def main():
'''Create and save model
'''
path_to_module = '../tools/'
sys.path.append(path_to_module)
# load s3 read & write functions
import bototools as bt
import nlp_preprocessing as nlp
# if pulling from s3
#path_to_data = '../.keys/'
#file_name = 'csv_to_classify.json'
#bucket, key = bt.load_s3_location(path_to_data, file_name)
#df = bt.load_df_from_s3(bucket, key, compression='gzip')
# local data
path_to_data = '../data/multiclass/'
data_file = 'labeled_large.csv'
df = pd.read_csv(os.path.join(path_to_data, data_file))
print('loaded CSV')
# remove samples not of interestc
# map roles to general labels
key_path = '../.keys'
role_mapper = load_pickle(os.path.join(key_path, 'roles.pickle'))
role_names = [n for n in role_mapper]
df = df[df['role'].isin(role_names)]
df['label'] = df['role'].map(role_mapper)
df = df.dropna()
cols_to_train = ['title', 'label']
df = df[cols_to_train]
# standardize text format
df = nlp.standardize_text(df, 'title')
# select data to predict from
X = df['title'].tolist()
y = df['label'].tolist()
#y = df['gig'].tolist()
# dividing X, y into train and test data
print('vectorizing bag of words model')
X_counts, count_vectorizer = nlp.create_count_vectorizer(X) # count vec
cv_to_write = '../models/CV_nb_bow_model.pckl'
pickle.dump(count_vectorizer, open(cv_to_write, 'wb'))
print('successfully serialized CV model')
# training a Naive Bayes classifier
print('testing model')
#model = MultinomialNB().fit(X_counts, y)
model = ComplementNB().fit(X_counts, y)
# save model for later use (locally & on s3)
file_to_write = '../models/complement_nb_model.pckl'
pickle.dump(model, open(file_to_write, 'wb'))
print('successfully serialized NB model')
def main():
'''Run labeling on job data
Process directly from partner feeds
** NOTE - this is a "dumb" method without NLP/ML **
Current configuration:
+ Load csv data from xml feed
+ Label based on job title
+ Write results to s3
'''
# define paths
key_path = '/home/ubuntu/job-classifier/.keys/'
#model_path = '/home/ubuntu/job-classifier/models/'
# dict with partner:url pairs
partners = load_pickle(os.path.join(key_path, 'partners.pickle'))
s3_details = load_pickle(os.path.join(key_path, 's3_config.pickle'))
bucket = s3_details['csv_bucket']
file_to_write = s3_details['target']
# pull xml from url and parse into df
for partner in partners:
url = partners[partner]
if url.endswith('.xml.gz'):
df = xt.xml_from_url_compressed(url)
else:
df = xt.xml_from_url(url)
# standardize text format
df = nlp.standardize_text(df, 'title')
# apply label function
df['label'] = df.title.apply(get_role)
label_cols = ['label', 'company','title','city','state','url']
df_to_keep = df[~(df['label']=='ignore')][label_cols]
role_dfs = {}
for label in df_to_keep.label.unique():
tmp_df = df_to_keep[df_to_keep['label']==label]
if len(tmp_df) > 100:
tmp_df = tmp_df.sample(n=100)
role_dfs[label] = tmp_df
df_to_write = pd.concat([role_dfs[x] for x in role_dfs])
# write labeled roles
label_key = partner + '/' + 'labeled_jobs.csv'
print('Writing {} jobs'.format(len(df_to_write)))
bt.write_df_to_s3(df_to_write, bucket, label_key, comp=False)
def main():
# text cleaning
df = pd.read_csv() # or something
df = nlp.standardize_text(df, 'text')
def main():
'''Run classification on job data
Process directly from partner feeds
Current configuration:
+ Load csv data from xml feed
+ Count Vectorize with Bag of Words
+ Label using Naive Bayes Classifier
+ Write results back to s3
'''
# define paths
key_path = '/home/ubuntu/job-classifier/.keys/'
model_path = '/home/ubuntu/job-classifier/models/'
# dict with partner:url pairs
partners = load_pickle(os.path.join(key_path, 'partners.pickle'))
s3_details = load_pickle(os.path.join(key_path, 's3_config.pickle'))
bucket = s3_details['csv_bucket']
