Autocorrelation

metalearn/metafeatures/metafeatures.py

@@ -17,6 +17,7 @@
 from metalearn.metafeatures.decision_tree_metafeatures import resources_info as dt_resources
 from metalearn.metafeatures.general_resource_computers import resources_info as general_resources
 from metalearn.metafeatures.text_metafeatures import resources_info as text_resources
+from metalearn.metafeatures.simple_metafeatures import resources_info as simple_resources
 
 from metalearn.metafeatures.decision_tree_metafeatures import metafeatures_info as dt_metafeatures
 from metalearn.metafeatures.information_theoretic_metafeatures import metafeatures_info as info_theoretic_metafeatures
@@ -38,6 +39,7 @@ class Metafeatures(object):
     _resources_info.update(dt_resources)
     _resources_info.update(general_resources)
     _resources_info.update(text_resources)
+    _resources_info.update(simple_resources)
 
     # noop resource computers for the user-provided resources
     # `_get_arguments` and `_resource_is_target_dependent` assumes ResourceComputer's

metalearn/metafeatures/simple_metafeatures.py

@@ -1,19 +1,27 @@
+from collections import Counter
+
 import numpy as np
 from pandas import DataFrame
 
 from metalearn.metafeatures.common_operations import *
-from metalearn.metafeatures.base import build_resources_info, MetafeatureComputer
+from metalearn.metafeatures.base import build_resources_info, MetafeatureComputer, ResourceComputer
 from metalearn.metafeatures.constants import ProblemType, MetafeatureGroup
 
 
-def get_dataset_stats(X, column_types):
+def get_dataset_stats(X, column_types, num_binary_numeric, num_binary_categorical):
     number_of_instances = X.shape[0]
     number_of_features = X.shape[1]
     numeric_features = len(get_numeric_features(X, column_types))
     categorical_features = number_of_features - numeric_features
+    ratio_of_binary_numeric_features = num_binary_numeric / number_of_features
+    ratio_of_binary_categorical_features = num_binary_categorical / number_of_features
     ratio_of_numeric_features = numeric_features / number_of_features
     ratio_of_categorical_features = categorical_features / number_of_features
-    return (number_of_instances, number_of_features, numeric_features, categorical_features, ratio_of_numeric_features, ratio_of_categorical_features)
+    return (
+        number_of_instances, number_of_features, numeric_features, categorical_features,
+        ratio_of_numeric_features, ratio_of_categorical_features, ratio_of_binary_numeric_features,
+        ratio_of_binary_categorical_features
+    )
 
 get_dataset_stats = MetafeatureComputer(
     get_dataset_stats,
@@ -23,11 +31,17 @@ def get_dataset_stats(X, column_types):
         "NumberOfNumericFeatures",
         "NumberOfCategoricalFeatures",
         "RatioOfNumericFeatures",
-        "RatioOfCategoricalFeatures"
+        "RatioOfCategoricalFeatures",
+        "RatioOfBinaryNumericFeatures",
+        "RatioOfBinaryCategoricalFeatures"
     ],
     ProblemType.ANY,
     [MetafeatureGroup.SIMPLE],
-    { "X": "X_raw" }
+    {
+        "X": "X_raw",
+        "num_binary_numeric": "NumberOfBinaryNumericFeatures",
+        "num_binary_categorical": "NumberOfBinaryCategoricalFeatures"
+    }
 )
 
 
@@ -88,8 +102,8 @@ def get_class_stats(Y):
     return (number_of_classes, *profile_distribution(probs), minority_class_size, majority_class_size)
 
