def main():
myco = Myco(file_path='data/gipsy.myc')
myco.load()
out_path = 'data/'
n_estimators, n_jobs = 60, 1
result = Result(dataset=myco, out_path=out_path)
model = Model(RandomForestClassifier,
out_path,
n_estimators=n_estimators,
verbose=True,
n_jobs=n_jobs)
partition = model._get_partition(myco, n_cores=n_jobs)
test_indices = [1, 23, 45, 12]
model.fit(partitions=partitions,
populations=myco.populations,
exclude_indices=test_indices)
probabilities = model.predict_proba(partitions=partitions,
include_indices=test_indices)
classes = model.classes_
top_class_ids = np.argmax(probabilities, axis=1)
top_probabilities = np.max(probabilities, axis=1)
predicted_origin = [classes[class_id] for class_id in top_class_ids]
result.set_q_pred_pops_params(predicted_origin,
top_probabilities,
classes,
test_indices=test_indices)
result.output_q()
mixture_plot(result)
def main():
myco = Myco(file_path='data/gipsy.myc')
myco.load()
cv = CrossValidate(dataset=myco, out_path='data/')
cv.run(n_partitions=1, n_loci=0, n_splits=5, n_estimators=60, n_cores=1)
mixture_plot(cv)
def main():
myco = Myco(file_path='data/myco.myc', is_str=True)
myco = Structure(file_path='data/myco.str', is_str=True)
myco.load()
cv = CrossValidate(dataset=myco, out_path='data/')
cv.run(n_partitions=1, n_loci=0, n_splits=5, n_estimators=60, n_cores=1)
mixture_plot(cv)
def main():
myco = Myco(file_path='data/gipsy.myc')
myco.load()
cv = Supervised(dataset=myco, out_path='data/')
cv.run(test=[1, 23, 45, 12],
n_partitions=1,
n_loci=0,
n_estimators=60,
n_cores=1)
mixture_plot(cv)
def main():
myco = Myco(file_path='examples/data/gipsy.myc')
myco.load()
populations = np.array(myco.populations)
out_path = 'data/'
n_splits = 5
n_estimators, n_jobs = 60, 1 # RandomForestClassifier parameters
n_partitions, n_loci = 1, 0 # _partitions parameters
# Check if population can be split in @n_splits splits
counts = Counter(populations)
for key in counts:
if counts[key] < n_splits:
raise ValueError(
"Population {0} has less samples ({1}) than the number of splits ({2})."
.format(key, counts[key], num_splits))
predicted_origin_out, mixture_estimate_out, ordering_out = [], [], []
# As a alternative - use shortcut: Model(...)._get_partition(...)
partitions = _partition(myco, out_path, n_partitions, n_loci, n_jobs)
# Splits dataset with @n_splits splits where each
kf = StratifiedKFold(n_splits=n_splits, shuffle=True)
# Each train/test set is near balanced split of data (label-wise)
for train, test in kf.split(partitions[0], y=populations):
mixture_estimates_per_partition = []
model = Model(RandomForestClassifier,
out_path,
n_estimators=n_estimators,
verbose=True,
n_jobs=n_jobs)
for part in partitions:
# Train
model.fit(partitions=part,
populations=populations,
include_indices=train)
# Predict probabilities for observed classes
mixture = model.predict_proba(partitions=part,
include_indices=test)
mixture_estimates_per_partition.append(mixture)
# average the results among different trials and find predicted origins
mixture_estimate = np.mean(mixture_estimates_per_partition, axis=0)
predicted_origin = [
model.classes_[ind] for ind in np.argmax(mixture_estimate, axis=1)
]
# accumulate the results
predicted_origin_out.extend(predicted_origin)
mixture_estimate_out.extend(mixture_estimate.tolist())
ordering_out.extend(test.tolist())
# find indices for smart sorting
indices = np.argsort(ordering_out)
# sort results according to the
pred_pops = np.array(predicted_origin_out)[indices]
q = np.array(mixture_estimate_out)[indices]
q_pops = sorted(set(populations))
# Save data to the Result object and output data to the @out_path
result = Result(dataset=myco, out_path=out_path)
result.set_pred_pops(pred_pops)
result.set_q(q, q_pops)
result.output_q()
result.output_accuracy()
# Plot the results
mixture_plot(result)
def supervised():
parser = argparse.ArgumentParser()
parser.add_argument('-i',
'--in-file',
type=str,
required=True,
help='Path to the input data file.')
parser.add_argument('-o',
'--out',
type=str,
required=True,
help='Path to the output folder.')
parser.add_argument('-P',
'--partitions',
type=int,
default=1,
help='The number of partitions (default 1).')
parser.add_argument(
'-M',
'--loci',
type=int,
default=0,
help='The number of randomly selected loci (0 for all).')
parser.add_argument(
'-e',
'--estimators',
type=int,
default=60,
help='Number of trees in the Random Forest classifier (default 60).')
parser.add_argument('-c',
'--cores',
type=int,
default=1,
help='The number of cores (default 1).')
parser.add_argument(
'-x',
'--splitstree',
type=str,
default=None,
help='Path to the SplitsTree executable (default PATH).')
parser.add_argument('-f',
'--format',
type=str,
default='myco',
choices=['myco', 'struct'],
help='Data file format (default myco).')
parser.add_argument('-t',
'--type',
type=str,
default='snp',
choices=['snp', 'str'],
help='SNP or microsatellite data.')
args = parser.parse_args()
if args.splitstree is not None:
const['__SPLITSTREE_PATH__'] = args.splitstree
if args.format == 'myco':
myco = Myco(file_path=args.in_file,
is_str=True if args.type == 'str' else False)
elif args.format == 'struct':
myco = Structure(file_path=args.in_file,
is_str=True if args.type == 'str' else False)
else:
raise ValueError('Invalid format.')
myco.load()
cv = Supervised(dataset=myco, out_path=args.out)
cv.run(n_partitions=args.partitions,
n_loci=args.loci,
n_estimators=args.estimators,
n_cores=args.cores)
mixture_plot(cv, predictionOnly=True)