file_to_write = s3_details['target']
cv_model = load_pickle(os.path.join(model_path, 'CV_nb_bow_model.pckl'))
# update cv model above
#mnb_model = load_pickle(os.path.join(model_path, 'multi_nb_model.pckl'))
cnb_model = load_pickle(
os.path.join(model_path, 'complement_nb_model.pckl'))
# cities & roles
city_roles = {
'driver': ['new york', 'los angeles'],
'nurse': ['dallas', 'los angeles'],
'tech': ['san francisco', 'boston']
}
# pull xml from url and parse into df
for partner in partners:
print('\nProcessing:', partner.upper())
url = partners[partner]
if url.endswith('.xml.gz'):
df = xt.xml_from_url_compressed(url)
else:
df = xt.xml_from_url(url)
# standardize text format
df = nlp.standardize_text(df, 'title')
# select data to predict from
X_classify = df['title'].tolist()
# get count vec
X_classify_counts = nlp.get_cv_test_counts(X_classify, cv_model)
# predict with model
#y_label = mnb_model.predict(X_classify_counts)
y_label = cnb_model.predict(X_classify_counts)
# assign predictions to jobs & prune dataframe
df['label'] = y_label
label_cols = ['label', 'company', 'title', 'city', 'state', 'url']
#labels_to_drop = ['ignore','driver','service']
labels_to_drop = ['ignore', 'service']
df_tmp = df[~(df['label'].isin(labels_to_drop))][label_cols]
for role in city_roles:
print('processing:', role.upper())
for city in city_roles[role]:
print('city:', city.upper())
df_to_write = get_city_df(df, city, role)
# write labeled roles
city_file = city.replace(' ', '_')
label_key = partner + '/' + role + '/' + city_file + '/' + 'jobs.csv'
bt.write_df_to_s3(df_to_write, bucket, label_key, comp=False)
def main():
'''Run classification on job data
Process directly from partner feeds
Current configuration:
+ Load csv data from xml feed
+ Count Vectorize with Bag of Words
+ Label using Naive Bayes Classifier
+ Write results back to s3
'''
# define paths
key_path = '/home/ubuntu/job-classifier/.keys/'
model_path = '/home/ubuntu/job-classifier/models/'
# dict with partner:url pairs
partners = load_pickle(os.path.join(key_path, 'partners.pickle'))
s3_details = load_pickle(os.path.join(key_path, 's3_config.pickle'))
bucket = s3_details['csv_bucket']
file_to_write = s3_details['target']
cv_model = load_pickle(os.path.join(model_path,'CV_nb_bow_model.pckl'))
# update cv model above
#clf_model = load_pickle(os.path.join(model_path, 'lr_bow_train_only_model.pckl'))
#mnb_model = load_pickle(os.path.join(model_path, 'multi_nb_model.pckl'))
cnb_model = load_pickle(os.path.join(model_path, 'complement_nb_model.pckl'))
# pull xml from url and parse into df
for partner in partners:
url = partners[partner]
if url.endswith('.xml.gz'):
df = xt.xml_from_url_compressed(url)
else:
df = xt.xml_from_url(url)
# standardize text format
df = nlp.standardize_text(df, 'title')
# select data to predict from
X_classify = df['title'].tolist()
# get count vec
X_classify_counts = nlp.get_cv_test_counts(X_classify, cv_model)
# predict with model
#y_label = mnb_model.predict(X_classify_counts)
y_label = cnb_model.predict(X_classify_counts)
# assign predictions to jobs & prune dataframe
df['label'] = y_label
label_cols = ['label', 'company','title','city','state','url']
labels_to_drop = ['ignore','driver','service']
df_to_write = df[~(df['label'].isin(labels_to_drop))][label_cols]
#df_to_write = df[~(df['label']=='ignore')][label_cols]
# SAMPLE DF_TO_WRITE for smaller dataset
df_to_write = df_to_write.sample(n=100)
# write labeled roles
label_key = partner + '/' + 'labeled_jobs.csv'
bt.write_df_to_s3(df_to_write, bucket, label_key, comp=False)
def main():
'''Run classification on job data
Process directly from partner feeds
** NOTE - this is a "dumb" method without NLP/ML **
Current configuration:
+ Load csv data from xml feed
+ Label based on job title
