def predict_e2e(pre_e2e, project):
prediction_li = []
nrcomp = len(models.fdm.comp_dict)
materials = np.unique(pre_e2e.material).tolist()
nrmat = len(materials)
mat_dict = dict([(mat, {'ix': materials.index(mat)}) for mat in materials])
Xnw = np.zeros(shape=(nrmat, nrcomp), dtype=float)
# matcmp_df = DataFrame(columns=models.fdm.components)
# components_set = set(models.fdm.components)
# matcmp_df.index.name = 'material'
# matcmp_df.columns.name = 'component'
for idx, row_srs in pre_e2e.iterrows():
if row_srs.component in models.fdm.comp_dict.keys():
matix = mat_dict[row_srs.material]['ix']
cmpix = models.fdm.comp_dict[row_srs.component]['ix']
Xnw[matix][cmpix] = row_srs.bompartp
# matcmp_df.set_value(row_srs.A0MATERIAL, row_srs.A0COMPONENT, row_srs.ZCPE2E002_EQ002_BOMPARTP)
# Xnw = np.array(matcmp_df)
print('lengths Xnw: ', len(Xnw))
print('Components/Features Xnw[0].size: ', Xnw[0].size)
for learn in models.fdm.learn_li:
Ynw = learn.clf.predict(Xnw)
print('Region {}, Classifier {}: WINs Yts {}, Ynw {}, accuracy {}: '.
format(learn.segment, learn.clf_name, learn.y_ts_wins,
len(Ynw[[Ynw == 'WIN']]), learn.accuracy))
print('Region {0}, Classifier {1}: Accuracy is: {2:5.2f}'.format(
learn.segment, learn.clf_name, learn.accuracy))
for material in mat_dict.keys():
if 'A0MATERIAL_T' in pre_e2e.columns:
descr = pre_e2e[pre_e2e['material'] ==
material]['A0MATERIAL_T'].iloc[0]
else:
descr = material
prediction1 = models.prediction_cl(
material=material,
descr=descr,
segment=learn.segment,
clf_name=learn.clf_name,
project=project,
fnresstssbm=Ynw[mat_dict[material]['ix']],
accuracy=learn.accuracy)
prediction_li.append(prediction1)
return (prediction_li)
def predict_formulas(gen_e2e):
# For generated E2E dataset execute the prediction
nrcomp = len(models.fdm.comp_dict)
materials = np.unique(gen_e2e.material).tolist()
nrmat = len(materials)
mat_dict = dict([(mat, {'ix': materials.index(mat)}) for mat in materials])
Xnw = np.zeros(shape=(nrmat, nrcomp), dtype=float)
# matcmp_df = DataFrame(columns=models.fdm.components)
# matcmp_df.index.name = 'material'
# matcmp_df.columns.name = 'component'
for idx, row_srs in gen_e2e.iterrows():
if row_srs.component in models.fdm.comp_dict.keys():
matix = mat_dict[row_srs.material]['ix']
cmpix = models.fdm.comp_dict[row_srs.component]['ix']
Xnw[matix][cmpix] = row_srs.bompartp
# matcmp_df.set_value(row_srs.material, row_srs.component, row_srs.bompartp)
# matcmp_df.fillna(0, inplace=True)
# Xnw = np.array(matcmp_df)
print('lengths Xnw: ', len(Xnw))
print('Components/Features Xnw[0].size: ', Xnw[0].size)
for learn in models.fdm.learn_li:
Ynw = learn.clf.predict(Xnw)
print('Region {}, Classifier {}: WINs Yts {}, Ynw {}, accuracy {}: '.
