Machine Learning to predict location of ice recrystallization

May - July 2022
UGA and IGE internship
M1 Statistics and Data Sciences (SSD)

Renan MANCEAUX
Supervisor : Thomas CHAUVE

Data Computing

8.2. Compute Triple Junction dataset#

import numpy as np
import pandas as pd
import sys
sys.path.append("../../scripts/")
import utils

from tqdm.notebook import tqdm
import pickle

import xarrayaita.aita as xa
from collections import Counter
from scipy.spatial import distance_matrix

name = 'CI09'

8.2.1. Loading data#

data = utils.load_data("../../data/for_learning_plus/"+name+".npy")
file = open('../../data/craft/'+name+'.xr', 'rb')
ds_data = pickle.load(file)
file.close()

8.2.2. Compute variables#

8.2.2.1. Compute map of TJ and distance#

maps = ds_data.aita.dist2eachTJ()
data['idTJ'] = np.argmin(np.array(maps), axis=2).flatten()
data['grainId'] = np.array(ds_data.grainId).flatten()
imshape = np.shape(ds_data.grainId)
ntj = len(data.idTJ.unique())
M = ds_data.aita.TJ_map()

8.2.2.2. Filter TJ to close to border#

tj_to_skip = []
for idtj in tqdm(np.unique(data.idTJ)):
    coord =  M[idtj][0:2]
    if (coord[0]-4.5<0)|(coord[1]-4.5<0)|(coord[0]+4.5>imshape[1]-1)|(coord[1]+4.5>imshape[0]-1):
        tj_to_skip.append(idtj)

8.2.2.3. Compute RX of each TJ (10%)#

RX = np.zeros((ntj))
for i in tqdm(data.idTJ.unique()):
    if i in tj_to_skip :
        RX[i] = np.nan
    else:
        m = np.array(maps[:,:,i]).flatten()
        n = len(np.where(m < 10)[0])
        if sum(data.loc[(np.where(m < 10)[0])].Y)>(n*10/100): #10%
            RX[i] = 1

8.2.2.4. Compute schmid factor of each 3 grains#

schmid_tj = np.zeros((ntj,3))
for i in tqdm(range(2,5)) :
    for j in range(ntj) :
        if j in tj_to_skip :
            schmid_tj[j][i-2] = np.nan
        else:
            id = int(M[j][i])
            schmid_tj[j][i-2] = data.loc[(data.grainId == id),'schmid'].unique()   

8.2.2.5. Compute schmid diff for each 3 GB#

diff_schmid_tj = np.zeros((ntj,3))
for i in tqdm(range(ntj)) :
    if i in tj_to_skip :
        diff_schmid_tj[i] = np.nan
    else:
        coord = M[i][0:2]

        pts_y = np.array([coord[0]-1.5,coord[0]-0.5,coord[0]+0.5,coord[0]+1.5]).astype(int)
        pts_x = np.array([coord[1]-1.5,coord[1]-0.5,coord[1]+0.5,coord[1]+1.5]).astype(int)

        val = []
        for x in pts_x:
            for y in pts_y:
                val.append(np.array((data.diff_schmid)).reshape(imshape)[x,y])

        if len(np.unique(val))<3:

            pts_y = np.append(pts_y,coord[0]-2.5).astype(int)
            pts_y = np.append(pts_y,coord[0]+2.5).astype(int)
            pts_x = np.append(pts_x,coord[1]-2.5).astype(int)
            pts_x = np.append(pts_x,coord[1]+2.5).astype(int)

            val = []
            for x in pts_x:
                for y in pts_y:
                    val.append(np.array((data.diff_schmid)).reshape(imshape)[x,y])

        if len(np.unique(val))<3:

            pts_y = np.append(pts_y,coord[0]-3.5).astype(int)
            pts_y = np.append(pts_y,coord[0]+3.5).astype(int)
            pts_x = np.append(pts_x,coord[1]-3.5).astype(int)
            pts_x = np.append(pts_x,coord[1]+3.5).astype(int)

            val = []
            for x in pts_x:
                for y in pts_y:
                    val.append(np.array((data.diff_schmid)).reshape(imshape)[x,y])

        val_sort = sorted(val, key=Counter(val).get, reverse=True)
        good_val= list(dict.fromkeys(val_sort))[0:3]

        diff_schmid_tj[i] = good_val

8.2.2.6. Compute misorientation angle of each 3 GB#

misangle_tj = np.zeros((ntj,3))
for i in tqdm(range(ntj)) :
    if i in tj_to_skip :
        misangle_tj[i] = np.nan
    else:
        coord = M[i][0:2]

        pts_y = np.array([coord[0]-1.5,coord[0]-0.5,coord[0]+0.5,coord[0]+1.5]).astype(int)
        pts_x = np.array([coord[1]-1.5,coord[1]-0.5,coord[1]+0.5,coord[1]+1.5]).astype(int)