+ Write results to s3
'''
# define paths
key_path = '/home/ubuntu/job-classifier/.keys/'
# dict with partner:url pairs
partners = load_pickle(os.path.join(key_path, 'partners.pickle'))
s3_details = load_pickle(os.path.join(key_path, 's3_config.pickle'))
city_roles = load_pickle(os.path.join(key_path, 'city_roles.pickle'))
bucket = s3_details['csv_bucket']
file_to_write = s3_details['target']
# pull xml from url and parse into df
for partner in partners:
print('\nProcessing:', partner.upper())
url = partners[partner]
if url.endswith('.xml.gz'):
df = xt.xml_from_url_compressed(url)
else:
df = xt.xml_from_url(url)
# standardize text format
df = nlp.standardize_text(df, 'title')
# assign labels to jobs & prune dataframe
df['label'] = df.title.apply(get_role)
label_cols = ['label', 'company', 'title', 'city', 'state', 'url']
df = df[~(df['label'] == 'ignore')][label_cols]
for role in city_roles:
print('processing:', role.upper())
for city in city_roles[role]:
print('city:', city.upper())
df_to_write = get_city_df(df, city, role)
# write labeled roles
city_file = city.replace(' ', '_')
label_key = partner + '/' + role + '/' + city_file + '/' + 'jobs.csv'
if len(df_to_write) > 0:
#print('write: {} to s3'.format(len(df_to_write)))
bt.write_df_to_s3(df_to_write,
bucket,
label_key,
comp=False)
else:
print('No matches found')
def main():
'''Run classification on job data
Current configuration:
+ Load csv data from xml feed
+ Count Vectorize with Bag of Words
+ Predict Linear Regression
+ Select results on LinReg class probability
+ Write results back to s3
'''
# add custom modules to path
path_to_module = '../tools/'
sys.path.append(path_to_module)
# load s3 connector & preprocessing functions
import bototools as bt
import xmltools as xt
import nlp_preprocessing as nlp
# load job data from s3 csv to dataframe
path_to_data = '../.keys/'
file_name = 'eda_data_file.json'
bucket, key, url, target = bt.load_s3_location(path_to_data, file_name)
# pull xml from url and parse into df
#df = bt.load_df_from_s3(bucket, key, comp='gzip')
df = xt.xml_from_url(url)
# standardize text format
cols_to_model = ['title']
for col in cols_to_model:
df = nlp.standardize_text(df, col)
# select data to predict from
X_classify = df['title'].tolist()
# define model path
path_to_models = '../models/'
# load count vectorizer & transform data to predict
cv_pickle = 'CV_lr_bow_train_only_model.pckl'
cv_path = os.path.join(path_to_models, cv_pickle)
cv_model = pickle.load(open(cv_path, 'rb'))
X_classify_counts = nlp.get_cv_test_counts(X_classify, cv_model)
# load, train, fit model
clf_pickle = 'lr_bow_train_only_model.pckl'
clf_path = os.path.join(path_to_models, clf_pickle)
clf_model = pickle.load(open(clf_path, 'rb'))
y_predicted = clf_model.predict(X_classify_counts)
#y_prob = clf_model.predict_proba(X_classify_counts)
# assign predictions to jobs & prune dataframe
df['gig'] = y_predicted
#df['prob'] = y_prob[:,0] # failed last test
cols_to_write = ['company','title','city','state','url']
#cols_to_write = ['company','title','city','state','posted_at','url']
# only keep listings with over 95% probability of being a gig job
# tighten/loosen requirement depending on model
#df_to_write = df[(df['gig']==1) & (df['prob']==0.95)][cols_to_write]
df_to_write = df[df['gig']==1][cols_to_write]
# write jobs to accessible location on s3
#file_to_write = 'gigs/streamed_full_daily_job_list.csv'
file_to_write = target
bt.write_df_to_s3(df_to_write, bucket, file_to_write, comp=False)
def main():
'''Test & Evaluate multiple models to select best option
Considerations:
+ Single-feature
+ Multiple features
+ Count vectorizer - BoW/Tfidf
+ k-fold x-validation
+ Choice of classifier (LR, RF, etc)
Metrics:
+ Accuracy
+ Variance
+ F1
+ Confusion matrix
