[ENH] add zoo example, resolve comments

examples/hed/README.md

@@ -0,0 +1,39 @@
+# Handwritten Equation Decipherment
+
+This notebook shows an implementation of [Handwritten Equation Decipherment](https://proceedings.neurips.cc/paper_files/paper/2019/file/9c19a2aa1d84e04b0bd4bc888792bd1e-Paper.pdf). In this task, the handwritten equations are given, which consist of sequential pictures of characters. The equations are generated with unknown operation rules from images of symbols ('0', '1', '+' and '='), and each equation is associated with a label indicating whether the equation is correct (i.e., positive) or not (i.e., negative). Also, we are given a knowledge base which involves the structure of the equations and a recursive definition of bit-wise operations. The task is to learn from a training set of above mentioned equations and then to predict labels of unseen equations. 
+
+## Run
+
+```bash
+pip install -r requirements.txt
+python main.py
+```
+
+## Usage
+
+```bash
+usage: main.py [-h] [--no-cuda] [--epochs EPOCHS] [--lr LR]
+               [--weight-decay WEIGHT_DECAY] [--batch-size BATCH_SIZE]
+               [--loops LOOPS] [--segment_size SEGMENT_SIZE]
+               [--save_interval SAVE_INTERVAL] [--max-revision MAX_REVISION]
+               [--require-more-revision REQUIRE_MORE_REVISION]
+               [--ground] [--max-err MAX_ERR]
+
+Handwritten Equation Decipherment example
+
+optional arguments:
+  -h, --help            show this help message and exit
+  --no-cuda             disables CUDA training
+  --epochs EPOCHS       number of epochs in each learning loop iteration
+                        (default : 1)
+  --lr LR               base model learning rate (default : 0.001)
+  --weight-decay WEIGHT_DECAY
+                        weight decay (default : 0.0001)
+  --batch-size BATCH_SIZE
+                        base model batch size (default : 32)
+  --save_interval SAVE_INTERVAL
+                        save interval (default : 1)
+  --max-revision MAX_REVISION
+                        maximum revision in reasoner (default : 10)
+
+```

examples/hed/datasets/equation_generator.py

@@ -1,173 +0,0 @@
-import os
-import itertools
-import random
-import numpy as np
-from PIL import Image
-import pickle
-
-def get_sign_path_list(data_dir, sign_names):
-    sign_num = len(sign_names)
-    index_dict = dict(zip(sign_names, list(range(sign_num))))
-    ret = [[] for _ in range(sign_num)]
-    for path in os.listdir(data_dir):
-        if (path in sign_names):
-            index = index_dict[path]
-            sign_path = os.path.join(data_dir, path)
-            for p in os.listdir(sign_path):
-                ret[index].append(os.path.join(sign_path, p))
-    return ret
-
-def split_pool_by_rate(pools, rate, seed = None):
-    if seed is not None:
-        random.seed(seed)
-    ret1 = []
-    ret2 = []
-    for pool in pools:
-        random.shuffle(pool)
-        num = int(len(pool) * rate)
-        ret1.append(pool[:num])
-        ret2.append(pool[num:])
-    return ret1, ret2
-
-def int_to_system_form(num, system_num):
-    if num == 0:
-        return "0"
-    ret = ""
-    while (num > 0):
-        ret += str(num % system_num)
-        num //= system_num
-    return ret[::-1]
-
-def generator_equations(left_opt_len, right_opt_len, res_opt_len, system_num, label, generate_type):
-    expr_len = left_opt_len + right_opt_len
-    num_list = "".join([str(i) for i in range(system_num)])
-    ret = []
-    if generate_type == "all":
-        candidates = itertools.product(num_list, repeat = expr_len)
-    else:
-        candidates = [''.join(random.sample(['0', '1'] * expr_len, expr_len))]
-        random.shuffle(candidates)
-    for nums in candidates:
-        left_num = "".join(nums[:left_opt_len])
-        right_num = "".join(nums[left_opt_len:])
-        left_value = int(left_num, system_num)
-        right_value = int(right_num, system_num)
-        result_value = left_value + right_value
-        if (label == 'negative'):
-            result_value += random.randint(-result_value, result_value)
-            if (left_value + right_value == result_value):
-                continue
-        result_num = int_to_system_form(result_value, system_num)
-        #leading zeros
-        if (res_opt_len != len(result_num)):
-            continue
-        if ((left_opt_len > 1 and left_num[0] == '0') or (right_opt_len > 1 and right_num[0] == '0')):
-            continue
-
-        #add leading zeros
-        if (res_opt_len < len(result_num)):
-            continue
-        while (len(result_num) < res_opt_len):
-            result_num = '0' + result_num
-            #continue
-        ret.append(left_num + '+' + right_num + '=' + result_num) # current only consider '+' and '='
-        #print(ret[-1])
-    return ret
-
-def generator_equation_by_len(equation_len, system_num = 2, label = 0, require_num = 1):
-    generate_type = "one"
-    ret = []
-    equation_sign_num = 2 # '+' and '='
-    while len(ret) < require_num:
-        left_opt_len = random.randint(1, equation_len - 1 - equation_sign_num)
-        right_opt_len = random.randint(1, equation_len - left_opt_len - equation_sign_num)
-        res_opt_len = equation_len - left_opt_len - right_opt_len - equation_sign_num
-        ret.extend(generator_equations(left_opt_len, right_opt_len, res_opt_len, system_num, label, generate_type))
-    return ret
-
-def generator_equations_by_len(equation_len, system_num = 2, label = 0, repeat_times = 1, keep = 1, generate_type = "all"):
-    ret = []
-    equation_sign_num = 2 # '+' and '='
