def __init__(self,
n_estimators=10, # 30
max_depth=0,
max_leaf_nodes=0,
max_features=0,
min_samples_leaf=1, # 5
# random_state=123456,
# feature_type="float",
# n_jobs=1,
model_file=None,
verbose=False
):
if model_file is None:
model_file = tempfile().name
self.model_file = model_file
self.fte_per_split = max_features
self.verbose = verbose
self.fit_argv = [WiseRFRegressor.wiserf_src]
self.fit_argv += ["learn-regress"]
self.fit_argv += ["--model-file", self.model_file]
self.fit_argv += ["--criteria", "variance"]
self.fit_argv += ["--num-trees", n_estimators]
self.fit_argv += ["--max-depth", max_depth]
self.fit_argv += ["--max-leaf-nodes", max_leaf_nodes]
self.fit_argv += ["--min-instances-per-node", min_samples_leaf]
self.pred_argv = [WiseRFRegressor.wiserf_src]
self.pred_argv += ["test-regress"]
self.pred_argv += ["--model-file", self.model_file]
def _make_wiserf_csvs(X, Y):
infile = tempfile().name
descfile = tempfile().name
tmpX = DataFrame(X).add_prefix("x")
if Y is not None:
tmpY = DataFrame(Y).add_prefix("y")
tmpX = tmpX.join(tmpY, how='outer')
columns = tmpX.columns
tmpX.to_csv(infile, header=False, index=False)
with open(descfile, 'w') as outfile:
outfile.write("label-type regression\n")
outfile.write("num-features %d\n" % WiseRFRegressor._max_fte(X))
for indx, column in enumerate(columns):
if column.startswith('y'):
outfile.write("class-column %d\n" % indx)
return [infile, descfile]
def _default(self, model, fit=False):
if fit:
self._modelfile = tempfile().name
model = model or self._modelfile
assert model is not None
argv = map(str, self._argv)
argv += ["--dimensionality", self._dim]
argv += [("--test_file", "--training_file")[fit], self._data]
argv += [("--model_in", "--model_out")[fit], model]
if not fit:
outfile = tempfile().name
argv += ["--cluster_mapping_out", outfile]
call(argv, verbose=self.verbose)
X, y = read_svmlight(outfile)
return X.todense()
else:
call(argv, verbose=self.verbose)
self._modelfile = model
def __init__(self,
learner_type="pegasos",
loop_type="stochastic",
eta_type="pegasos",
prediction_type="linear",
bias_term=True,
iterations=100000,
lambda_=0.1,
passive_aggressive_c=None,
passive_aggressive_lambda=0,
perceptron_margin_size=1.0,
rank_step_probability=0.5,
hash_mask_bits=0,
random_state=0,
model_file=None,
verbose=False
):
assert learner_type in ["pegasos", "sgd-svm", "passive-aggressive", "margin-perceptron", "romma", "logreg-pegasos", "logreg", "least-mean-squares"]
assert loop_type in ["stochastic", "balanced-stochastic", "rank", "roc", "query-norm-rank", "combined-ranking", "combined-roc"]
assert eta_type in ["basic", "pegasos", "constant"]
assert prediction_type in ["linear", "logistic"]
if model_file is None:
model_file = tempfile().name
self.verbose = verbose
self.fit_argv = [SofiaML.sofia_src]
self.fit_argv += ["--learner_type", learner_type]
self.fit_argv += ["--loop_type", loop_type]
self.fit_argv += ["--eta_type", eta_type]
self.fit_argv += ["--iterations", iterations]
self.fit_argv += ["--prediction_type", prediction_type]
if learner_type in ["pegasos", "sgd-svm", "logreg-pegasos"]:
self.fit_argv += ["--lambda", lambda_]
elif learner_type == "passive-aggressive":
if passive_aggressive_c is not None:
self.fit_argv += ["--passive_aggressive_c", passive_aggressive_c]
self.fit_argv += ["--passive_aggressive_lambda", passive_aggressive_lambda]
elif learner_type == "margin-perceptron":
self.fit_argv += ["--perceptron-margin-size", perceptron_margin_size]
if loop_type in ["combined-ranking", "combined-roc"]:
self.fit_argv += ["--rank_step_probability", rank_step_probability]
if not bias_term:
self.fit_argv += ["--no_bias_term"]
self.fit_argv += ["--hash_mask_bits", hash_mask_bits]
self.fit_argv += ["--random_seed", random_state]
self.fit_argv += ["--model_out", model_file]
self.pred_argv = [SofiaML.sofia_src]
if not bias_term:
self.pred_argv += ["--no_bias_term"]
self.pred_argv += ["--hash_mask_bits", hash_mask_bits]
self.pred_argv += ["--random_seed", random_state]
self.pred_argv += ["--model_in", model_file]
def predict(self, X, Y=None):
"""Returns model prediction of data points.
Parameters
----------
X: numpy array, shape = [n_samples, n_features]
Testing vectors, where n_samples is the number of samples and n_features is the number of features.
Y: Not used for computation, optional
"""
argv = map(str, self.pred_argv)
pred_file = tempfile().name
argv += ["--predictions-file", pred_file]
self._default(argv, X)
tmpY = read_csv(pred_file, header=None, index_col=False, sep=" ")
tmpY = tmpY.as_matrix()[0, :-1]
print X.shape
print tmpY.shape
return tmpY
def predict(self, X, Y=None):
"""Returns model prediction of data points.
Parameters
----------
X: numpy array, shape = [n_samples, n_features]
Testing vectors, where n_samples is the number of samples and n_features is the number of features.
Y: Not used for computation, optional
"""
assert self._dim == X.shape[1]
data = make_svmlight(X, Y).name
results_file = tempfile().name
argv = map(str, self.pred_argv)
argv += ["--test_file", data]
argv += ["--results_file", results_file]
self._default(argv)
Y = read_csv(results_file, sep='\t', header=None, index_col=False)
return Y.as_matrix()[:, 0]