+ Create graphics
+ Feature importance
'''
path_to_module = '../tools/'
sys.path.append(path_to_module)
# load s3 read & write functions
import bototools as bt
import nlp_preprocessing as nlp
print('classifying new jobs...\n')
# if pulling from s3
#path_to_data = '../.keys/'
#file_name = 'csv_to_classify.json'
#bucket, key = bt.load_s3_location(path_to_data, file_name)
#df = bt.load_df_from_s3(bucket, key, compression='gzip')
# local data
path_to_data = '../data/multiclass/'
data_file = 'labeled_large.csv'
df = pd.read_csv(os.path.join(path_to_data, data_file))
print('loaded CSV')
# map roles to general labels
key_path = '../.keys'
role_mapper = load_pickle(os.path.join(key_path, 'roles.pickle'))
role_names = [n for n in role_mapper]
df = df[df['role'].isin(role_names)]
df['label'] = df['role'].map(role_mapper)
df = df.dropna()
cols_to_train = ['title', 'label']
df = df[cols_to_train]
# standardize text format
df = nlp.standardize_text(df, 'title')
## run a 10-fold stratified cross validation
skf = StratifiedKFold(n_splits=10, random_state=0)
df = df.reset_index(drop=True)
X = df['title'].values.astype('U')
y = df['label']
current_split = 1
acc_lr = []
prec_lr = []
rec_lr = []
f1_lr = []
acc_mnb = []
prec_mnb = []
rec_mnb = []
f1_mnb = []
acc_cnb = []
prec_cnb = []
rec_cnb = []
f1_cnb = []
for train_index, test_index in skf.split(X, y):
print('CURRENT SPLIT:', current_split)
# get splits & assign data
X_train, X_test = X[train_index], X[test_index]
y_train, y_test = y[train_index], y[test_index]
#print('successful split')
# vectorize word counts
X_train_counts, count_vectorizer = nlp.create_count_vectorizer(X_train)
X_test_counts = count_vectorizer.transform(X_test)
# Logistic regression
lr = LogisticRegression(C=30.0,
class_weight='balanced',
solver='newton-cg',
multi_class='multinomial',
n_jobs=-1,
random_state=40)
lr.fit(X_train_counts, y_train)
lr_predictions = lr.predict(X_test_counts)
accuracy, precision, recall, f1 = get_metrics(y_test, lr_predictions)
print(
"LR: accuracy = %.3f, precision = %.3f, recall = %.3f, f1 = %.3f" %
(accuracy, precision, recall, f1))
acc_lr.append(accuracy)
prec_lr.append(precision)
rec_lr.append(recall)
f1_lr.append(f1)
# multinomial NB
mnb = MultinomialNB().fit(X_train_counts, y_train)
mnb_predictions = mnb.predict(X_test_counts)
accuracy, precision, recall, f1 = get_metrics(y_test, mnb_predictions)
print(
"MNB: accuracy = %.3f, precision = %.3f, recall = %.3f, f1 = %.3f"
% (accuracy, precision, recall, f1))
acc_mnb.append(accuracy)
prec_mnb.append(precision)
rec_mnb.append(recall)
f1_mnb.append(f1)
# complement NB
cnb = ComplementNB().fit(X_train_counts, y_train)
cnb_predictions = cnb.predict(X_test_counts)
accuracy, precision, recall, f1 = get_metrics(y_test, cnb_predictions)
print(
"CNB: accuracy = %.3f, precision = %.3f, recall = %.3f, f1 = %.3f"
% (accuracy, precision, recall, f1))
acc_cnb.append(accuracy)
prec_cnb.append(precision)
rec_cnb.append(recall)
f1_cnb.append(f1)
current_split += 1
# Summarize
print('\nFinal Perfomance')
print('\nLogistic Regression Perfomance')
print('Accuracy: mean %.3f, variance %.3f' %
(np.mean(acc_lr), np.var(acc_lr)))
print('Precision: mean %.3f, variance %.3f' %
(np.mean(prec_lr), np.var(prec_lr)))
print('Recall: mean %.3f, variance %.3f' %
(np.mean(rec_lr), np.var(rec_lr)))
print('F1 Score: mean %.3f, variance %.3f' %
(np.mean(f1_lr), np.var(f1_lr)))
print('\nMultinomial Naive Bayes Perfomance')
print('Accuracy: mean %.3f, variance %.3f' %
(np.mean(acc_mnb), np.var(acc_mnb)))
print('Precision: mean %.3f, variance %.3f' %
(np.mean(prec_mnb), np.var(prec_mnb)))
print('Recall: mean %.3f, variance %.3f' %
(np.mean(rec_mnb), np.var(rec_mnb)))
print('F1 Score: mean %.3f, variance %.3f' %
(np.mean(f1_mnb), np.var(f1_mnb)))