-    for left_opt_len in range(1, equation_len - (2 + equation_sign_num) + 1):
-        for right_opt_len in range(1, equation_len - left_opt_len - (1 + equation_sign_num) + 1):
-            res_opt_len = equation_len - left_opt_len - right_opt_len - equation_sign_num
-            for i in range(repeat_times): #generate more equations
-                if random.random() > keep ** (equation_len):
-                    continue
-                ret.extend(generator_equations(left_opt_len, right_opt_len, res_opt_len, system_num, label, generate_type))
-    return ret
-
-def generator_equations_by_max_len(max_equation_len, system_num = 2, label = 0, repeat_times = 1, keep = 1, generate_type = "all", num_per_len = None):
-    ret = []
-    equation_sign_num = 2 # '+' and '='
-    for equation_len in range(3 + equation_sign_num, max_equation_len + 1):
-        if (num_per_len is None):
-            ret.extend(generator_equations_by_len(equation_len, system_num, label, repeat_times, keep, generate_type))
-        else:
-            ret.extend(generator_equation_by_len(equation_len, system_num, label, require_num = num_per_len))
-    return ret
-
-def generator_equation_images(image_pools, equations, signs, shape, seed, is_color):
-    if (seed is not None):
-        random.seed(seed)
-    ret = []
-    sign_num = len(signs)
-    sign_index_dict = dict(zip(signs, list(range(sign_num))))
-    for equation in equations:
-        data = []
-        for sign in equation:
-            index = sign_index_dict[sign]
-            pick = random.randint(0, len(image_pools[index]) - 1)
-            if is_color:
-                image = Image.open(image_pools[index][pick]).convert('RGB').resize(shape)
-            else:
-                image = Image.open(image_pools[index][pick]).convert('I').resize(shape)
-            image_array = np.array(image)
-            image_array = (image_array-127)*(1./128)
-            data.append(image_array)
-        ret.append(np.array(data))
-    return ret
-
-def get_equation_std_data(data_dir, sign_dir_lists, sign_output_lists, shape = (28, 28), train_max_equation_len = 10, test_max_equation_len = 10, system_num = 2, tmp_file_prev =
-None, seed = None, train_num_per_len = 10, test_num_per_len = 10, is_color = False):
-    tmp_file = ""
-    if (tmp_file_prev is not None):
-        tmp_file = "%s_train_len_%d_test_len_%d_sys_%d_.pk" % (tmp_file_prev, train_max_equation_len, test_max_equation_len, system_num)
-    if (os.path.exists(tmp_file)):
-        return pickle.load(open(tmp_file, "rb"))
-
-    image_pools = get_sign_path_list(data_dir, sign_dir_lists)
-    train_pool, test_pool = split_pool_by_rate(image_pools, 0.8, seed)
-
-    ret = {}
-    for label in ["positive", "negative"]:
-        print("Generating equations.")
-        train_equations = generator_equations_by_max_len(train_max_equation_len, system_num, label, num_per_len = train_num_per_len)
-        test_equations = generator_equations_by_max_len(test_max_equation_len, system_num, label, num_per_len = test_num_per_len)
-        print(train_equations)
-        print(test_equations)
-        print("Generated equations.")
-        print("Generating equation image data.")
-        ret["train:%s" % (label)] = generator_equation_images(train_pool, train_equations, sign_output_lists, shape, seed, is_color)
-        ret["test:%s" % (label)] = generator_equation_images(test_pool, test_equations, sign_output_lists, shape, seed, is_color)
-        print("Generated equation image data.")
-
-    if (tmp_file_prev is not None):
-        pickle.dump(ret, open(tmp_file, "wb"))
-    return ret
-
-if __name__ == "__main__":
-    data_dirs = ["./dataset/hed/mnist_images", "./dataset/hed/random_images"] #, "../dataset/cifar10_images"]
-    tmp_file_prevs = ["mnist_equation_data", "random_equation_data"] #, "cifar10_equation_data"]
-    for data_dir, tmp_file_prev in zip(data_dirs, tmp_file_prevs):
-        data = get_equation_std_data(data_dir = data_dir,\
-                                     sign_dir_lists = ['0', '1', '10', '11'],\
-                                     sign_output_lists = ['0', '1', '+', '='],\
-                                     shape = (28, 28),\
-                                     train_max_equation_len = 26, \
-                                     test_max_equation_len = 26, \
-                                     system_num = 2, \
-                                     tmp_file_prev = tmp_file_prev, \
-                                     train_num_per_len = 300, \
-                                     test_num_per_len = 300, \
-                                     is_color = False)

examples/hed/hed.ipynb

@@ -344,7 +344,6 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Set up metrics\n",
     "metric_list = [SymbolAccuracy(prefix=\"hed\"), ReasoningMetric(kb=kb, prefix=\"hed\")]"
    ]
   },
@@ -390,7 +389,7 @@
     "log_dir = ABLLogger.get_current_instance().log_dir\n",
     "weights_dir = osp.join(log_dir, \"weights\")\n",
     "\n",
-    "bridge.pretrain(\"./weights\")\n",
+    "bridge.pretrain(weights_dir)\n",
     "bridge.train(train_data, val_data)\n",
     "bridge.test(test_data)"
    ]

examples/hed/main.py

@@ -0,0 +1,101 @@
+import os.path as osp
+import argparse
+
+import torch
+import torch.nn as nn
+
+from examples.hed.datasets import get_dataset, split_equation
+from examples.models.nn import SymbolNet
+from abl.learning import ABLModel, BasicNN
+from examples.hed.reasoning import HedKB, HedReasoner
+from abl.data.evaluation import ReasoningMetric, SymbolAccuracy
+from abl.utils import ABLLogger, print_log
+from examples.hed.bridge import HedBridge
+
+def main():