 get_class_stats = MetafeatureComputer(
-    computer=get_class_stats,
-    returns=[
+    get_class_stats,
+    [
         "NumberOfClasses",
         "MeanClassProbability",
         "StdevClassProbability",
@@ -103,17 +117,55 @@ def get_class_stats(Y):
         "MinorityClassSize",
         "MajorityClassSize"
     ],
-    problem_type=ProblemType.CLASSIFICATION,
-    groups=[MetafeatureGroup.SIMPLE]
+    ProblemType.CLASSIFICATION,
+    [MetafeatureGroup.SIMPLE]
+)
+
+
+def get_categorical_cardinalities_at_values(CategoricalCardinalities):
+    counts = Counter(CategoricalCardinalities)
+    return counts.get(2, 0), counts.get(3, 0), counts.get(4, 0)
+
+get_categorical_cardinalities_at_values = MetafeatureComputer(
+    get_categorical_cardinalities_at_values,
+    [
+        "NumberOfBinaryCategoricalFeatures",
+        "CategoricalCardinalityAtThree",
+        "CategoricalCardinalityAtFour"
+    ],
+    ProblemType.ANY,
+    [MetafeatureGroup.SIMPLE]
+)
+
+
+def get_numeric_cardinalities_at_values(NumericCardinalities):
+    counts = Counter(NumericCardinalities)
+    return counts.get(2, 0), counts.get(3, 0), counts.get(4, 0)
+
+get_numeric_cardinalities_at_values = MetafeatureComputer(
+    get_numeric_cardinalities_at_values,
+    [
+        "NumberOfBinaryNumericFeatures",
+        "NumericCardinalityAtThree",
+        "NumericCardinalityAtFour"
+    ],
+    ProblemType.ANY,
+    [MetafeatureGroup.SIMPLE]
 )
 
 
 def get_categorical_cardinalities(X, column_types):
-    cardinalities = [X[feature].unique().shape[0] for feature in get_categorical_features(X, column_types)]
-    return profile_distribution(cardinalities)
+    cat_cards = [X[feature].unique().shape[0] for feature in get_categorical_features(X, column_types)]
+    return (cat_cards,)
 
-get_categorical_cardinalities = MetafeatureComputer(
+get_categorical_cardinalities = ResourceComputer(
     get_categorical_cardinalities,
+    ["CategoricalCardinalities"]
+)
+
+
+profile_categorical_cardinalities = MetafeatureComputer(
+    profile_distribution,
     [
         "MeanCardinalityOfCategoricalFeatures",
         "StdevCardinalityOfCategoricalFeatures",
@@ -127,15 +179,21 @@ def get_categorical_cardinalities(X, column_types):
     ],
     ProblemType.ANY,
     [MetafeatureGroup.SIMPLE],
+    { "data": "CategoricalCardinalities" }
 )
 
 
 def get_numeric_cardinalities(X, column_types):
-    cardinalities = [X[feature].unique().shape[0] for feature in get_numeric_features(X, column_types)]
-    return profile_distribution(cardinalities)
+    num_cards = [X[feature].unique().shape[0] for feature in get_numeric_features(X, column_types)]
+    return (num_cards,)
 
-get_numeric_cardinalities = MetafeatureComputer(
+get_numeric_cardinalities = ResourceComputer(
     get_numeric_cardinalities,
+    ["NumericCardinalities"]
+)
+
+profile_numeric_cardinalities = MetafeatureComputer(
+    profile_distribution,
     [
         "MeanCardinalityOfNumericFeatures",
         "StdevCardinalityOfNumericFeatures",
@@ -148,9 +206,18 @@ def get_numeric_cardinalities(X, column_types):
         "MaxCardinalityOfNumericFeatures"
     ],
     ProblemType.ANY,
-    [MetafeatureGroup.SIMPLE]
+    [MetafeatureGroup.SIMPLE],
+    { "data": "NumericCardinalities" }
 )
 
+"""
+A list of all ResourceComputer
+instances in this module.
+"""
+resources_info = build_resources_info(
+    get_categorical_cardinalities,
+    get_numeric_cardinalities
+)
 
 """
 A list of all MetafeatureComputer
@@ -159,8 +226,10 @@ def get_numeric_cardinalities(X, column_types):
 metafeatures_info = build_resources_info(
     get_dataset_stats,
     get_class_stats,
+    get_categorical_cardinalities_at_values,
+    get_numeric_cardinalities_at_values,
     get_dimensionality,
     get_missing_values,
-    get_categorical_cardinalities,
-    get_numeric_cardinalities
+    profile_categorical_cardinalities,
+    profile_numeric_cardinalities
 )

metalearn/metafeatures/statistical_metafeatures.py

@@ -14,6 +14,7 @@
 import metalearn.metafeatures.constants as consts
 
 
+
 def get_numeric_means(numeric_features_array):
     means = [feature.mean() for feature in numeric_features_array]
     return profile_distribution(means)
@@ -213,6 +214,25 @@ def preprocess(series):
 
     return correlations
 
+def autocorrelation(Y, column_types):
+    if Y.size == 0:
+        ac = None
+    elif Y.size == 1:
+        ac = 1.0
+    elif column_types[Y.name] == consts.NUMERIC:
+        ac = Y.autocorr()
+    elif column_types[Y.name] == consts.CATEGORICAL:
+        ac = np.equal(Y[:-1].values, Y[1:].values).mean()
+    else:
+        ac = None
+    return (ac,)
+
+autocorrelation = MetafeatureComputer(
+    autocorrelation,
+    ["Autocorrelation"],
+    ProblemType.ANY,
+    [MetafeatureGroup.STATISTICAL]
+)
 
