def start(username):
# register start time for this user
utils_users.get_experiment_start_time(username, 'pairwise')
# initialise acquirer
acquirer = DiscreteAcquirer(
input_domain=data_traffic.get_traffic_results(),
query_type='clustering',
seed=specs_traffic.SEED)
# get the starting points from the acquirer
item1, item2 = -1 * np.array(acquirer.get_start_points())
# get the objective names
obj_names = data_traffic.get_objective_names()
obj_abbrev = data_traffic.get_objective_abbrev()
return render_template("query_pairwise_traffic.html",
username=username,
item1=item1,
item2=item2,
side_clicked=-1,
obj_names=obj_names,
obj_abbrev=obj_abbrev)
def start(username):
# register start time for this user
utils_users.get_experiment_start_time(username, 'pairwise')
# get the starting points from the acquirer
job1, job2 = utils_jobs.get_next_start_jobs(username)
# transform the two jobs into dictionaries
job1 = utils_jobs.job_array_to_job_dict(job1)
job2 = utils_jobs.job_array_to_job_dict(job2)
return render_template("query_pairwise_jobs.html", username=username, job1=job1, job2=job2, side_clicked=-1)
def submit_ranking():
# get the username and their dataset
username = request.form['username']
user_dataset = utils_users.get_gp_dataset(
username, 'ranking', num_objectives=specs_jobs.NUM_OBJECTIVES)
# get the ranking the user submitted
ranking = np.fromstring(request.form['rankingResult'],
sep=',',
dtype=np.int)
# save it
utils_users.save_ranking(username, ranking)
# get the actual jobs (ranking returned the indiced in the dataset)
jobs_ranked = user_dataset.datapoints[ranking]
# add the ranking to the dataset
user_dataset.add_ranked_preferences(jobs_ranked)
# save
utils_users.update_gp_dataset(username, user_dataset, 'ranking')
# get the start time for this user
start_time = utils_users.get_experiment_start_time(username, 'ranking')
if time.time() - start_time < specs_jobs.TIME_EXPERIMENT_SEC:
return redirect(url_for('.start', username=username))
else:
utils_users.update_experiment_status(username=username,
query_type='ranking')
return redirect('start_experiment/{}'.format(username))
def submit_clustering_jobs():
# get the username and their dataset
username = request.form['username']
user_dataset = utils_users.get_gp_dataset(username, 'clustering', num_objectives=specs_jobs.NUM_OBJECTIVES)
# get the clusters the user submitted
top_job = [int(request.form['top-cluster'])]
good_jobs = np.fromstring(request.form['good-cluster'], sep=',', dtype=np.int)
bad_jobs = np.fromstring(request.form['bad-cluster'], sep=',', dtype=np.int)
clustering = [top_job, good_jobs, bad_jobs]
# save it
utils_users.save_clustering(username, clustering)
# get the actual jobs (ranking returned the indiced in the dataset)
top_job = user_dataset.datapoints[top_job]
good_jobs = user_dataset.datapoints[good_jobs]
bad_jobs = user_dataset.datapoints[bad_jobs]
clustering = [top_job, good_jobs, bad_jobs]
# add the ranking to the dataset
user_dataset.add_clustered_preferences(clustering)
# save
utils_users.update_gp_dataset(username, user_dataset, 'clustering')
# get the start time for this user
start_time = utils_users.get_experiment_start_time(username, 'clustering')
if time.time()-start_time < specs_jobs.TIME_EXPERIMENT_SEC:
return redirect(url_for('.start', username=username))
else:
utils_users.update_experiment_status(username=username, query_type='clustering')
return redirect('start_experiment/{}'.format(username))
def continue_pairwise(username, side_clicked):
# get the dataset for this user
dataset_user = utils_users.get_gp_dataset(
username, 'pairwise', num_objectives=specs_jobs.NUM_OBJECTIVES)
# initialise the acquirer which picks new datapoints
acquirer = DiscreteAcquirer(input_domain=utils_jobs.get_jobs(),
query_type='pairwise',
seed=specs_jobs.SEED)
# intialise the GP
gp = GPPairwise(num_objectives=specs_jobs.NUM_OBJECTIVES,
seed=specs_jobs.SEED)