+    parser = argparse.ArgumentParser(description="Handwritten Equation Decipherment example")
+    parser.add_argument(
+        "--no-cuda", action="store_true", default=False, help="disables CUDA training"
+    )
+    parser.add_argument(
+        "--epochs",
+        type=int,
+        default=1,
+        help="number of epochs in each learning loop iteration (default : 1)",
+    )
+    parser.add_argument(
+        "--lr", type=float, default=1e-3, help="base model learning rate (default : 0.001)"
+    )
+    parser.add_argument(
+        "--weight-decay", type=float, default=1e-4, help="weight decay (default : 0.0001)"
+    )
+    parser.add_argument(
+        "--batch-size", type=int, default=32, help="base model batch size (default : 32)"
+    )
+    parser.add_argument(
+        "--segment_size", type=int or float, default=1000, help="segment size (default : 1000)"
+    )
+    parser.add_argument("--save_interval", type=int, default=1, help="save interval (default : 1)")
+    parser.add_argument(
+        "--max-revision",
+        type=int or float,
+        default=10,
+        help="maximum revision in reasoner (default : 10)",
+    )
+
+    args = parser.parse_args()
+
+    ### Working with Data
+    total_train_data = get_dataset(train=True)
+    train_data, val_data = split_equation(total_train_data, 3, 1)
+    test_data = get_dataset(train=False)
+
+    ### Building the Learning Part
+    # Build necessary components for BasicNN
+    cls = SymbolNet(num_classes=4)
+    loss_fn = nn.CrossEntropyLoss()
+    optimizer = torch.optim.RMSprop(cls.parameters(), lr=args.lr, weight_decay=args.weight_deccay)
+    use_cuda = not args.no_cuda and torch.cuda.is_available()
+    device = torch.device("cuda" if use_cuda else "cpu")
+
+    # Build BasicNN
+    base_model = BasicNN(
+        cls,
+        loss_fn,
+        optimizer,
+        device,
+        batch_size=args.batch_size,
+        num_epochs=args.epochs,
+        stop_loss=None,
+    )
+
+    # Build ABLModel
+    model = ABLModel(base_model)
+
+    ### Building the Reasoning Part
+    # Build knowledge base
+    kb = HedKB()
+
+    # Create reasoner
+    reasoner = HedReasoner(kb, dist_func="hamming", use_zoopt=True, max_revision=args.max_revision)
+
+    ### Building Evaluation Metrics
+    metric_list = [SymbolAccuracy(prefix="hed"), ReasoningMetric(kb=kb, prefix="hed")]
+    
+    ### Bridge Learning and Reasoning
+    bridge = HedBridge(model, reasoner, metric_list)
+
+    # Build logger
+    print_log("Abductive Learning on the HED example.", logger="current")
+
+    # Retrieve the directory of the Log file and define the directory for saving the model weights.
+    log_dir = ABLLogger.get_current_instance().log_dir
+    weights_dir = osp.join(log_dir, "weights")
+
+    bridge.pretrain(weights_dir)
+    bridge.train(train_data, val_data)
+    bridge.test(test_data)
+
+
+if __name__ == "__main__":
+    main()

examples/hwf/README.md

@@ -1,4 +1,4 @@
-# MNIST Addition Example
+# Handwritten Formula
 
 This example shows a simple implementation of [Handwritten Formula](https://arxiv.org/abs/2006.06649) task, where handwritten images of decimal formulas and their computed results are given, alongwith a domain knowledge base containing information on how to compute the decimal formula. The task is to recognize the symbols (which can be digits or operators '+', '-', '×', '÷') of handwritten images and accurately determine their results.
 
@@ -13,13 +13,13 @@ python main.py
 
 ```bash
 usage: main.py [-h] [--no-cuda] [--epochs EPOCHS] [--lr LR]
-               [--weight-decay WEIGHT_DECAY] [--batch-size BATCH_SIZE]
+               [--batch-size BATCH_SIZE]
                [--loops LOOPS] [--segment_size SEGMENT_SIZE]
                [--save_interval SAVE_INTERVAL] [--max-revision MAX_REVISION]
                [--require-more-revision REQUIRE_MORE_REVISION]
                [--ground] [--max-err MAX_ERR]
 
-MNIST Addition example
+Handwritten Formula example
 
 optional arguments:
   -h, --help            show this help message and exit

examples/hwf/main.py

@@ -67,7 +67,7 @@ class HwfGroundKB(GroundKB):
 
 
 def main():
-    parser = argparse.ArgumentParser(description="MNIST Addition example")
+    parser = argparse.ArgumentParser(description="Handwritten Formula example")
     parser.add_argument(
         "--no-cuda", action="store_true", default=False, help="disables CUDA training"
     )

examples/mnist_add/README.md

@@ -1,4 +1,4 @@
-# MNIST Addition Example
+# MNIST Addition
 
 This example shows a simple implementation of [MNIST Addition](https://arxiv.org/abs/1805.10872) task, where pairs of MNIST handwritten images and their sums are given, alongwith a domain knowledge base containing information on how to perform addition operations. The task is to recognize the digits of handwritten images and accurately determine their sum.
 

examples/mnist_add/mnist_add.ipynb

@@ -450,7 +450,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.8.18"
+   "version": "3.8.13"
   },
   "orig_nbformat": 4,
   "vscode": {

examples/zoo/README.md

@@ -0,0 +1,23 @@
+# Zoo Example
+
+This example shows a simple implementation of [Zoo](https://archive.ics.uci.edu/dataset/111/zoo). In this task, attributes of animals (such as presence of hair, eggs, etc.) and their targets (the animal class they belong to) are given, along with a knowledge base which contain information about the relations between attributes and targets, e.g., Implies(milk == 1, mammal == 1). The goal of this task is to develop a learning model that can predict the targets of animals based on their attributes.
