Renamed package "italia" to "ital"

italia/adapt_al.py → ital/adapt_al.py

@@ -55,11 +55,11 @@ def __init__(self, data = None, queries = [], length_scale = 0.1, var = 1.0, noi
         
         - queries: list of query data points to initially fit the GP to. If empty, the GP will not be fitted.
         
-        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - var: the `var` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - var: the `var` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
         - subsample: if set to a positive integer, the set of candidates will be restricted to a random subsample
                      of unlabeled instance of the given size.

italia/ital.py → ital/ital.py

@@ -24,11 +24,11 @@ def __init__(self, data = None, queries = [], length_scale = 0.1, var = 1.0, noi
         
         - queries: list of query data points to initially fit the GP to. If empty, the GP will not be fitted.
         
-        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - var: the `var` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - var: the `var` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
         - label_prob: the assumed probability that the user annotates a sample.
         

italia/ital_regression.py → ital/ital_regression.py

@@ -25,11 +25,11 @@ def __init__(self, data = None, train_init = [], y_init = [], length_scale = 0.1
         
         - y_init: initial training values for the samples in train_init.
         
-        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - var: the `var` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - var: the `var` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
         - label_prob: the assumed probability that the user annotates a sample.
         
@@ -96,7 +96,8 @@ def fetch_unlabelled(self, k, show_progress = False):
 class MutualInformation(object):
     """ Helper class for ITAL_KL computing mutual information between the value distribution and the feedback distribution of a set of samples.
     
-    \mathrm{MI}(R,F\ |\ A) = \int_{-\infty}^{\infty} { p(F=f\ |\ A) \cdot \mathrm{KL}\Bigl(p(R=r\ |\ F=f,A) \Bigm\| p(R=r\ |\ A)\Bigr) }
+    .. math::
+        \mathrm{MI}(R,F\ |\ A) = \int_{-\infty}^{\infty} { p(F=f\ |\ A) \cdot \mathrm{KL}\Bigl(p(R=r\ |\ F=f,A) \Bigm\| p(R=r\ |\ A)\Bigr) }
     """
 
     def __init__(self, learner, eps = 1e-6):

italia/mcmi.py → ital/mcmi.py

@@ -27,11 +27,11 @@ def __init__(self, data = None, queries = [], length_scale = 0.1, var = 1.0, noi
         
         - queries: list of query data points to initially fit the GP to. If empty, the GP will not be fitted.
         
-        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - var: the `var` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - var: the `var` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
         - subsample: if set to a positive integer, the set of candidates will be restricted to a random subsample
                      of unlabeled instance of the given size.

italia/regression_base.py → ital/regression_base.py

@@ -20,11 +20,11 @@ def __init__(self, data = None, train_init = [], y_init = [], length_scale = 0.1
         
         - y_init: initial training values for the samples in train_init.
         
-        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - var: the `var` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - var: the `var` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         """
 
         self.length_scale = length_scale
@@ -144,7 +144,7 @@ def updated_prediction(self, feedback, test_ind, cov_mode = 'full'):
 
         ind, val, _ = self.labeled_feedback(feedback)
         if len(ind) == 0:
-            return self.predict_stored(test_ind, cov_mode = cov_mode)
+            return self.gp.predict_stored(test_ind, cov_mode = cov_mode)
         else:
             ind.sort()
             return self.gp.updated_prediction(ind, val, test_ind, cov_mode = cov_mode)

italia/retrieval_base.py → ital/retrieval_base.py

@@ -18,11 +18,11 @@ def __init__(self, data = None, queries = [], length_scale = 0.1, var = 1.0, noi
         
         - queries: list of query data points to initially fit the GP to. If empty, the GP will not be fitted.
         
-        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - length_scale: the `sigma` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - var: the `var` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - var: the `var` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         
-        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of italia.gp.GaussianProcess).
+        - noise: the `sigma_noise` hyper-parameter of the kernel (see documentation of ital.gp.GaussianProcess).
         """
 
         self.length_scale = length_scale
@@ -152,7 +152,7 @@ def updated_prediction(self, feedback, test_ind, cov_mode = 'full'):
 
         rel, irr, _ = self.partition_feedback(feedback)
         if len(rel) + len(irr) == 0:
-            return self.predict_stored(test_ind, cov_mode=cov_mode)
+            return self.gp.predict_stored(test_ind, cov_mode=cov_mode)
         else:
             rel.sort()
             irr.sort()

optimize_parameters.py

@@ -9,7 +9,7 @@
 
 import utils
 from datasets import RegressionDataset, MultitaskRetrievalDataset
-from italia.gp import GaussianProcess
+from ital.gp import GaussianProcess
 
 
 

run_experiment.py

@@ -9,7 +9,7 @@
 
 import utils, viz_utils
 from datasets import MultitaskRetrievalDataset
-from italia.regression_base import ActiveRegressionBase
+from ital.regression_base import ActiveRegressionBase
 
 
 

utils.py

@@ -4,7 +4,7 @@
 
 import numpy as np
 
-from italia import *
+from ital import *
 from datasets import load_dataset, RetrievalDataset, MultitaskRetrievalDataset, RegressionDataset
 
 
@@ -92,8 +92,8 @@ def load_config(config_file, section, overrides = {}):
         a (parser, dataset, learner) tuple whose individual components are:
         - parser: a configparser.ConfigParser instance,
         - dataset: a dataset.Dataset instance,
-        - learner: either an italia.retrieval_base.ActiveRetrievalBase instance, an
-                   italia.regression_base.ActiveRegressionBase instance.
+        - learner: either an ital.retrieval_base.ActiveRetrievalBase instance, an
+                   ital.regression_base.ActiveRegressionBase instance.
                    The learner is usually initialized with the data from the dataset,
                    except in the case of a MultitaskDataset, where an uninitialized
                    learner will be returned.

viz_utils.py

@@ -124,7 +124,7 @@ def plot_learning_step(dataset, queries, relevance, learner, ret, fb):
     
     - relevance: the ground-truth relevance labels of all samples in dataset.X_train.
     
-    - learner: an italia.retrieval_base.ActiveRetrievalBase instance.
+    - learner: an ital.retrieval_base.ActiveRetrievalBase instance.
     
     - ret: the indices of the samples selected for the current active learning batch.
     
@@ -193,7 +193,7 @@ def plot_regression_step(dataset, init, learner, ret, fb):
     
     - init: list of indices of the initial training samples in dataset.X_train.
     
-    - learner: an italia.regression_base.ActiveRegressionBase instance.
+    - learner: an ital.regression_base.ActiveRegressionBase instance.
     
     - ret: the indices of the samples selected for the current active learning batch.