# add collected datapoints to acquirer
acquirer.history = dataset_user.datapoints
# add collected datapoints to GP
gp.update(dataset_user)
# get the best job so far
job_best_idx = dataset_user.comparisons[-1, 0]
job_best = dataset_user.datapoints[job_best_idx]
# let acquirer pick new point
job_new = acquirer.get_next_point(gp, dataset_user)
# sort according to what user did last round
if side_clicked == "1":
job1 = job_best
job2 = job_new
else:
job1 = job_new
job2 = job_best
# transform the two jobs into dictionaries
job1 = utils_jobs.job_array_to_job_dict(job1)
job2 = utils_jobs.job_array_to_job_dict(job2)
# get the start time for this user
start_time = utils_users.get_experiment_start_time(username, 'pairwise')
if time.time() - start_time < specs_jobs.TIME_EXPERIMENT_SEC:
return render_template("query_pairwise_jobs.html",
username=username,
job1=job1,
job2=job2,
side_clicked=side_clicked)
else:
utils_users.update_experiment_status(username=username,
query_type='pairwise')
return redirect('start_experiment/{}'.format(username))
def submit_ranking():
# get the username and their dataset
username = request.form['username']
user_dataset = utils_users.get_gp_dataset(username, 'ranking', num_objectives=specs_traffic.NUM_OBJECTIVES)
# get the ranking the user submitted
ranking = np.fromstring(request.form['rankingResult'], sep=',', dtype=np.int)
# save it
utils_users.save_ranking(username, ranking)
# get the actual traffic (ranking returned the indiced in the dataset)
traffic_ranked = user_dataset.datapoints[ranking]
# add the ranking to the dataset
user_dataset.add_ranked_preferences(traffic_ranked)
print("comparisons from curr ranking", user_dataset.comparisons)
# save
utils_users.update_gp_dataset(username, user_dataset, 'ranking')
# get the start time for this user
utils_users.get_experiment_start_time(username, 'ranking')
button_type = request.form['buttonType']
if button_type == 'next':
return redirect(url_for('.start', username=username))
elif button_type == 'end':
# save end time
utils_users.save_experiment_end_time(username, 'ranking')
# register that this experiment was done
utils_users.update_experiment_status(username=username, query_type='ranking')
return redirect('start_experiment/{}'.format(username))
else:
raise NotImplementedError('Button type not defined')
def start(username):
# get the start time for this user
utils_users.get_experiment_start_time(username, 'ranking')
# get the dataset for this user
user_dataset = utils_users.get_gp_dataset(username, 'ranking', num_objectives=specs_traffic.NUM_OBJECTIVES)
# initialise acquirer
acquirer = DiscreteAcquirer(input_domain=data_traffic.get_traffic_results(), query_type='clustering',
seed=specs_traffic.SEED)
# if no data has been collected yet, we only display two starting traffic
if user_dataset.comparisons.shape[0] == 0:
# delete any datapoint in the user's dataset (in case experiment was aborted)
user_dataset.datapoints = np.empty((0, specs_traffic.NUM_OBJECTIVES))
# get the starting points from the acquirer
item1, item2 = acquirer.get_start_points()
# add traffic to dataset of user
item1_ID = user_dataset._add_single_datapoint(item1)
item2_ID = user_dataset._add_single_datapoint(item2)
# save dataset
utils_users.update_gp_dataset(username, user_dataset, 'ranking')
# put traffic we want to display in the respective lists
traffic_unranked = [item1, item2]
IDs_unranked = [item1_ID, item2_ID]
traffic_ranked = []
IDs_ranked = []
# otherwise, we show the previous ranking and pick a new point according to that
else:
# add collected datapoints to acquirer
acquirer.history = user_dataset.datapoints
# add virtual comparisons in first few queries
for i in range(np.min([user_dataset.datapoints.shape[0], 6])):
user_dataset.add_single_comparison(user_dataset.datapoints[i], data_traffic.get_traffic_min())
user_dataset.add_single_comparison(data_traffic.get_traffic_max(), user_dataset.datapoints[i])