+
+## Run
+
+```bash
+pip install -r requirements.txt
+python main.py
+```
+
+## Usage
+
+```bash
+usage: main.py [-h] [--loops LOOPS] 
+
+Zoo example
+
+optional arguments:
+  -h, --help            show this help message and exit
+  --loops LOOPS         number of loop iterations (default : 3)
+
+```

examples/zoo/kb.py

@@ -13,8 +13,8 @@ class ZooKB(KBBase):
         X, y, categorical_indicator, attribute_names = dataset.get_data(target=dataset.default_target_attribute)
         self.attribute_names = attribute_names
         self.target_names = y.cat.categories.tolist()
-        print("Attribute names are: ", self.attribute_names)
-        print("Target names are: ", self.target_names)
+        # print("Attribute names are: ", self.attribute_names)
+        # print("Target names are: ", self.target_names)
         # self.attribute_names = ["hair", "feathers", "eggs", "milk", "airborne", "aquatic", "predator", "toothed", "backbone", "breathes", "venomous", "fins", "legs", "tail", "domestic", "catsize"]
         # self.target_names = ["mammal", "bird", "reptile", "fish", "amphibian", "insect", "invertebrate"]
 

examples/zoo/main.py

@@ -1,119 +1,19 @@
 import os.path as osp
+import argparse
 
 import numpy as np
 from sklearn.ensemble import RandomForestClassifier
-from sklearn.metrics import accuracy_score
-from z3 import Solver, Int, If, Not, Implies, Sum, sat
-import openml
 
+from examples.zoo.get_dataset import load_and_preprocess_dataset, split_dataset
 from abl.learning import ABLModel
-from abl.reasoning import KBBase, Reasoner
+from examples.zoo.kb import ZooKB
+from abl.reasoning import Reasoner
 from abl.data.evaluation import ReasoningMetric, SymbolAccuracy
+from abl.utils import ABLLogger, print_log, confidence_dist
 from abl.bridge import SimpleBridge
-from abl.utils.utils import confidence_dist
-from abl.utils import ABLLogger, print_log
-
-# Build logger
-print_log("Abductive Learning on the Zoo example.", logger="current")
-
-# Retrieve the directory of the Log file and define the directory for saving the model weights.
-log_dir = ABLLogger.get_current_instance().log_dir
-weights_dir = osp.join(log_dir, "weights")
-
-# Learning Part
-rf = RandomForestClassifier()
-model = ABLModel(rf)
-
-# %% [markdown]
-# ### Logic Part
-
-
-# %%
-class ZooKB(KBBase):
-    def __init__(self):
-        super().__init__(pseudo_label_list=list(range(7)), use_cache=False)
-
-        # Use z3 solver
-        self.solver = Solver()
-
-        # Load information of Zoo dataset
-        dataset = openml.datasets.get_dataset(
-            dataset_id=62,
-            download_data=False,
-            download_qualities=False,
-            download_features_meta_data=False,
-        )
-        X, y, categorical_indicator, attribute_names = dataset.get_data(
-            target=dataset.default_target_attribute
-        )
-        self.attribute_names = attribute_names
-        self.target_names = y.cat.categories.tolist()
-
-        # Define variables
-        for name in self.attribute_names + self.target_names:
-            exec(
-                f"globals()['{name}'] = Int('{name}')"
-            )  ## or use dict to create var and modify rules
-        # Define rules
-        rules = [
-            Implies(milk == 1, mammal == 1),
-            Implies(mammal == 1, milk == 1),
-            Implies(mammal == 1, backbone == 1),
-            Implies(mammal == 1, breathes == 1),
-            Implies(feathers == 1, bird == 1),
-            Implies(bird == 1, feathers == 1),
-            Implies(bird == 1, eggs == 1),
-            Implies(bird == 1, backbone == 1),
-            Implies(bird == 1, breathes == 1),
-            Implies(bird == 1, legs == 2),
-            Implies(bird == 1, tail == 1),
-            Implies(reptile == 1, backbone == 1),
-            Implies(reptile == 1, breathes == 1),
-            Implies(reptile == 1, tail == 1),
-            Implies(fish == 1, aquatic == 1),
-            Implies(fish == 1, toothed == 1),
-            Implies(fish == 1, backbone == 1),
-            Implies(fish == 1, Not(breathes == 1)),
-            Implies(fish == 1, fins == 1),
-            Implies(fish == 1, legs == 0),
-            Implies(fish == 1, tail == 1),
-            Implies(amphibian == 1, eggs == 1),
-            Implies(amphibian == 1, aquatic == 1),
-            Implies(amphibian == 1, backbone == 1),
-            Implies(amphibian == 1, breathes == 1),
-            Implies(amphibian == 1, legs == 4),
-            Implies(insect == 1, eggs == 1),
-            Implies(insect == 1, Not(backbone == 1)),
-            Implies(insect == 1, legs == 6),
-            Implies(invertebrate == 1, Not(backbone == 1)),
-        ]
-        # Define weights and sum of violated weights
-        self.weights = {rule: 1 for rule in rules}
-        self.total_violation_weight = Sum(
-            [If(Not(rule), self.weights[rule], 0) for rule in self.weights]
-        )
-
-    def logic_forward(self, pseudo_label, data_point):
-        attribute_names, target_names = self.attribute_names, self.target_names
-        solver = self.solver
-        total_violation_weight = self.total_violation_weight
-        pseudo_label, data_point = pseudo_label[0], data_point[0]
-
-        self.solver.reset()
-        for name, value in zip(attribute_names, data_point):