print('\nComplement Naive Bayes Perfomance')
print('Accuracy: mean %.3f, variance %.3f' %
(np.mean(acc_cnb), np.var(acc_cnb)))
print('Precision: mean %.3f, variance %.3f' %
(np.mean(prec_cnb), np.var(prec_cnb)))
print('Recall: mean %.3f, variance %.3f' %
(np.mean(rec_cnb), np.var(rec_cnb)))
print('F1 Score: mean %.3f, variance %.3f' %
(np.mean(f1_cnb), np.var(f1_cnb)))
def main():
'''Run classification on job data
Current configuration:
+ Load csv data from s3
+ Count Vectorize with Bag of Words
+ Predict Linear Regression
+ Select results on LinReg class probability
+ Write results back to s3
'''
# add custom modules to path
path_to_module = '/home/ubuntu/job-classifier/tools/'
sys.path.append(path_to_module)
# load s3 connector & preprocessing functions
import bototools as bt
import xmltools as xt
import nlp_preprocessing as nlp
# load job data from s3 csv to dataframe
path_to_data = '/home/ubuntu/job-classifier/.keys/'
file_name = 'cron_data_file.json'
bucket, key, url, target = bt.load_s3_location(path_to_data, file_name)
# pull xml from url and parse into df
df = xt.xml_from_url(url)
# standardize text format
cols_to_model = ['title']
for col in cols_to_model:
df = nlp.standardize_text(df, col)
# select data to predict from
X_classify = df['title'].tolist()
# define model path
path_to_models = '/home/ubuntu/job-classifier/models/'
# load count vectorizer & transform data to predict
cv_pickle = 'CV_lr_bow_train_only_model.pckl'
cv_path = os.path.join(path_to_models, cv_pickle)
cv_model = pickle.load(open(cv_path, 'rb'))
X_classify_counts = nlp.get_cv_test_counts(X_classify, cv_model)
# load, train, fit model
clf_pickle = 'lr_bow_train_only_model.pckl'
clf_path = os.path.join(path_to_models, clf_pickle)
clf_model = pickle.load(open(clf_path, 'rb'))
y_predicted = clf_model.predict(X_classify_counts)
y_prob = clf_model.predict_proba(X_classify_counts)
# assign predictions to jobs & prune dataframe
df['gig'] = y_predicted
cols_to_write = ['company','title','city','state','url']
df_to_write = df[df['gig']==1][cols_to_write]
# write jobs to accessible location on s3
# custom name by date -- test overlap between days
timestr = time.strftime("%Y-%m-%d")
prefix, fn = target.split('/')
file_to_write = prefix + '/' + timestr + '-' + fn
bt.write_df_to_s3(df_to_write, bucket, file_to_write, comp=False)
# add labeled samples to validate for future training
df_positive = df[df['gig']==1].sample(1000)
file_positive = 'positive' + '/' + timestr + '-' + fn
bt.write_df_to_s3(df_positive, bucket, file_positive, comp=False)
df_negative = df[df['gig']==0].sample(1000)
file_negative = 'negative' + '/' + timestr + '-' + fn
bt.write_df_to_s3(df_negative, bucket, file_negative, comp=False)
def main():
'''Test & Evaluate multiple models to select best option
Considerations:
+ Single-feature
+ Multiple features
+ Count vectorizer - BoW/Tfidf
+ k-fold x-validation
+ Choice of classifier (LR, RF, etc)
Metrics:
+ Accuracy
+ Variance
+ F1
+ Confusion matrix
+ Create graphics
+ Feature importance
'''
path_to_module = '../tools/'
sys.path.append(path_to_module)
# load s3 read & write functions
import bototools as bt
import nlp_preprocessing as nlp
print('classifying new jobs...\n')
# if pulling from s3
#path_to_data = '../.keys/'
#file_name = 'csv_to_classify.json'
#bucket, key = bt.load_s3_location(path_to_data, file_name)
#df = bt.load_df_from_s3(bucket, key, compression='gzip')
# local data
path_to_data = '../data/multiclass/'
data_file = 'labeled_large.csv'
df = pd.read_csv(os.path.join(path_to_data, data_file))
print('loaded CSV')
# map roles to general labels
key_path = '../.keys'
role_mapper = load_pickle(os.path.join(key_path, 'roles.pickle'))
role_names = [n for n in role_mapper]
df = df[df['role'].isin(role_names)]
