def train_sblc(
idflabelspath,
algo,
outputDir="../working-directories/4-pre-trained-classifiers/",
savePickle=False,
):
"""Train the specified algorithm on the specified dataset. Save the
trained classifier into a Pickle object.
Parameters
----------
idflabelspath: str
Path to the dataset with identified labels
algo: str
Name of the supervised algorithm to use. Possible choices:
{RandomForestClassifier, KNeighborsClassifier, DecisionTreeClassifier,
AdaBoostClassifier, LabelSpreading}
For more details, refer to the documentation of scikit-learn 0.22
outputDir: str
Directory where will be stored the outputs
savePickle: bool, default=False
If False, the trained classifier is not saved
Returns
-------
Generate `algo.pkl` object in `outputDir`
clf: sklearn Classifier object
Trained classifier. To be used with `predict` method
Example
-------
>>> from blusc.supervisedfit import train_sblc
>>> inputDir = "../working-directories/3-identified-labels/"
>>> idfname = "IDFLABELS_2015_0219.PASSY2015_BT-T_linear_dz40_dt30_zmax2000.nc"
>>> clf = train_sblc(inputDir + idfname, algo ="KNeighborsClassifier")
Classifier not saved because savePickle= False
>>> clf.classes_
array([0, 1, 2], dtype=int32)
"""
# Load dataset
# ------------
X_raw, z_common, t_common, rawlabl, lablid, lablnames = utils.load_dataset(
idflabelspath,
variables_to_load=["X_raw", "altitude", "time", "rawlabels"],
fields_to_load=["label_identification", "label_long_names"],
)
# Normalization
# -------------
scaler = StandardScaler()
scaler.fit(X_raw)
X = scaler.transform(X_raw)
# Instantiate classifiers
# -----------------------
if algo in ["rf", "RandomForest", "RandomForestClassifier"]:
from sklearn.ensemble import RandomForestClassifier
clf = RandomForestClassifier(n_estimators=50, max_depth=3)
elif algo in ["knn", "nearestneighbors", "KNeighborsClassifier"]:
from sklearn.neighbors import KNeighborsClassifier
clf = KNeighborsClassifier(n_neighbors=6)
elif algo in ["dt", "DecisionTree", "DecisionTreeClassifier"]:
from sklearn.tree import DecisionTreeClassifier
clf = DecisionTreeClassifier(max_depth=5)
elif algo in ["ab", "adab", "AdaBoost", "AdaBoostClassifier"]:
from sklearn.ensemble import AdaBoostClassifier
from sklearn.tree import DecisionTreeClassifier
clf = AdaBoostClassifier(
base_estimator=DecisionTreeClassifier(max_depth=4), n_estimators=50
)
elif algo in ["ls", "LabelSpreading"]:
from sklearn.semi_supervised import LabelSpreading
clf = LabelSpreading(kernel="knn", alpha=0.2)
else:
raise ValueError("Not supported algorithm:", algo)
# Fit supervised model
# ---------------------
clf.fit(X, rawlabl)
# Exports
# -----------
clf.label_identification_ = lablid
clf.label_long_names_ = lablnames
clf.scaler = scaler
n_classes = clf.classes_.size
centroids = np.zeros((n_classes, X.shape[1]))
for k in range(n_classes):
idx = np.where(rawlabl==k)[0]
centroids[k,:] = np.mean(X[idx,:],axis=0)
clf.training_class_centroids_ = centroids
idflabelsname = idflabelspath.split("/")[-1]
prefx, prepkey, dotnc = idflabelsname.split(".")
dropfilename = str(clf).split("(")[0] + "." + prepkey + ".pkl"
if savePickle:
fc = open(outputDir + dropfilename, "wb")
pickle.dump(clf, fc)
fc.close()
print("Trained classifier saved in ", outputDir + dropfilename)
else:
print("Classifier not saved because savePickle=", savePickle)
return clf