-            solver.add(eval(f"{name} == {value}"))
-        for cate, name in zip(self.pseudo_label_list, target_names):
-            value = 1 if (cate == pseudo_label) else 0
-            solver.add(eval(f"{name} == {value}"))
-
-        if solver.check() == sat:
-            model = solver.model()
-            total_weight = model.evaluate(total_violation_weight)
-            return total_weight.as_long()
-        else:
-            # No solution found
-            return 1e10
 
+def transform_tab_data(X, y):
+    return ([[x] for x in X], [[y_item] for y_item in y], [0] * len(y))
 
 def consitency(data_example, candidates, candidate_idxs, reasoning_results):
     pred_prob = data_example.pred_prob
@@ -122,94 +22,54 @@ def consitency(data_example, candidates, candidate_idxs, reasoning_results):
     scores = model_scores + rule_scores
     return scores
 
-
-kb = ZooKB()
-reasoner = Reasoner(kb, dist_func=consitency)
-
-# %% [markdown]
-# ### Datasets and Evaluation Metrics
-
-
-# %%
-# Function to load and preprocess the dataset
-def load_and_preprocess_dataset(dataset_id):
-    dataset = openml.datasets.get_dataset(
-        dataset_id, download_data=True, download_qualities=False, download_features_meta_data=False
+def main():
+    parser = argparse.ArgumentParser(description="Zoo example")
+    parser.add_argument(
+        "--loops", type=int, default=3, help="number of loop iterations (default : 3)"
     )
-    X, y, _, attribute_names = dataset.get_data(target=dataset.default_target_attribute)
-    # Convert data types
-    for col in X.select_dtypes(include="bool").columns:
-        X[col] = X[col].astype(int)
-    y = y.cat.codes.astype(int)
-    X, y = X.to_numpy(), y.to_numpy()
-    return X, y
-
-
-# Function to split data (one shot)
-def split_dataset(X, y, test_size=0.3):
-    # For every class: 1 : (1-test_size)*(len-1) : test_size*(len-1)
-    label_indices, unlabel_indices, test_indices = [], [], []
-    for class_label in np.unique(y):
-        idxs = np.where(y == class_label)[0]
-        np.random.shuffle(idxs)
-        n_train_unlabel = int((1 - test_size) * (len(idxs) - 1))
-        label_indices.append(idxs[0])
-        unlabel_indices.extend(idxs[1 : 1 + n_train_unlabel])
-        test_indices.extend(idxs[1 + n_train_unlabel :])
-    X_label, y_label = X[label_indices], y[label_indices]
-    X_unlabel, y_unlabel = X[unlabel_indices], y[unlabel_indices]
-    X_test, y_test = X[test_indices], y[test_indices]
-    return X_label, y_label, X_unlabel, y_unlabel, X_test, y_test
-
-
-# Load and preprocess the Zoo dataset
-X, y = load_and_preprocess_dataset(dataset_id=62)
-
-# Split data into labeled/unlabeled/test data
-X_label, y_label, X_unlabel, y_unlabel, X_test, y_test = split_dataset(X, y, test_size=0.3)
-
-
-# Transform tabluar data to the format required by ABL, which is a tuple of (X, ground truth of X, reasoning results)
-# For tabular data in abl, each example contains a single instance (a row from the dataset).
-# For these tabular data examples, the reasoning results are expected to be 0, indicating no rules are violated.
-def transform_tab_data(X, y):
-    return ([[x] for x in X], [[y_item] for y_item in y], [0] * len(y))
-
-
-label_data = transform_tab_data(X_label, y_label)
-test_data = transform_tab_data(X_test, y_test)
-train_data = transform_tab_data(X_unlabel, y_unlabel)
-
-# %%
-# Set up metrics
-metric_list = [SymbolAccuracy(prefix="zoo"), ReasoningMetric(kb=kb, prefix="zoo")]
-
-# %% [markdown]
-# ### Bridge Machine Learning and Logic Reasoning
-
-# %%
-bridge = SimpleBridge(model, reasoner, metric_list)
-
-# %% [markdown]
-# ### Train and Test
-
-# %%
-# Pre-train the machine learning model
-rf.fit(X_label, y_label)
-
-# %%
-# Test the initial model
-print("------- Test the initial model -----------")
-bridge.test(test_data)
-print("------- Use ABL to train the model -----------")
-# Use ABL to train the model
-bridge.train(
-    train_data=train_data,
-    label_data=label_data,
-    loops=3,
-    segment_size=len(X_unlabel),
-    save_dir=weights_dir,
-)
-print("------- Test the final model -----------")
-# Test the final model
-bridge.test(test_data)
+    args = parser.parse_args()
+
+    ### Working with Data
+    X, y = load_and_preprocess_dataset(dataset_id=62)
+    X_label, y_label, X_unlabel, y_unlabel, X_test, y_test = split_dataset(X, y, test_size=0.3)
+    label_data = transform_tab_data(X_label, y_label)
+    test_data = transform_tab_data(X_test, y_test)
+    train_data = transform_tab_data(X_unlabel, y_unlabel)
+
+    ### Building the Learning Part
+    base_model = RandomForestClassifier()
+
+    # Build ABLModel
+    model = ABLModel(base_model)
+
+    ### Building the Reasoning Part
+    # Build knowledge base
+    kb = ZooKB()
+    
+    # Create reasoner
+    reasoner = Reasoner(kb, dist_func=consitency)
+
+    ### Building Evaluation Metrics
+    metric_list = [SymbolAccuracy(prefix="zoo"), ReasoningMetric(kb=kb, prefix="zoo")]
+    
+    # Build logger
+    print_log("Abductive Learning on the ZOO example.", logger="current")
+    log_dir = ABLLogger.get_current_instance().log_dir
+    weights_dir = osp.join(log_dir, "weights")
+    
+    ### Bridging learning and reasoning
+    bridge = SimpleBridge(model, reasoner, metric_list)
