First commit

Author: mhmotallebi

.gitignore

@@ -0,0 +1,5 @@
+__pycache__/
+ouput_bfs*
+*.zip
+.idea/
+backup_stuff/

README.md

@@ -0,0 +1,28 @@
+# SigDirect
+This is a python implementation of SigDirect [1] classifier. 
+The classifier does not have any specific hyper-parameter to tune and 
+the only one is the p-value of the statistical significance, 
+which is set to 0.05 (you can change it in the config file, though this is the value used in most scientific works)
+
+## Running
+You first need to use the requirements.txt file to install the dependencies:
+```
+pip3 install -r requirements.txt
+```
+
+You can test the dataset using one of the provided UCI datasets:
+```
+python3 sigdirect_test.py iris
+```
+In order to use the classifier, you can call the code similar to scikit-learn classifiers:
+You should instantiate it first, and then call the ```fit``` method to train a model. 
+To get predictions, you can either call ```predict``` or ```predict_proba``` methods where 
+the former will provide classes and the latter will provide the probability distributions. 
+
+The input to fit method should be a 2-d numpy array where each feature can be either 0 or 1. 
+For labels, a 1-d array should be used where each element should also be integers (starting from 0 to n-1 for n classes.)
+
+Note: You can provide ```get_logs=std.out``` or ```get_logs=std.err``` as an argument to the constructor
+of the classifier so it will print some logs about creating the model. 
+
+[1]: Li, Jundong, and Osmar R. Zaiane. "Exploiting statistically significant dependent rules for associative classification." Intelligent Data Analysis 21.5 (2017): 1155-1172.

__init__.py

@@ -0,0 +1,10 @@
+random_seed = 1
+
+ALPHA = 0.05
+ALPHA_LOG = -3.0000
+
+DEFAULT_LAYER_SIZE = 1
+
+BASE_VALUE_THRESHOLD = 1000000
+
+BATCH_SIZE = 1000

config.py

@@ -0,0 +1,5 @@
+numpy==1.17.4
+pandas==0.25.3
+psutil==5.6.7
+scikit-learn==0.22
+scipy==1.3.2

requirements.txt

@@ -0,0 +1,49 @@
+#!/usr/bin/env python3
+
+import math
+
+
+class Rule:
+    """ Represents a final Rule that is extracted from the tree. """
+
+    def __init__(self, items, label, confidence, ss, support):
+        self._items = items
+        self._label = label
+        self._confidence = confidence
+        self._ss = ss
+        self._support = support
+
+    def get_items(self):
+        return self._items
+
+    def get_label(self):
+        return self._label
+
+    def get_confidence(self):
+        return self._confidence
+
+    def get_ss(self):
+        return self._ss
+
+    def get_support(self):
+        return self._support
+
+    def set_items(self, new_items):
+        self._items = new_items
+    
+    def __str__(self):
+        try:
+            return "{} {};{:.4f},{:.3f},{:.3f}".format(' '.join(map(str, self._items)),
+                                          self._label,
+                                          self._support,
+                                          self._confidence,
+                                          math.log(self._ss),
+                                          )
+        except Exception as e:
+            print(repr(e), self._items, self._label, self._ss)
+            return "{} {};{:.4f},{:.3f},{:.3f}".format(' '.join(map(str, self._items)),
+                                          self._label,
+                                          float(self._support),
+                                          float(self._confidence),
+                                          0.0,
+                                          )

rule.py

@@ -0,0 +1,15 @@
+#!/bin/bash
+
+INDEX=80
+INTERPRETER=python3
+declare -a NAMES=(  "breast" "flare" "glass" "heart" "iris" 
+					"led7" "anneal" "pageBlocks" "pima" "wine" 
+					"zoo" "hepati" "horse" "adult" "mushroom" 
+					"penDigits" "letRecog" "soybean" "ionosphere" "cylBands")
+
+for NAME in ${NAMES[@]};
+do
+	echo $NAME
+	INTERPRETER sigdirect_test.py $NAME  >> output_bfs_$INDEX;
+	echo "done"
+done