def data_from_config(config_file):
"""
Returns features, scores, elapsed time in seconds etc.
and searchspace from a config file
"""
config = Config(config_file)
session = config.trials()
searchspace = config.search_space()
history = [[t.parameters, t.test_scores, t.status, t.elapsed]
for t in session.query(Trial).all()]
return get_data(history, searchspace) + (searchspace, )
print('usage: sample_db.py config.yaml sample_size n_samples')
exit(1)
inp_file = argv[1]
num = int(argv[2]) # TOTAL size of samples to use e.g. 100
iter = int(argv[3]) # Number of splits e.g. 5
# This will give 5 splits of 20 samples each.100
if num % iter != 0:
print('sample_size not strictly divisible by n_samples')
exit(1)
# Get original database and history
config1 = Config(inp_file)
df1 = config1.trial_results()
hist1 = config1.trials().query(Trial).all()
# Main loop
for name, group in df1.groupby('project_name'):
# Sample the group
sample = group.sample(num, random_state=42)
cv = KFold(n_splits=iter, random_state=42)
all_keep = sample['id'].values
for i, (_, test_idx) in enumerate(cv.split(all_keep)):
keep = all_keep[test_idx]
db2 = make_session('sqlite:///osprey-trials-{0}-{1}.db'.format(int(num/iter), i), project_name=name)
# Get the relevant trial objects from original db