+    
+    # Performing training and testing
+    print_log("------- Use labeled data to pretrain the model -----------", logger="current")
+    base_model.fit(X_label, y_label)
+    print_log("------- Test the initial model -----------", logger="current")
+    bridge.test(test_data)
+    print_log("------- Use ABL to train the model -----------", logger="current")
+    bridge.train(train_data=train_data, label_data=label_data, loops=args.loops, segment_size=len(X_unlabel), save_dir=weights_dir)
+    print_log("------- Test the final model -----------", logger="current")
+    bridge.test(test_data)
+
+
+if __name__ == "__main__":
+    main()

examples/zoo/zoo.ipynb

@@ -6,9 +6,9 @@
    "source": [
     "# ZOO\n",
     "\n",
-    "This notebook shows an implementation of [MNIST Addition](https://arxiv.org/abs/1805.10872). In this task, pairs of MNIST handwritten images and their sums are given, alongwith a domain knowledge base containing information on how to perform addition operations. The task is to recognize the digits of handwritten images and accurately determine their sum.\n",
+    "This notebook shows an implementation of [Zoo](https://archive.ics.uci.edu/dataset/111/zoo). In this task, attributes of animals (such as presence of hair, eggs, etc.) and their targets (the animal class they belong to) are given, along with a knowledge base which contain information about the relations between attributes and targets, e.g., Implies(milk == 1, mammal == 1). \n",
     "\n",
-    "Intuitively, we first use a machine learning model (learning part) to convert the input images to digits (we call them pseudo-labels), and then use the knowledge base (reasoning part) to calculate the sum of these digits. Since we do not have ground-truth of the digits, in Abductive Learning, the reasoning part will leverage domain knowledge and revise the initial digits yielded by the learning part through abductive reasoning. This process enables us to further update the machine learning model."
+    "The goal of this task is to develop a learning model that can predict the targets of animals based on their attributes. In the initial stages, when the model is under-trained, it may produce incorrect predictions that conflict with the relations contained in the knowledge base. When this happens, abductive reasoning can be employed to adjust these results and retrain the model accordingly. This process enables us to further update the learning model."
    ]
   },
   {
@@ -36,7 +36,7 @@
    "source": [
     "## Working with Data\n",
     "\n",
-    "First, we get the training and testing datasets:"
+    "First, we load and preprocess the [Zoo dataset](https://archive.ics.uci.edu/dataset/111/zoo), and split it into labeled/unlabeled/test data"
    ]
   },
   {
@@ -45,10 +45,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Load and preprocess the Zoo dataset\n",
     "X, y = load_and_preprocess_dataset(dataset_id=62)\n",
-    "\n",
-    "# Split data into labeled/unlabeled/test data\n",
     "X_label, y_label, X_unlabel, y_unlabel, X_test, y_test = split_dataset(X, y, test_size=0.3)"
    ]
   },
@@ -56,9 +53,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "`train_data` and `test_data` share identical structures: tuples with three components: X (list where each element is a list of two images), gt_pseudo_label (list where each element is a list of two digits, i.e., pseudo-labels) and Y (list where each element is the sum of the two digits). The length and structures of datasets are illustrated as follows.\n",
-    "\n",
-    "Note: ``gt_pseudo_label`` is only used to evaluate the performance of the learning part but not to train the model."
+    "Zoo dataset consist of tabular data. The attributes contains 17 boolean values (e.g., hair, feathers, eggs, milk, airborne, aquatic, etc.) and the target is a integer value in range [0,6] representing 7 classes (e.g., mammal, bird, reptile, fish, amphibian, insect, and other). Below is an illustration:"
    ]
   },
   {
@@ -99,16 +94,12 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "Transform tabluar data to the format required by ABL-Package, which is a tuple of (X, gt_pseudo_label, Y)\n",
-    "\n",
-    "For tabular data in abl, each example contains a single instance (a row from the dataset).\n",
-    "\n",
-    "For these tabular data samples, the reasoning results are expected to be 0, indicating no rules are violated."
+    "Next, we transform the tabular data to the format required by ABL-Package, which is a tuple of (X, gt_pseudo_label, Y). In this task, we treat the attributes as X and the targets as gt_pseudo_label (ground truth pseudo-labels). Y (reasoning results) are expected to be 0, indicating no rules are violated."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 4,
+   "execution_count": 3,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -131,28 +122,14 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "To build the learning part, we need to first build a machine learning base model. We use a [Random Forest](https://en.wikipedia.org/wiki/Random_forest) as the base model"
+    "To build the learning part, we need to first build a machine learning base model. We use a [Random Forest](https://en.wikipedia.org/wiki/Random_forest) as the base model."
