def random_forest():
return ml_technique.MLTechnique(name='Random Forest Predictor', data_file=data_file,
model=ensemble.RandomForestClassifier(),
features=features,
ml_code='random',
pickle_file=pickle_file,
nlp=nlp,
data_frame=data_frame)
def k_neighbours():
return ml_technique.MLTechnique(name='k-Nearest Neighbours Predictor', data_file=data_file,
model=neighbors.KNeighborsClassifier(),
features=features,
ml_code='k',
pickle_file=pickle_file,
nlp=nlp,
data_frame=data_frame)
def naive():
return ml_technique.MLTechnique(name='Naive Bayes Predictor', data_file=data_file,
model=naive_bayes.GaussianNB(),
features=features,
ml_code='naive',
pickle_file=pickle_file,
nlp=nlp,
data_frame=data_frame)
def log_reg():
return ml_technique.MLTechnique(name='Logistic Regression Predictor', data_file=data_file,
model=linear_model.LogisticRegression(),
features=features,
ml_code='log_reg',
pickle_file=pickle_file,
nlp=nlp,
data_frame=data_frame)
def d_tree():
return ml_technique.MLTechnique(name='Decision Tree Predictor', data_file=data_file,
model=tree.DecisionTreeClassifier(),
features=features,
ml_code='d_tree',
pickle_file=pickle_file,
nlp=nlp,
data_frame=data_frame)
def svm_tech():
return ml_technique.MLTechnique(name='SVM Predictor', data_file=data_file,
model=svm.SVC(kernel='linear', C=1, gamma=1),
features=features,
ml_code='svm',
pickle_file=pickle_file,
nlp=nlp,
data_frame=data_frame)
def calculate_statistics(input_path, output_path, features, forced=False):
file = pd.ExcelFile(input_path)
stat_data = file.parse('Balance', index_col='jobid')
file.close()
stat_nlp = read_instance_pickle('pickle/stat_nlp.pickle')
if stat_nlp is None:
stat_nlp = NLPTechnique(name='Stat NLP',
nlp_vader=True,
nlp_bow=True,
pp_lemma=True,
pp_pos=True,
pickle_file='pickle/stat_nlp.pickle')
stat_data = stat_nlp.process_data(stat_data, forced=forced)
attr = features + ['category']
stat_data = stat_data[attr]
# stat_data[features] = stat_data[features].apply(pd.to_numeric, axis='columns')
# stat_data = stat_data.fillna(method='ffill')
stat.bucketing(stat_data, output_path)
svm_tech = read_instance_pickle('pickle/svm_tech.pickle')
dt_tech = read_instance_pickle('pickle/dt_tech.pickle')
log_reg_tech = read_instance_pickle('pickle/log_reg_tech.pickle')
k_tech = read_instance_pickle('pickle/k_tech.pickle')
nb_tech = read_instance_pickle('pickle/nb_tech.pickle')
rf_tech = read_instance_pickle('pickle/rf_tech.pickle')
if None in (svm_tech, dt_tech, log_reg_tech, k_tech, nb_tech, rf_tech):
# Technique instances
print("cannot find instances")
svm_tech = ml_technique.MLTechnique(name='Support Vector Machine (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=svm.SVC(kernel='linear', C=1, gamma=1),
pickle_file='pickle/svm_tech.pickle',
features=features)
dt_tech = ml_technique.MLTechnique(name='Decision Tree (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=tree.DecisionTreeClassifier(),
pickle_file='pickle/dt_tech.pickle',
features=features)
log_reg_tech = ml_technique.MLTechnique(name='Logistic Regression (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=linear_model.LogisticRegression(),
pickle_file='pickle/log_reg_tech.pickle',
features=features)
k_tech = ml_technique.MLTechnique(name='k-Nearest Neighbour (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=neighbors.KNeighborsClassifier(),
pickle_file='pickle/k_tech.pickle',
features=features)
nb_tech = ml_technique.MLTechnique(name='Naive Bayes (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=naive_bayes.GaussianNB(),
pickle_file='pickle/nb_tech.pickle',
features=features)
rf_tech = ml_technique.MLTechnique(name='Random Forest (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=ensemble.RandomForestClassifier(),
pickle_file='pickle/rf_tech.pickle',
features=features)
# DUP
elif getattr(svm_tech, 'features') != features:
print('instance has different features')
svm_tech = ml_technique.MLTechnique(name='Support Vector Machine (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=svm.SVC(kernel='linear', C=1, gamma=1),
pickle_file='pickle/svm_tech.pickle',
features=features)
dt_tech = ml_technique.MLTechnique(name='Decision Tree (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=tree.DecisionTreeClassifier(),
pickle_file='pickle/dt_tech.pickle',
features=features)
log_reg_tech = ml_technique.MLTechnique(name='Logistic Regression (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=linear_model.LogisticRegression(),
pickle_file='pickle/log_reg_tech.pickle',
features=features)
k_tech = ml_technique.MLTechnique(name='k-Nearest Neighbour (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=neighbors.KNeighborsClassifier(),