# add linear prior in first few queries
if acquirer.history.shape[0] < 6:
prior_mean_type = 'linear'
else:
prior_mean_type = 'zero'
print("comparisons after adding stuff", user_dataset.comparisons)
# intialise the GP
gp = GPPairwise(num_objectives=specs_traffic.NUM_OBJECTIVES, seed=specs_traffic.SEED, prior_mean_type=prior_mean_type)
# add collected datapoints to GP
gp.update(user_dataset)
# let acquirer pick new point
job_new = acquirer.get_next_point(gp, user_dataset)
# add that point to the dataset and save
job_new_ID = user_dataset._add_single_datapoint(job_new)
utils_users.update_gp_dataset(username, user_dataset, 'ranking')
# put into list of traffic that need to be ranked
traffic_unranked = [job_new]
IDs_unranked = [job_new_ID]
# get ranking so far
IDs_ranked = utils_users.get_ranking(username)
# get the job information from that ranking and convert to dictionaries
traffic_ranked = user_dataset.datapoints[IDs_ranked]
# get names of objectives
obj_names = data_traffic.get_objective_names()
obj_abbrev = data_traffic.get_objective_abbrev()
return render_template("query_ranking_traffic.html", username=username, traffic_unranked=-1*np.array(traffic_unranked),
traffic_ranked=-1*np.array(traffic_ranked), IDs_ranked=IDs_ranked, IDs_unranked=IDs_unranked,
obj_names=obj_names, obj_abbrev=obj_abbrev)
def start(username):
# get the start time for this user
utils_users.get_experiment_start_time(username, 'ranking')
# get the dataset for this user
user_dataset = utils_users.get_gp_dataset(
username, 'ranking', num_objectives=specs_jobs.NUM_OBJECTIVES)
# if no data has been collected yet, we only display two starting jobs
if user_dataset.comparisons.shape[0] == 0:
# delete any datapoint in the user's dataset (in case experiment was aborted)
user_dataset.datapoints = np.empty((0, specs_jobs.NUM_OBJECTIVES))
# get the starting points from the acquirer
job1, job2 = utils_jobs.get_next_start_jobs(username)
# add jobs to dataset of user
job1_idx = user_dataset._add_single_datapoint(job1)
job2_idx = user_dataset._add_single_datapoint(job2)
# save dataset
utils_users.update_gp_dataset(username, user_dataset, 'ranking')
# convert into displayable format
job1 = utils_jobs.job_array_to_job_dict(job1)
job2 = utils_jobs.job_array_to_job_dict(job2)
# add ID to the above dictionaries (equals the index in the dataset
job1['ID'] = job1_idx
job2['ID'] = job2_idx
# put jobs we want to display in the respective lists
jobs_unranked = [job1, job2]
jobs_ranked = []
# otherwise, we show the previous ranking and pick a new point according to that
else:
# intialise the GP
gp = GPPairwise(num_objectives=specs_jobs.NUM_OBJECTIVES,
seed=specs_jobs.SEED)
# initialise acquirer
acquirer = DiscreteAcquirer(input_domain=utils_jobs.get_jobs(),
query_type='clustering',
seed=specs_jobs.SEED)
# add collected datapoints to acquirer
acquirer.history = user_dataset.datapoints
# add collected datapoints to GP
gp.update(user_dataset)
# let acquirer pick new point
job_new = acquirer.get_next_point(gp, user_dataset)
# add that point to the dataset and save
job_new_idx = user_dataset._add_single_datapoint(job_new)
utils_users.update_gp_dataset(username, user_dataset, 'ranking')
# convert job to dictionary
job_new = utils_jobs.job_array_to_job_dict(job_new)
# add the ID
job_new['ID'] = job_new_idx
# put into list of jobs that need to be ranked
jobs_unranked = [job_new]
# get ranking so far
ranking = utils_users.get_ranking(username)
# get the job information from that ranking and convert to dictionaries
jobs_ranked = user_dataset.datapoints[ranking]
jobs_ranked = [
utils_jobs.job_array_to_job_dict(job) for job in jobs_ranked
]
# add the IDs
for i in range(len(ranking)):
jobs_ranked[i]['ID'] = ranking[i]
return render_template("query_ranking_jobs.html",
username=username,
jobs_unranked=jobs_unranked,
jobs_ranked=jobs_ranked)