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 5,
    "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/html": [
-       "<style>#sk-container-id-1 {color: black;}#sk-container-id-1 pre{padding: 0;}#sk-container-id-1 div.sk-toggleable {background-color: white;}#sk-container-id-1 label.sk-toggleable__label {cursor: pointer;display: block;width: 100%;margin-bottom: 0;padding: 0.3em;box-sizing: border-box;text-align: center;}#sk-container-id-1 label.sk-toggleable__label-arrow:before {content: \"▸\";float: left;margin-right: 0.25em;color: #696969;}#sk-container-id-1 label.sk-toggleable__label-arrow:hover:before {color: black;}#sk-container-id-1 div.sk-estimator:hover label.sk-toggleable__label-arrow:before {color: black;}#sk-container-id-1 div.sk-toggleable__content {max-height: 0;max-width: 0;overflow: hidden;text-align: left;background-color: #f0f8ff;}#sk-container-id-1 div.sk-toggleable__content pre {margin: 0.2em;color: black;border-radius: 0.25em;background-color: #f0f8ff;}#sk-container-id-1 input.sk-toggleable__control:checked~div.sk-toggleable__content {max-height: 200px;max-width: 100%;overflow: auto;}#sk-container-id-1 input.sk-toggleable__control:checked~label.sk-toggleable__label-arrow:before {content: \"▾\";}#sk-container-id-1 div.sk-estimator input.sk-toggleable__control:checked~label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-1 div.sk-label input.sk-toggleable__control:checked~label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-1 input.sk-hidden--visually {border: 0;clip: rect(1px 1px 1px 1px);clip: rect(1px, 1px, 1px, 1px);height: 1px;margin: -1px;overflow: hidden;padding: 0;position: absolute;width: 1px;}#sk-container-id-1 div.sk-estimator {font-family: monospace;background-color: #f0f8ff;border: 1px dotted black;border-radius: 0.25em;box-sizing: border-box;margin-bottom: 0.5em;}#sk-container-id-1 div.sk-estimator:hover {background-color: #d4ebff;}#sk-container-id-1 div.sk-parallel-item::after {content: \"\";width: 100%;border-bottom: 1px solid gray;flex-grow: 1;}#sk-container-id-1 div.sk-label:hover label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-1 div.sk-serial::before {content: \"\";position: absolute;border-left: 1px solid gray;box-sizing: border-box;top: 0;bottom: 0;left: 50%;z-index: 0;}#sk-container-id-1 div.sk-serial {display: flex;flex-direction: column;align-items: center;background-color: white;padding-right: 0.2em;padding-left: 0.2em;position: relative;}#sk-container-id-1 div.sk-item {position: relative;z-index: 1;}#sk-container-id-1 div.sk-parallel {display: flex;align-items: stretch;justify-content: center;background-color: white;position: relative;}#sk-container-id-1 div.sk-item::before, #sk-container-id-1 div.sk-parallel-item::before {content: \"\";position: absolute;border-left: 1px solid gray;box-sizing: border-box;top: 0;bottom: 0;left: 50%;z-index: -1;}#sk-container-id-1 div.sk-parallel-item {display: flex;flex-direction: column;z-index: 1;position: relative;background-color: white;}#sk-container-id-1 div.sk-parallel-item:first-child::after {align-self: flex-end;width: 50%;}#sk-container-id-1 div.sk-parallel-item:last-child::after {align-self: flex-start;width: 50%;}#sk-container-id-1 div.sk-parallel-item:only-child::after {width: 0;}#sk-container-id-1 div.sk-dashed-wrapped {border: 1px dashed gray;margin: 0 0.4em 0.5em 0.4em;box-sizing: border-box;padding-bottom: 0.4em;background-color: white;}#sk-container-id-1 div.sk-label label {font-family: monospace;font-weight: bold;display: inline-block;line-height: 1.2em;}#sk-container-id-1 div.sk-label-container {text-align: center;}#sk-container-id-1 div.sk-container {/* jupyter's `normalize.less` sets `[hidden] { display: none; }` but bootstrap.min.css set `[hidden] { display: none !important; }` so we also need the `!important` here to be able to override the default hidden behavior on the sphinx rendered scikit-learn.org. See: https://github.com/scikit-learn/scikit-learn/issues/21755 */display: inline-block !important;position: relative;}#sk-container-id-1 div.sk-text-repr-fallback {display: none;}</style><div id=\"sk-container-id-1\" class=\"sk-top-container\"><div class=\"sk-text-repr-fallback\"><pre>RandomForestClassifier()</pre><b>In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook. <br />On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.</b></div><div class=\"sk-container\" hidden><div class=\"sk-item\"><div class=\"sk-estimator sk-toggleable\"><input class=\"sk-toggleable__control sk-hidden--visually\" id=\"sk-estimator-id-1\" type=\"checkbox\" checked><label for=\"sk-estimator-id-1\" class=\"sk-toggleable__label sk-toggleable__label-arrow\">RandomForestClassifier</label><div class=\"sk-toggleable__content\"><pre>RandomForestClassifier()</pre></div></div></div></div></div>"
-      ],
-      "text/plain": [
-       "RandomForestClassifier()"
-      ]
-     },
-     "execution_count": 5,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
+   "outputs": [],
    "source": [
     "base_model = RandomForestClassifier()"
    ]
@@ -166,7 +143,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 6,
+   "execution_count": 7,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -184,12 +161,12 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "In the reasoning part, we first build a knowledge base which contain information on how to perform addition operations. We build it by creating a subclass of `KBBase`. In the derived subclass, we initialize the `pseudo_label_list` parameter specifying list of possible pseudo-labels, and override the `logic_forward` function defining how to perform (deductive) reasoning."
+    "In the reasoning part, we first build a knowledge base which contains information about the relations between attributes (X) and targets (pseudo-labels), e.g., Implies(milk == 1, mammal == 1). The knowledge base is built in the `ZooKB` class within file `kb.py`, and is derived from the `KBBase` class."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
+   "execution_count": 8,
    "metadata": {},
    "outputs": [
     {
@@ -209,36 +186,46 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "The knowledge base can perform logical reasoning (both deductive reasoning and abductive reasoning). Below is an example of performing (deductive) reasoning, and users can refer to [Documentation]() for details of abductive reasoning."