 """
 A list of all MetafeatureComputer
@@ -223,5 +243,6 @@ def preprocess(series):
     get_numeric_stdev,
     get_numeric_skewness,
     get_numeric_kurtosis,
-    get_pca
+    get_pca,
+    autocorrelation
 )

tests/data/dataset_metafeatures/38_sick_train_data_mf.json

@@ -1,4 +1,13 @@
 {
+    "Autocorrelation": {
+        "value": 0.8865022540440202
+    },
+    "CategoricalCardinalityAtFour": {
+        "value": 0
+    },
+    "CategoricalCardinalityAtThree": {
+        "value": 1
+    },
     "CategoricalNoiseToSignalRatio": {
         "value": 65.12375165878576
     },
@@ -290,6 +299,12 @@
     "NaiveBayesKappa": {
         "value": 0.04421998456446913
     },
+    "NumberOfBinaryCategoricalFeatures": {
+        "value": 19
+    },
+    "NumberOfBinaryNumericFeatures": {
+        "value": 0
+    },
     "NumberOfCategoricalFeatures": {
         "value": 22
     },
@@ -323,6 +338,12 @@
     "NumberOfTokensContainingNumericChar": {
         "value": 0
     },
+    "NumericCardinalityAtFour": {
+        "value": 0
+    },
+    "NumericCardinalityAtThree": {
+        "value": 0
+    },
     "NumericNoiseToSignalRatio": {
         "value": 43.499206497970064
     },
@@ -545,6 +566,12 @@
     "RandomTreeDepth3Kappa": {
         "value": 0.16934246039579826
     },
+    "RatioOfBinaryCategoricalFeatures": {
+        "value": 0.6551724137931034
+    },
+    "RatioOfBinaryNumericFeatures": {
+        "value": 0.0
+    },
     "RatioOfCategoricalFeatures": {
         "value": 0.7586206896551724
     },
@@ -692,4 +719,4 @@
     "kNN1NKappa": {
         "value": 0.18924686295501464
     }
-}
+}

tests/data/dataset_metafeatures/small_test_dataset_mf.json

@@ -1,4 +1,13 @@
 {
+    "Autocorrelation": {
+        "value": 0.6666666666666666
+    },
+    "CategoricalCardinalityAtFour": {
+        "value": 0
+    },
+    "CategoricalCardinalityAtThree": {
+        "value": 0
+    },
     "CategoricalNoiseToSignalRatio": {
         "value": 1.0834895643494507
     },
@@ -290,6 +299,12 @@
     "NaiveBayesKappa": {
         "value": 0.1875
     },
+    "NumberOfBinaryCategoricalFeatures": {
+        "value": 1
+    },
+    "NumberOfBinaryNumericFeatures": {
+        "value": 0
+    },
     "NumberOfCategoricalFeatures": {
         "value": 2
     },
@@ -323,6 +338,12 @@
     "NumberOfTokensContainingNumericChar": {
         "value": 0
     },
+    "NumericCardinalityAtFour": {
+        "value": 0
+    },
+    "NumericCardinalityAtThree": {
+        "value": 0
+    },
     "NumericNoiseToSignalRatio": {
         "value": 0.8185194089132799
     },
@@ -545,6 +566,12 @@
     "RandomTreeDepth3Kappa": {
         "value": 0.11764705882352944
     },
+    "RatioOfBinaryCategoricalFeatures": {
+        "value": 0.2
+    },
+    "RatioOfBinaryNumericFeatures": {
+        "value": 0.0
+    },
     "RatioOfCategoricalFeatures": {
         "value": 0.4
     },
@@ -692,4 +719,4 @@
     "kNN1NKappa": {
         "value": 0.20833333333333337
     }
-}
+}