format(learn.segment, learn.clf_name, learn.y_ts_wins,
len(Ynw[[Ynw == 'WIN']]), learn.accuracy))
print('Region {0}, Classifier {1}: Accuracy is: {2:5.2f}'.format(
learn.segment, learn.clf_name, learn.accuracy))
for experiment in mat_dict.keys():
gen1 = [
exp for exp in models.fdm.generate_li
if exp.experiment == experiment
][0]
ix = mat_dict[experiment]['ix']
prediction1 = models.prediction_cl(material=experiment,
descr='',
segment=learn.segment,
clf_name=learn.clf_name,
project='FDM',
fnresstssbm=Ynw[ix],
accuracy=learn.accuracy)
gen1.predictions.append(prediction1)
def generate_ml(nrexp, region_choices, category_choices, corp_choices):
# Read E2E data
e2e = models.fdm.e2e.copy(deep=True)
if len(region_choices):
e2e = e2e[e2e['region'].isin(region_choices)]
if len(category_choices):
e2e = e2e[e2e['catclass'].isin(category_choices)]
if len(corp_choices):
e2e = e2e[e2e['custcorp'].isin(corp_choices)]
# Determine nr of compoents and dosage distribution in filtered E2E
matcmp_df = e2e.groupby(
['material', 'component',
'bompartp']).size().reset_index().rename(columns={0: 'count'})
gen_comp_dict = dict(matcmp_df['component'].value_counts())
for component in gen_comp_dict.keys():
dosage = matcmp_df[matcmp_df['component'] == component]['bompartp']
nrcomp = gen_comp_dict[component]
dosage_mu, dosage_std = stats.norm.fit(dosage)
gen_comp_dict[component] = [nrcomp, dosage_mu, dosage_std]
# X is the number of components used by materials, based on normal distribution calc components for generated experiments
nrcomp_X = np.array(matcmp_df['material'].value_counts())
nrcomp_mu, nrcomp_std = stats.norm.fit(nrcomp_X)
nrcomp_Y = stats.truncnorm.rvs(1, (1000 - nrcomp_mu) / nrcomp_std,
nrcomp_mu,
nrcomp_std,
size=nrexp + 1).astype(int)
# Generate new E2E dataset for new experimentes
gen_e2e = DataFrame(columns=['material', 'component', 'bompartp'])
models.fdm.generate_li = None # cleanup old data
models.fdm.generate_li = []
segment = 'regions:'
for region in region_choices:
segment = segment + ' ' + region
for expnr in range(1, nrexp + 1):
experiment = 'FDM%0.9d' % expnr
nrcomp = nrcomp_Y[expnr]
bom = draw_sample(experiment, nrcomp, gen_comp_dict)
e2eexp = pd.DataFrame(bom,
columns=['component', 'comp_name', 'bompartp'])
e2eexp['material'] = experiment
gen_e2e = gen_e2e.append(e2eexp, ignore_index=True)
gen1 = models.generate_cl(experiment, segment, nrcomp)
gen1.bom = bom
models.fdm.generate_li.append(gen1)
# For generated E2E dataset execute the prediction
nrcomp = len(models.fdm.comp_dict)
materials = np.unique(gen_e2e.material).tolist()
nrmat = len(materials)
mat_dict = dict([(mat, {'ix': materials.index(mat)}) for mat in materials])
Xnw = np.zeros(shape=(nrmat, nrcomp), dtype=float)
# matcmp_df = DataFrame(columns=models.fdm.components)
# matcmp_df.index.name = 'material'
# matcmp_df.columns.name = 'component'
for idx, row_srs in gen_e2e.iterrows():
if row_srs.component in models.fdm.comp_dict.keys():
matix = mat_dict[row_srs.material]['ix']
cmpix = models.fdm.comp_dict[row_srs.component]['ix']
Xnw[matix][cmpix] = row_srs.bompartp
# matcmp_df.set_value(row_srs.material, row_srs.component, row_srs.bompartp)
# matcmp_df.fillna(0, inplace=True)
# Xnw = np.array(matcmp_df)
print('lengths Xnw: ', len(Xnw))
print('Components/Features Xnw[0].size: ', Xnw[0].size)
for learn in models.fdm.learn_li:
Ynw = learn.clf.predict(Xnw)
print('Region {}, Classifier {}: WINs Yts {}, Ynw {}, accuracy {}: '.
format(learn.segment, learn.clf_name, learn.y_ts_wins,
len(Ynw[[Ynw == 'WIN']]), learn.accuracy))
print('Region {0}, Classifier {1}: Accuracy is: {2:5.2f}'.format(
learn.segment, learn.clf_name, learn.accuracy))
for experiment in mat_dict.keys():
gen1 = [
exp for exp in models.fdm.generate_li
if exp.experiment == experiment
][0]
ix = mat_dict[experiment]['ix']
prediction1 = models.prediction_cl(material=experiment,
descr='',
segment=learn.segment,
clf_name=learn.clf_name,
project='FDM',
fnresstssbm=Ynw[ix],
accuracy=learn.accuracy)
gen1.predictions.append(prediction1)
return