+    "As mentioned, for all attributes and targets in the dataset, the reasoning results are expected to be 0 since there should be no violations of the established knowledge in real data. As shown below:"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": 15,
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Reasoning result of pseudo-label example [1, 2] is 3.\n"
+      "Example 0: the attributes are: [True False False True False False True True True True False False 4 False\n",
+      " False True], and the target is 0.\n",
+      "Reasoning result is 0.\n",
+      "\n",
+      "Example 1: the attributes are: [True False False True False False False True True True False False 4 True\n",
+      " False True], and the target is 0.\n",
+      "Reasoning result is 0.\n",
+      "\n",
+      "Example 2: the attributes are: [False False True False False True True True True False False True 0 True\n",
+      " False False], and the target is 3.\n",
+      "Reasoning result is 0.\n",
+      "\n",
+      "Example 3: the attributes are: [True False False True False False True True True True False False 4 False\n",
+      " False True], and the target is 0.\n",
+      "Reasoning result is 0.\n",
+      "\n",
+      "Example 4: the attributes are: [True False False True False False True True True True False False 4 True\n",
+      " False True], and the target is 0.\n",
+      "Reasoning result is 0.\n",
+      "\n"
      ]
     }
    ],
    "source": [
-    "pseudo_label = [0]\n",
-    "data_point = [np.array([1,0,0,1,0,0,1,1,1,1,0,0,4,0,0,1,1])]\n",
-    "print(kb.logic_forward(pseudo_label, data_point))\n",
-    "for x, y_item in zip(X, y):\n",
-    "    print(x,y_item)\n",
-    "    print(kb.logic_forward([y_item], [x]))"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Note: In addition to building a knowledge base based on `KBBase`, we can also establish a knowledge base with a ground KB using `GroundKB`, or a knowledge base implemented based on Prolog files using `PrologKB`. The corresponding code for these implementations can be found in the `main.py` file. Those interested are encouraged to examine it for further insights."
+    "for idx, (x, y_item) in enumerate(zip(X[:5], y[:5])):\n",
+    "    print(f\"Example {idx}: the attributes are: {x}, and the target is {y_item}.\")\n",
+    "    print(f\"Reasoning result is {kb.logic_forward([y_item], [x])}.\")\n",
+    "    print()"
    ]
   },
   {
@@ -250,7 +237,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 11,
+   "execution_count": 16,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -281,7 +268,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": 17,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -300,7 +287,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": 18,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -316,28 +303,54 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 22,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "12/22 11:01:21 - abl - INFO - Abductive Learning on the ZOO example.\n",
+      "12/22 11:01:21 - abl - INFO - ------- Use labeled data to pretrain the model -----------\n",
+      "12/22 11:01:21 - abl - INFO - ------- Test the initial model -----------\n",
+      "12/22 11:01:21 - abl - INFO - Evaluation ended, zoo/character_accuracy: 0.935 zoo/reasoning_accuracy: 0.935 \n",
+      "12/22 11:01:21 - abl - INFO - ------- Use ABL to train the model -----------\n",
+      "12/22 11:01:21 - abl - INFO - loop(train) [1/3] segment(train) [1/1] \n",
+      "12/22 11:01:23 - abl - INFO - Evaluation start: loop(val) [1]\n",
+      "12/22 11:01:23 - abl - INFO - Evaluation ended, zoo/character_accuracy: 0.984 zoo/reasoning_accuracy: 1.000 \n",
+      "12/22 11:01:23 - abl - INFO - loop(train) [2/3] segment(train) [1/1] \n",
+      "12/22 11:01:24 - abl - INFO - Evaluation start: loop(val) [2]\n",
+      "12/22 11:01:25 - abl - INFO - Evaluation ended, zoo/character_accuracy: 0.984 zoo/reasoning_accuracy: 1.000 \n",
+      "12/22 11:01:25 - abl - INFO - loop(train) [3/3] segment(train) [1/1] \n",
+      "12/22 11:01:26 - abl - INFO - Evaluation start: loop(val) [3]\n",
+      "12/22 11:01:26 - abl - INFO - Evaluation ended, zoo/character_accuracy: 0.984 zoo/reasoning_accuracy: 1.000 \n",
+      "12/22 11:01:26 - abl - INFO - ------- Test the final model -----------\n",
+      "12/22 11:01:27 - abl - INFO - Evaluation ended, zoo/character_accuracy: 0.903 zoo/reasoning_accuracy: 0.935 \n"
+     ]
+    }
+   ],
    "source": [
     "# Build logger\n",
     "print_log(\"Abductive Learning on the ZOO example.\", logger=\"current\")\n",
     "log_dir = ABLLogger.get_current_instance().log_dir\n",
     "weights_dir = osp.join(log_dir, \"weights\")\n",
     "\n",
-    "# Pre-train the machine learning model\n",
+    "print_log(\"------- Use labeled data to pretrain the model -----------\", logger=\"current\")\n",
     "base_model.fit(X_label, y_label)\n",
-    "\n",
-    "# Test the initial model\n",
-    "print(\"------- Test the initial model -----------\")\n",
+    "print_log(\"------- Test the initial model -----------\", logger=\"current\")\n",
     "bridge.test(test_data)\n",
-    "print(\"------- Use ABL to train the model -----------\")\n",
-    "# Use ABL to train the model\n",
+    "print_log(\"------- Use ABL to train the model -----------\", logger=\"current\")\n",
     "bridge.train(train_data=train_data, label_data=label_data, loops=3, segment_size=len(X_unlabel), save_dir=weights_dir)\n",
-    "print(\"------- Test the final model -----------\")\n",
-    "# Test the final model\n",
+    "print_log(\"------- Test the final model -----------\", logger=\"current\")\n",
     "bridge.test(test_data)"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We may see from the results, after undergoing training with ABL, the model's accuracy has improved."
+   ]
   }
  ],
  "metadata": {
@@ -356,7 +369,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.8.18"
+   "version": "3.8.13"
   },
   "orig_nbformat": 4,
   "vscode": {