# =====
money: float = 0.0
while True:
# Get user selection
selection = functions.user_selection()
if selection == "":
continue
# Take action to perform user selection
elif selection == 3:
functions.switch_off()
elif selection == 4:
functions.print_report(money)
else:
# Check sufficient ingredients to make the drink
sufficient = functions.check_ingredients(selection)
if not sufficient:
print("Maintenance required")
if sufficient:
# Sufficient ingredients to make the drink
# Ask for payment
result = functions.get_payment(selection, money)
paid = result[0]
money = result[1]
if paid:
#ETHNIC_SLURS = set(load_ethnic_slurs())
training_data_path = build_data_path('augmented_train.csv')
df = pd.read_csv(training_data_path)
X = df['comment_text']
y = df[LABEL_COLS]
X_train, X_valid, y_train, y_valid = train_test_split(X, y, test_size=0.33)
clf = OneVsRestClassifier(MultinomialNB())
tfidf = TfidfVectorizer(strip_accents='ascii',
stop_words='english',
ngram_range=(1, 1),
norm='l2')
bad_word_counter = CountVectorizer(vocabulary=BAD_WORDS)
#slur_counter = CountVectorizer(vocabulary=ETHNIC_SLURS)
#union = make_union(tfidf, bad_word_counter, slur_counter)
union = make_union(tfidf, bad_word_counter)
pipeline = make_pipeline(union, clf)
optimizer = pipeline
optimizer.fit(X_train, y_train)
y_predictions = optimizer.predict(X_valid)
# best_estimator_ = optimizer.best_estimator_
print_report(y_valid, y_predictions)
y = df[LABEL_COLS]
X_train, X_valid, y_train, y_valid = train_test_split(X, y, test_size=0.33)
label_counts = y_train.sum()
total = label_counts.sum()
label_freqs = defaultdict(lambda: 0)
for key in label_counts.index:
label_freqs[key] = label_counts[key] / total
def predict(X_values):
predictions = []
for example in X_values:
prediction = []
for key in LABEL_COLS:
rand_value = random()
frequency = label_freqs[key]
prediction.append(1 if rand_value < frequency else 0)
predictions.append(prediction)
return np.array(predictions)
random_predictions = predict(X_valid)
print('Baseline Data')
print_report(y_valid, random_predictions)
print()
scoring='f1_weighted',
verbose=3)
fit_params = {
'randomforestclassifier__sample_weights':
compute_sample_weight('balanced', y_train)
}
optimizer.fit(X_train, y_train)
y_predictions = optimizer.predict(X_valid)
print(y_predictions.shape, y_valid.shape)
# best_estimator_ = optimizer.best_estimator_
metrics.roc_auc_score(y_valid, y_predictions)
test_data = build_data_path('test.csv')
data_df = pd.read_csv(test_data)
test_labels = build_data_path('test_labels.csv')
label_df = pd.read_csv(test_labels)
test_df = data_df.set_index('id').join(label_df.set_index('id'))
CONDITIONS = [f'{label} != -1' for label in LABEL_COLS]
QUERY_STRING = ' & '.join(CONDITIONS)
test_df = test_df.query(QUERY_STRING)
X_test = test_df['comment_text']
y_test = test_df[LABEL_COLS]
y_predictions = optimizer.predict(X_test)
print_report(y_test, y_predictions, data_type='TESTING')
# The Inspiration for the Code
# has been taken from
# https://github.com/SaltyQuetzals/Toxicroak