diff --git a/examples/ifcb_experiments.py b/examples/ifcb_experiments.py
new file mode 100644
index 0000000..fff13ef
--- /dev/null
+++ b/examples/ifcb_experiments.py
@@ -0,0 +1,24 @@
+import quapy as qp
+from sklearn.linear_model import LogisticRegression
+from quapy.evaluation import evaluation_report
+
+
+def newLR():
+ return LogisticRegression(n_jobs=-1)
+
+quantifiers = {'CC':qp.method.aggregative.CC(newLR()),
+ 'ACC':qp.method.aggregative.ACC(newLR()),
+ 'PCC':qp.method.aggregative.PCC(newLR()),
+ 'PACC':qp.method.aggregative.PACC(newLR()),
+ 'HDy':qp.method.aggregative.DistributionMatching(newLR()),
+ 'EMQ':qp.method.aggregative.EMQ(newLR())}
+
+for quant_name, quantifier in quantifiers.items():
+ print("Experiment with "+quant_name)
+
+ train, test_gen = qp.datasets.fetch_IFCB()
+
+ quantifier.fit(train)
+
+ report = evaluation_report(quantifier, protocol=test_gen, error_metrics=['mae'], verbose=True)
+ print(report.mean())
diff --git a/quapy/data/_ifcb.py b/quapy/data/_ifcb.py
new file mode 100644
index 0000000..87bb030
--- /dev/null
+++ b/quapy/data/_ifcb.py
@@ -0,0 +1,34 @@
+import os
+import pandas as pd
+from quapy.protocol import AbstractProtocol
+
+class IFCBTrainSamplesFromDir(AbstractProtocol):
+
+ def __init__(self, path_dir:str, classes: list):
+ self.path_dir = path_dir
+ self.classes = classes
+ self.samples = []
+ for filename in os.listdir(path_dir):
+ if filename.endswith('.csv'):
+ self.samples.append(filename)
+
+ def __call__(self):
+ for sample in self.samples:
+ s = pd.read_csv(os.path.join(self.path_dir,sample))
+ # all columns but the first where we get the class
+ X = s.iloc[:, 1:].to_numpy()
+ y = s.iloc[:, 0].to_numpy()
+ yield X, y
+
+class IFCBTestSamples(AbstractProtocol):
+
+ def __init__(self, path_dir:str, test_prevalences_path: str):
+ self.path_dir = path_dir
+ self.test_prevalences = pd.read_csv(os.path.join(path_dir, test_prevalences_path))
+
+ def __call__(self):
+ for _, test_sample in self.test_prevalences.iterrows():
+ #Load the sample from disk
+ X = pd.read_csv(os.path.join(self.path_dir,test_sample['sample']+'.csv')).to_numpy()
+ prevalences = test_sample.iloc[1:].to_numpy().astype(float)
+ yield X, prevalences
\ No newline at end of file
diff --git a/quapy/data/datasets.py b/quapy/data/datasets.py
index b01dcec..2c82b0e 100644
--- a/quapy/data/datasets.py
+++ b/quapy/data/datasets.py
@@ -732,3 +732,82 @@ def fetch_lequa2022(task, data_home=None):
return train, val_gen, test_gen
+
+def fetch_IFCB(single_sample_train=True, data_home=None):
+ """
+ Loads the IFCB dataset for quantification <https://zenodo.org/records/10036244>`. For more
+ information on this dataset check the zenodo site.
+ This dataset is based on the data available publicly at <https://github.com/hsosik/WHOI-Plankton>.
+ The scripts for the processing are available at <https://github.com/pglez82/IFCB_Zenodo>
+
+ Basically, this is the IFCB dataset with precomputed features for testing quantification algorithms.
+
+ The datasets are downloaded only once, and stored for fast reuse.
+
+ :param single_sample_train: boolean. If True (default), it returns the train dataset as an instance of
+ :class:`quapy.data.base.LabelledCollection` (all examples together).
+ If False, a generator of training samples will be returned.
+ Each example in the training set has an individual class label.
+ :param data_home: specify the quapy home directory where collections will be dumped (leave empty to use the default
+ ~/quay_data/ directory)
+ :return: a tuple `(train, test_gen)` where `train` is an instance of
+ :class:`quapy.data.base.LabelledCollection`, if `single_sample_train` is True or
+ :class:`quapy.data._ifcb.IFCBTrainSamplesFromDir` otherwise, i.e. a sampling protocol that
+ returns a series of samples labelled example by example.
+ test_gen is an instance of :class:`quapy.data._ifcb.IFCBTestSamples`,
+ i.e., a sampling protocol that returns a series of samples labelled by prevalence.
+ """
+
+ from quapy.data._ifcb import IFCBTrainSamplesFromDir, IFCBTestSamples
+
+ if data_home is None:
+ data_home = get_quapy_home()
+
+ URL_TRAIN=f'https://zenodo.org/records/10036244/files/IFCB.train.zip'
+ URL_TEST=f'https://zenodo.org/records/10036244/files/IFCB.test.zip'
+ URL_TEST_PREV=f'https://zenodo.org/records/10036244/files/IFCB.test_prevalences.zip'
+
+ ifcb_dir = join(data_home, 'ifcb')
+ os.makedirs(ifcb_dir, exist_ok=True)
+
+ def download_unzip_and_remove(unzipped_path, url):
+ tmp_path = join(ifcb_dir, 'ifcb_tmp.zip')
+ download_file_if_not_exists(url, tmp_path)
+ with zipfile.ZipFile(tmp_path) as file:
+ file.extractall(unzipped_path)
+ os.remove(tmp_path)
+
+ if not os.path.exists(os.path.join(ifcb_dir,'train')):
+ download_unzip_and_remove(ifcb_dir, URL_TRAIN)
+ if not os.path.exists(os.path.join(ifcb_dir,'test')):
+ download_unzip_and_remove(ifcb_dir, URL_TEST)
+ if not os.path.exists(os.path.join(ifcb_dir,'test_prevalences.csv')):
+ download_unzip_and_remove(ifcb_dir, URL_TEST_PREV)
+
+ # Load test prevalences and classes
+ test_true_prev_path = join(ifcb_dir, 'test_prevalences.csv')
+ test_true_prev = pd.read_csv(test_true_prev_path)
+ classes = test_true_prev.columns[1:]
+
+ #Load train samples
+ train_samples_path = join(ifcb_dir,'train')
+ train_gen = IFCBTrainSamplesFromDir(path_dir=train_samples_path, classes=classes)
+
+ #Load test samples
+ test_samples_path = join(ifcb_dir,'test')
+ test_gen = IFCBTestSamples(path_dir=test_samples_path, test_prevalences_path=test_true_prev_path)
+
+ # In the case the user wants it, join all the train samples in one LabelledCollection
+ if single_sample_train:
+ X = []
+ y = []
+ for X_, y_ in train_gen():
+ X.append(X_)
+ y.append(y_)
+
+ X = np.vstack(X)
+ y = np.concatenate(y)
+ train = LabelledCollection(X,y, classes=classes)
+ return train, test_gen
+ else:
+ return train_gen, test_gen