def get_positive_neighbors_models(in_svm_data_file_template, output_file, wl_iter_range,
k_range=None, radius_range=None, override_tergets_function=None):
assert bool(k_range) ^ bool(radius_range)
def compute_model(X, y, param, param_is_k):
_, avg_count, avg_prop = get_positive_neighbors_counts(X, y, k=param if param_is_k else None, radius=None if param_is_k else param)
model = {"wl_iterations": w, "avg_count": avg_count, "avg_prop": avg_prop}
if param_is_k:
model["k"] = param
else:
model["radius"] = param
return model
param_range = k_range if k_range else radius_range
param_is_k = bool(k_range)
with open(output_file, "w") as fl:
for w in wl_iter_range:
data_file = in_svm_data_file_template.format(w)
X, y = dataset_manager.read_svm_light_bool_data_to_sparse(data_file)
if override_tergets_function:
y = override_tergets_function(y)
for param in param_range:
model = compute_model(X, y, param, param_is_k)
fl.write("{0},\n".format(model))
fl.flush()
print model
yield model
if __name__ == '__main__':
# path = "/media/ivan/204C66C84C669874/Uni-Bonn/Thesis/Main/6_Results/svm/mutagenicity/"
# path = "/media/ivan/204C66C84C669874/Uni-Bonn/Thesis/Main/6_Results/svm/nci_hiv/"
path = "/home/stud/ivanovi/Thesis/svm/nci_hiv/"
# wl_props = helpers.svm_light_format_datasets["mutagenicity"]
# wl_props = helpers.svm_light_format_datasets["nci-hiv"]["A-vs-M"]
wl_props = helpers.svm_light_format_datasets["nci-hiv"]["AM-vs-I"]
# wl_props = helpers.svm_light_format_datasets["nci-hiv"]["A-vs-I"]
output_file = path + "positive_neighbors_AM_vs_I"
print "Start"
with open(output_file, "w") as fl:
for w in range(0, 12):
data_file = path + wl_props["file_template"].format(w)
X, y = dataset_manager.read_svm_light_bool_data_to_sparse(data_file)
# y = np.vectorize(lambda t: 1 if t == 2 else -1)(y) # Only for A_vs_M
y = np.vectorize(lambda t: 1 if t == 2 else t)(y) # Only for AM_vs_I and A_vs_I
for n in range(1, 500):
prediction = PositiveNeighbors.cross_validate(X, y, n_neighbors=n, folds_count=10, approximate=False)
print w, n, prediction
fl.write("{0}, {1}, {2}\n".format(w, n, prediction))
fl.flush()
if prediction >= 0.95:
break
print "Done"
# path = "/media/ivan/204C66C84C669874/Uni-Bonn/Thesis/Main/6_Results/svm/nci_hiv/"
path = "/home/stud/ivanovi/Thesis/svm/nci_hiv/"
# wl_props = helpers.svm_light_format_datasets["mutagenicity"]
# wl_props = helpers.svm_light_format_datasets["nci-hiv"]["A-vs-M"]
wl_props = helpers.svm_light_format_datasets["nci-hiv"]["AM-vs-I"]
# wl_props = helpers.svm_light_format_datasets["nci-hiv"]["A-vs-I"]
output_file = path + "positive_neighbors_AM_vs_I"
print "Start"
with open(output_file, "w") as fl:
for w in range(0, 12):
data_file = path + wl_props["file_template"].format(w)
X, y = dataset_manager.read_svm_light_bool_data_to_sparse(
data_file)
# y = np.vectorize(lambda t: 1 if t == 2 else -1)(y) # Only for A_vs_M
y = np.vectorize(lambda t: 1 if t == 2 else t)(
y) # Only for AM_vs_I and A_vs_I
for n in range(1, 500):
prediction = PositiveNeighbors.cross_validate(
X, y, n_neighbors=n, folds_count=10, approximate=False)
print w, n, prediction
fl.write("{0}, {1}, {2}\n".format(w, n, prediction))
fl.flush()
if prediction >= 0.95:
break
print "Done"
