3)
P_A1 = A1
P_A2 = A2
P_A3 = A3

curiter = maxiter

while curiter>0:
    curiter -= 1
    Q1_A1 = np.kron(A2,A3)
    Q1 = Y1 @ Q1_A1
    U, S1, VH = np.linalg.svd(Q1)
    A1 = U[:, range(S_lim)]
    
    Q2_A2 = np.kron(A1, A3)
    Q2 = Y2 @ Q2_A2
    U, S2, VH = np.linalg.svd(Q2)
    A2 = U[:, range(S_lim)]
    
    Q3_A3 = np.kron(A1, A2)
    Q3 = Y3 @ Q3_A3
    U, S3, VH = np.linalg.svd(Q3)
    A3 = U[:, range(S_lim)]
    
    gobreak = [False, False, False]
    diff = []
    A_sqr_sum = []
    for i, TMP_A, TMP_PA in ( (0, A1, P_A1), (1, A2, P_A2), (2, A3, P_A3) ):
        colmax = max(TMP_A.shape[1],TMP_PA.shape[1])
        if TMP_A.shape[1] < colmax:
            TMP_A = np.concatenate( (TMP_A, np.zeros( ( TMP_A.shape[0], colmax - TMP_A.shape[1] ) ) ), axis = 1 )
        if TMP_PA.shape[1] < colmax:
            TMP_PA = np.concatenate( (TMP_PA, np.zeros( ( TMP_PA.shape[0], colmax - TMP_PA.shape[1] ) ) ), axis = 1 )
        gobreak[i] = np.sum(np.square(TMP_A-TMP_PA))/TMP_A.size < epsilon
        diff.append(np.sum(np.square(TMP_A-TMP_PA))/TMP_A.size)
        A_sqr_sum.append(np.sum(np.square(TMP_A)))
    if all(gobreak):
        break
    P_A1 = A1
    P_A2 = A2
    P_A3 = A3



df_A_dict = {}
for Aname, A, new_index in (('A1', A1, df_group_by_sphere_pzk.index), ('A2', A2, df_sphere_by_group_pzk.index), ('A3', A3, df_pzk_by_group_sphere.index) ):
    print()
    #Манхеттен-норма'
    manhetten = np.sum(np.absolute(A), axis=1)
    mrank = A.shape[0] - scipy.stats.rankdata(manhetten, axis=0) + 1
    #Евклидова-норма
    euclid = np.sqrt(np.sum(np.square(A), axis=1))
    erank = A.shape[0] - scipy.stats.rankdata(euclid, axis=0) + 1
    #Бесконечность-норма
    linf = np.max(np.absolute(A), axis=1)
    lrank = A.shape[0] - scipy.stats.rankdata(linf, axis=0) + 1
    # Норма Гёльдера
    gelder = np.power(np.sum(np.power(A,4), axis=1),0.25)
    grank = A.shape[0] - scipy.stats.rankdata(gelder, axis=0) + 1
    sumrank = mrank + erank + lrank + grank
    finrank = scipy.stats.rankdata(sumrank, axis=0)
    dfA = pd.DataFrame(np.concatenate( (A, manhetten, mrank, euclid, erank, linf, lrank, gelder, grank, sumrank, finrank), axis=1 ))
    dfA.index = new_index
    dfA.columns = list(dfA.columns[:A.shape[1]]) + ['Манхеттен', 'Мранг', 'Евклид', 'Еранг', 'LБесконечность', 'Lранг', 'Гёльдер', 'Гранг', 'Сум ранг', 'Итог ранг']
    df_A_dict[Aname] = dfA


print('ouput:')
for key, data in df_A_dict.items():
    print(key)
    print(data)
    print()


C = np.kron(np.kron(A1,A2),A3)


Z = C.T @ Y3.reshape((Y3.size, 1))
df_Z_dict = {}

for i in range(1,S_lim + 1):
    df_Z_dict['Z' + str(i)] = pd.DataFrame(Z[range(S_lim*S_lim*(i-1),S_lim*S_lim*i)].reshape((S_lim, S_lim)))

print()
print('tensor')
for key, data in df_Z_dict.items():
    print(key)
    print(data)
    print()

with open('out.csv', 'w', newline='', encoding='utf-8') as csvfile:
    writer = csv.writer(csvfile)
    
    writer.writerow(['ouput'])
    for key, data in df_A_dict.items():
        rows = data.values.tolist();
        for i in range(len(rows)):
            rows[i].insert(0, data.index[i])
        rows.insert(0, [''] + list(data.columns));
        writer.writerow([key])
        writer.writerows(rows)
        writer.writerow([])
        
    writer.writerow(['tensor'])
    for key, data in df_Z_dict.items():
        rows = data.values.tolist();
        for i in range(len(rows)):
            rows[i].insert(0, data.index[i])
        rows.insert(0, [''] + list(data.columns));
        writer.writerow([key])
        writer.writerows(rows)
        writer.writerow([])



#