pickle_file='pickle/k_tech.pickle',
features=features)
nb_tech = ml_technique.MLTechnique(name='Naive Bayes (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=naive_bayes.GaussianNB(),
pickle_file='pickle/nb_tech.pickle',
features=features)
rf_tech = ml_technique.MLTechnique(name='Random Forest (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=ensemble.RandomForestClassifier(),
pickle_file='pickle/rf_tech.pickle',
features=features)
# Calculate statistics
svm_a = stat.get_accuracy(svm_tech, forced=forced)
dt_a = stat.get_accuracy(dt_tech, forced=forced)
log_reg_a = stat.get_accuracy(log_reg_tech, forced=forced)
k_a = stat.get_accuracy(k_tech, forced=forced)
nb_a = stat.get_accuracy(nb_tech, forced=forced)
rf_a = stat.get_accuracy(rf_tech, forced=forced)
svm_f = stat.get_f1_score(svm_tech, forced=forced)
dt_f = stat.get_f1_score(dt_tech, forced=forced)
log_reg_f = stat.get_f1_score(log_reg_tech, forced=forced)
k_f = stat.get_f1_score(k_tech, forced=forced)
nb_f = stat.get_f1_score(nb_tech, forced=forced)
rf_f = stat.get_f1_score(rf_tech, forced=forced)
# Add time stamp to the result
tt = datetime.datetime.now().strftime('%H:%M %d/%m/%Y')
print("prediction done")
return (
{
'svm': svm_a,
'dt': dt_a,
'log_reg': log_reg_a,
'k': k_a,
'nb': nb_a,
'rf': rf_a
},
{
'svm': svm_f,
'dt': dt_f,
'log_reg': log_reg_f,
'k': k_f,
'nb': nb_f,
'rf': rf_f
},
tt
)
def ml_analysis():
category_column = 'category'
category_map = {
"Detractor": 0,
"Passive": 1,
"Promoter": 2
}
# Get the saved ML techniques
svm_tech = read_instance_pickle('pickle/svm_tech.pickle')
dt_tech = read_instance_pickle('pickle/dt_tech.pickle')
log_reg_tech = read_instance_pickle('pickle/log_reg_tech.pickle')
k_tech = read_instance_pickle('pickle/k_tech.pickle')
nb_tech = read_instance_pickle('pickle/nb_tech.pickle')
rf_tech = read_instance_pickle('pickle/rf_tech.pickle')
# If the ML technique instances does not exist, create new instances
if None in (svm_tech, dt_tech, log_reg_tech, k_tech, nb_tech, rf_tech):
# Technique instances
print("cannot find instances")
features = ['neu', 'neg', 'pos', 'compound', 'bag_vector', 'on_time_in_full', 'deliveryday', 'region']
svm_tech = ml_technique.MLTechnique(name='Support Vector Machine (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=svm.SVC(kernel='linear', C=1, gamma=1),
pickle_file='pickle/svm_tech.pickle',
features=features)
dt_tech = ml_technique.MLTechnique(name='Decision Tree (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=tree.DecisionTreeClassifier(),
pickle_file='pickle/dt_tech.pickle',
features=features)
log_reg_tech = ml_technique.MLTechnique(name='Logistic Regression (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=linear_model.LogisticRegression(),
pickle_file='pickle/log_reg_tech.pickle',
features=features)
k_tech = ml_technique.MLTechnique(name='k-Nearest Neighbour (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=neighbors.KNeighborsClassifier(),
pickle_file='pickle/k_tech.pickle',
features=features)
nb_tech = ml_technique.MLTechnique(name='Naive Bayes (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=naive_bayes.GaussianNB(),
pickle_file='pickle/nb_tech.pickle',
features=features)
rf_tech = ml_technique.MLTechnique(name='Random Forest (w SA)',
data_file='artifacts/sel_data/master.xlsx',
model=ensemble.RandomForestClassifier(),
pickle_file='pickle/rf_tech.pickle',
features=features)
# Cross Validation
svm_cv = svm_tech.cross_validate(cat_col=category_column, cat_map=category_map)
dt_cv = dt_tech.cross_validate(cat_col=category_column, cat_map=category_map)
log_reg_cv = log_reg_tech.cross_validate(cat_col=category_column, cat_map=category_map)
k_cv = k_tech.cross_validate(cat_col=category_column, cat_map=category_map)
nb_cv = nb_tech.cross_validate(cat_col=category_column, cat_map=category_map)
rf_cv = rf_tech.cross_validate(cat_col=category_column, cat_map=category_map)
# Local "packaging/parsing" of cross validation results to be rendered
def analyze_result(name, cv_result):
cm = confusion_matrix(cv_result['true_array'], cv_result['predict_array'])
return {
'name': name,
'labels': cv_result['categories'],
'matrix': cm,
'analysis': {
'input_size': cv_result['total_size']
}
}
return {
'techniques': ['svm', 'dt', 'log_reg', 'k', 'nb', 'rf'],
'results': {
'svm': analyze_result(svm_tech.get_name(), svm_cv),
'dt': analyze_result(dt_tech.get_name(), dt_cv),
'log_reg': analyze_result(log_reg_tech.get_name(), log_reg_cv),
'k': analyze_result(k_tech.get_name(), k_cv),
'nb': analyze_result(nb_tech.get_name(), nb_cv),
'rf': analyze_result(rf_tech.get_name(), rf_cv)
}
}