df['label'] = df['role'].map(role_mapper)
df = df.dropna()
cols_to_train = ['title', 'label']
df = df[cols_to_train]
# standardize text format
df = nlp.standardize_text(df, 'title')
# select data to predict from
X = df['title'].tolist()
y = df['label'].tolist()
# dividing X, y into train and test data
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0)
print('vectorizing bag of words model')
X_train_counts, count_vectorizer = nlp.create_count_vectorizer(
X_train) # count vec
X_test_counts = count_vectorizer.transform(X_test)
# training a Naive Bayes classifier
print('testing model')
# Logistic regression
lr = LogisticRegression(C=30.0,
class_weight='balanced',
solver='newton-cg',
multi_class='multinomial',
n_jobs=-1,
random_state=40)
lr.fit(X_train_counts, y_train)
lr_predictions = lr.predict(X_test_counts)
# multinomial NB
mnb = MultinomialNB().fit(X_train_counts, y_train)
mnb_predictions = mnb.predict(X_test_counts)
# complement NB
cnb = ComplementNB().fit(X_train_counts, y_train)
cnb_predictions = cnb.predict(X_test_counts)
# evaluate performance
from sklearn.metrics import confusion_matrix
roles = ['tech', 'nurse', 'service', 'driver', 'ignore']
cm_lr = confusion_matrix(y_test, lr_predictions, labels=roles)
cm_mnb = confusion_matrix(y_test, mnb_predictions, labels=roles)
cm_cnb = confusion_matrix(y_test, cnb_predictions, labels=roles)
print('\n--- Logistic Regression Confusion Matrix ---')
print(roles)
print(cm_lr)
get_metrics(y_test, lr_predictions)
print('\n--- Multinomial Bayes Confusion Matrix ---')
print(roles)
print(cm_mnb)
get_metrics(y_test, mnb_predictions)
print('\n--- Complement Bayes Confusion Matrix ---')
print(roles)
print(cm_cnb)
get_metrics(y_test, cnb_predictions)
def main():
'''Run classification
'''
path_to_module = '../tools/'
sys.path.append(path_to_module)
# load s3 read & write functions
import bototools as bt # add these to actual path
print('classifying new jobs...\n')
path_to_data = '../.keys/'
#file_name = 'csv_to_classify.json'
file_name = 'eda_data_file.json'
#bucket, key = bt.load_s3_location(path_to_data, file_name)
# use a sample file for testing -- avoid large files for now
bucket, _ = bt.load_s3_location(path_to_data, file_name)
key = 'eda_sample_data_file.csv'
df = bt.load_df_from_s3(bucket, key) # sample not gzipped
#df = bt.load_df_from_s3(bucket, key, comp='gzip')
# import preprocessing tools
import nlp_preprocessing as nlp
# cleanup the dataframe to prepare for classification
# while only SOME columns are used, ALL need to be returned for ops team
cols_to_model = ['title']
for col in cols_to_model:
df = nlp.standardize_text(df, col)
X_classify = df['title'].tolist()
# load count vectorizer
path_to_models = '../models/'
cv_pickle = 'CV_lr_bow_train_only_model.pckl' # use private file too
cv_path = os.path.join(path_to_models, cv_pickle)
cv_model = pickle.load(open(cv_path, 'rb'))
X_classify_counts = nlp.get_cv_test_counts(X_classify, cv_model)
from sklearn.linear_model import LogisticRegression # move outside main
# load pre-trained model
#path_to_model = '../models/'
# functionalize loading & training
clf_pickle = 'lr_bow_train_only_model.pckl' # use private file too
clf_path = os.path.join(path_to_models, clf_pickle)
clf_model = pickle.load(open(clf_path, 'rb'))
y_predicted = clf_model.predict(X_classify_counts)
df['gig'] = y_predicted
cols_to_write = ['company','title','city','state','posted_at','url']
df_to_write = df[df['gig']==1][cols_to_write]
#df_to_write = df[df['gig']==1]
#df[df['gig']==1].to_csv('../data/classified_gig_jobs.csv', index=False)
#print('Gig jobs found: {}'.format(df[df['gig']==1].shape[0]))
# write output (use a prefix!)
file_to_write = 'gigs/sample_daily_job_list.csv'
bt.write_df_to_s3(df_to_write, bucket, file_to_write)