        val = []
        for x in pts_x:
            for y in pts_y:
                val.append(np.array((data.misangle)).reshape(imshape)[x,y])

        if len(np.unique(val))<3:

            pts_y = np.append(pts_y,coord[0]-2.5).astype(int)
            pts_y = np.append(pts_y,coord[0]+2.5).astype(int)
            pts_x = np.append(pts_x,coord[1]-2.5).astype(int)
            pts_x = np.append(pts_x,coord[1]+2.5).astype(int)

            val = []
            for x in pts_x:
                for y in pts_y:
                    val.append(np.array((data.misangle)).reshape(imshape)[x,y])
        
        if len(np.unique(val))<3:

            pts_y = np.append(pts_y,coord[0]-3.5).astype(int)
            pts_y = np.append(pts_y,coord[0]+3.5).astype(int)
            pts_x = np.append(pts_x,coord[1]-3.5).astype(int)
            pts_x = np.append(pts_x,coord[1]+3.5).astype(int)

            val = []
            for x in pts_x:
                for y in pts_y:
                    val.append(np.array((data.diff_schmid)).reshape(imshape)[x,y])

        val_sort = sorted(val, key=Counter(val).get, reverse=True)
        good_val= list(dict.fromkeys(val_sort))[0:3]

        misangle_tj[i] = good_val

8.2.2.7. Compute volume ratio anisotropy of each TJ#

volratio_an_tj = np.zeros((ntj))

for i in tqdm(range(ntj)) :
    if i in tj_to_skip :
        volratio_an_tj[i] = np.nan
    else:
        coord = M[i][0:2]
        volratio_an_tj[i] = np.array((data.volratio_an)).reshape(imshape)[(np.array(coord)[1]-0.5).astype(int),(np.array(coord)[0]-0.5).astype(int)]

8.2.2.8. Compute mean of craft variables for each TJ#

work_mean_tj = np.zeros((ntj))
eqStrain_mean_tj = np.zeros((ntj))
eqStress_mean_tj = np.zeros((ntj))
act_py_mean_tj = np.zeros((ntj))
act_pr_mean_tj = np.zeros((ntj))

for i in tqdm(range(ntj)) :
    if i in tj_to_skip :
        work_mean_tj[i] = np.nan
        eqStrain_mean_tj[i] = np.nan
        eqStress_mean_tj[i] = np.nan
        act_py_mean_tj[i] = np.nan
        act_pr_mean_tj[i] = np.nan
    else:
        coord = M[i][0:2]

        pts_y = np.array([coord[0]-1.5,coord[0]-0.5,coord[0]+0.5,coord[0]+1.5]).astype(int)
        pts_x = np.array([coord[1]-1.5,coord[1]-0.5,coord[1]+0.5,coord[1]+1.5]).astype(int)

        val_w = []
        val_strain = []
        val_stress = []
        val_py = []
        val_pr = []
        for x in pts_x:
            for y in pts_y:
                val_w.append(np.array((data.work)).reshape(imshape)[x,y])
                val_strain.append(np.array((data.eqStrain)).reshape(imshape)[x,y])
                val_stress.append(np.array((data.eqStress)).reshape(imshape)[x,y])
                val_py.append(np.array((data.act_py)).reshape(imshape)[x,y])
                val_pr.append(np.array((data.act_pr)).reshape(imshape)[x,y])

        work_mean_tj[i] = np.mean(val_w)
        eqStrain_mean_tj[i] = np.mean(val_strain)
        eqStress_mean_tj[i] = np.mean(val_stress)
        act_py_mean_tj[i] = np.mean(val_py)
        act_pr_mean_tj[i] = np.mean(val_pr)

8.2.2.9. Compute Distance to other TJ#

Ma = ds_data.aita.TJ_map().T[0:2].T
dist = pd.DataFrame(distance_matrix(Ma,Ma))

dist2oTJ = np.zeros((ntj))

for i in tqdm(range(ntj)) :
    if i in tj_to_skip :
        dist2oTJ[i] = np.nan
    else:
        dist2oTJ[i] = np.array(dist[i].sort_values()[1:2])

8.2.2.10. Compute number of pixel of the 3 grains of each TJ#

nb_pix_g = []

for id in tqdm(np.unique(ds_data.grainId)) :
    nb_pix_g.append(int(sum(sum(ds_data.grainId == id))))

nb_pix_g = pd.Series(nb_pix_g,index=np.unique(ds_data.grainId).astype(int))

nb_pix_g_tj = np.zeros((ntj,3))

for i in tqdm(range(ntj)) :
    if i in tj_to_skip :
        nb_pix_g_tj[i] = np.nan
    else:
        idTJ = M[i][2:]
        nb_pix_g_tj[i] = np.array(nb_pix_g[np.array(idTJ)])

8.2.3. Building dataset#

TJ_data = pd.DataFrame(
    (
        RX,
        schmid_tj.T[0],
        schmid_tj.T[1],
        schmid_tj.T[2],
        diff_schmid_tj.T[0],
        diff_schmid_tj.T[1],
        diff_schmid_tj.T[2],
        misangle_tj.T[0],
        misangle_tj.T[1],
        misangle_tj.T[2],
        volratio_an_tj,
        eqStrain_mean_tj,
        eqStress_mean_tj,
        act_pr_mean_tj,
        act_py_mean_tj,
        work_mean_tj,
        dist2oTJ,
        nb_pix_g_tj.T[0],
        nb_pix_g_tj.T[1],
        nb_pix_g_tj.T[2]
    )
).T

TJ_data.columns=["RX","schmid1","schmid2","schmid3","diff_schmid1","diff_schmid2","diff_schmid3","misangle1","misangle2","misangle3",
        "volratio_an","eqStrain","eqStress","act_pr","act_py","work","dist1neigh","nb_pix_g1","nb_pix_g2","nb_pix_g3"]

TJ_data

	RX	schmid1	schmid2	schmid3	diff_schmid1	diff_schmid2	diff_schmid3	misangle1	misangle2	misangle3	volratio_an	eqStrain	eqStress	act_pr	act_py	work	dist1neigh	nb_pix_g1	nb_pix_g2	nb_pix_g3
0	1.0	0.433383	0.314354	0.464099	0.030716	0.119029	0.149745	0.726541	1.124480	0.706312	0.976792	0.018254	1.331305	0.000026	4.466442e-07	0.002944	21.377558	5824.0	3173.0	552.0
1	0.0	0.499819	0.433383	0.464099	0.066436	0.035720	0.030716	1.399579	1.130727	0.340875	0.964957	0.017129	1.155659	0.000018	1.703326e-08	0.001886	21.377558	5931.0	5824.0	552.0
2	1.0	0.056638	0.220494	0.244690	0.024196	0.188052	0.163855	1.397844	1.482392	0.242694	0.950161	0.011322	1.132815	0.000006	4.944418e-08	0.001001	141.014184	36321.0	1386.0	3667.0
3	1.0	0.491447	0.056638	0.314354	0.434809	0.177093	0.257716	0.636235	0.283226	0.357118	0.999528	0.020613	1.025976	0.000004	1.232579e-08	0.001333	41.231056	27290.0	36321.0	3173.0
4	1.0	0.491447	0.056638	0.326571	0.164876	0.434809	0.269933	0.636235	0.983226	0.397189	0.982284	0.004298	1.074021	0.000004	8.676976e-10	0.000441	61.400326	27290.0	36321.0	5481.0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
57	0.0	0.066017	0.492388	0.437152	0.055236	0.426370	0.371135	1.131787	0.382646	0.764838	0.992830	0.028784	1.937496	0.000114	6.602019e-06	0.008137	27.202941	3422.0	967.0	3726.0
58	0.0	0.043031	0.195062	0.338734	0.295703	0.143672	0.152031	1.014145	1.250826	0.249974	0.994848	0.043997	2.038062	0.000110	1.538746e-05	0.011282	5.000000	3440.0	1015.0	3007.0
59	1.0	0.195062	0.066017	0.148617	0.046446	0.129045	0.082599	0.359894	0.233268	0.134345	0.999837	0.015061	1.956725	0.000011	7.114007e-06	0.005079	55.659680	1015.0	3422.0	20561.0
60	0.0	0.074013	0.043031	0.059130	0.030982	0.014883	0.016099	0.284868	0.250435	0.050341	0.999907	0.016632	1.084423	0.000032	3.118853e-07	0.002953	72.560320	2194.0	3440.0	16226.0
61	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN

62 rows × 20 columns

8.2.3.1. Save New Dataset#

#np.save(file="../../data/TJ/TJ_"+name,arr=TJ_data)

Machine Learning to predict RX localization

Compute Triple Junction dataset

Contents

Machine Learning to predict location of ice recrystallization

Data Computing

8.2. Compute Triple Junction dataset#

8.2.1. Loading data#

8.2.2. Compute variables#

8.2.2.1. Compute map of TJ and distance#

8.2.2.2. Filter TJ to close to border#

8.2.2.3. Compute RX of each TJ (10%)#

8.2.2.4. Compute schmid factor of each 3 grains#

8.2.2.5. Compute schmid diff for each 3 GB#

8.2.2.6. Compute misorientation angle of each 3 GB#

8.2.2.7. Compute volume ratio anisotropy of each TJ#

8.2.2.8. Compute mean of craft variables for each TJ#

8.2.2.9. Compute Distance to other TJ#

8.2.2.10. Compute number of pixel of the 3 grains of each TJ#

8.2.3. Building dataset#

8.2.3.1. Save New Dataset#