Merge pull request #16 from Julian-Theis/development

Sampling hyperamater k, variant log size and biased variant log investigations

.gitignore

@@ -1,4 +1,5 @@
 .idea/
+conf/settings.py
 
 data/pns/*
 !data/pns/.gitkeep
@@ -14,3 +15,54 @@ data/avatar/variants/*
 
 data/avatar/train_data/*
 !data/avatar/train_data/.gitkeep
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+
+# C extensions
+*.so
+
+# Distribution / packaging
+bin/
+build/
+develop-eggs/
+dist/
+eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+*.egg-info/
+.installed.cfg
+*.egg
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+.tox/
+.coverage
+.cache
+nosetests.xml
+coverage.xml
+
+# Translations
+*.mo
+
+# Mr Developer
+.mr.developer.cfg
+.project
+.pydevproject
+
+# Rope
+.ropeproject
+
+# Django stuff:
+*.log
+*.pot
+
+# Sphinx documentation
+docs/_build/

README.md

@@ -34,6 +34,10 @@ This leads conclusively to a closer expected generalization score when assessing
  </center>
 
 # How To
+## Install
+AVATAR has been developed using Python 3.6. The required Python libraries are available in the [requirements-cpu.txt](requirements-cpu.txt) and [requirements-gpu.txt](requirements-gpu.txt) file, depending if you plan on using a CPU or GPU environment.
+Please run the [install/install_script.py](install/install_script.py) after creating your Python environment.
+
 ## Systems
 ### System Variant Playout
 To playout a unique system variant log from e.g. *PA System 11 3*, simply run the following command from the base directory. The script will create the variant log of the system, a train variant log, and a test variant log as txt files. Moreover, the train variants are used to create a CSV and an XES based event log suitable for process discovery.
@@ -91,7 +95,7 @@ python -m avatar.sampling --system <...> --job <0/1> --gpu <...> --suffix <...>
 Evaluate the sampling. 
 
 ```python
-python -m avatar.evaluation --system <...> --job <0/1> --gpu <...> --suffix <...> --strategy <naive/mh>
+python -m avatar.evaluation --system <...> --job <0/1> --suffix <...> --strategy <naive/mh>
 ```
 
 ### Generalization

avatar/cleanup.py

@@ -0,0 +1,46 @@
+import numpy as np
+from tqdm import tqdm
+import os
+from os import listdir
+from os.path import isfile, join
+import argparse
+from conf.settings import DATA_PATH
+
+WORK_PATH = os.path.abspath(os.getcwd())
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument('-sfx', '--suffix', help='Suffix (selected epoch, e.g. 1981, to keep)', required=True)
+    parser.add_argument('-s', '--system', help='System name', required=True)
+    parser.add_argument('-j', '--job', help='Job (0,1)', required=True)
+
+    args = parser.parse_args()
+
+    system = args.system
+    suffix = int(args.suffix)
+    job = int(args.job)
+
+    if DATA_PATH is None:
+        directory = os.path.join(WORK_PATH, "data", "avatar", "sgans", system, str(job), "tf_logs", "ckpt")
+    else:
+        directory = os.path.join(DATA_PATH, "avatar", "sgans", system, str(job), "tf_logs", "ckpt")
+
+    allfiles = [f for f in listdir(directory) if isfile(join(directory, f))]
+    to_delete = []
+    for f in allfiles:
+        if f == "checkpoint" or "iw_dict_" in f:
+            continue
+
+        if ".pre_model-" in f:
+            to_delete.append(f)
+            continue
+
+        if system in f and ".adv_model-" in f:
+            index = int(str(f.split(system + ".adv_model-")[1]).split(".")[0])
+            if index != suffix:
+                to_delete.append(f)
+
+    for f in tqdm(to_delete):
+        os.remove(os.path.join(directory, f))
+
+    print("Done.")

avatar/evaluation.py

@@ -1,6 +1,10 @@
 import numpy as np
+import os
 import argparse
 from util.playout import readVariantFile
+from conf.settings import DATA_PATH
+
+WORK_PATH = os.path.abspath(os.getcwd())
 
 def readFile(f_name, unique=False):
     traces = []
@@ -38,15 +42,25 @@ def intersection(lst1, lst2):
     return list(set(ls1) & set(ls2))
 
 def load_data(system, suffix, job, strategy=None):
-    f_train = "data/variants/" + system + "_train.txt"
-    f_test = "data/variants/" + system + "_test.txt"
-    f_pop = "data/variants/" + system + "_pop.txt"
-    f_eval = "data/avatar/train_data/" + system + "_eval.txt"
 
-    if strategy is None:
-        f_gan = "data/avatar/variants/" + system + "_relgan_" + str(suffix) + "_j" + str(job) + ".txt"
+    if DATA_PATH is None:
+        f_train = os.path.join(WORK_PATH, "data", "variants", system + "_train.txt")
+        f_test = os.path.join(WORK_PATH, "data", "variants", system + "_test.txt")
+        f_pop = os.path.join(WORK_PATH, "data", "variants", system + "_pop.txt")
+        f_eval = os.path.join(WORK_PATH, "data", "avatar", "train_data", system + "_eval.txt")
+        if strategy is None:
+            f_gan = os.path.join(WORK_PATH, "data", "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + ".txt")
+        else:
+            f_gan = os.path.join(WORK_PATH, "data", "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + ".txt")
     else:
-        f_gan = "data/avatar/variants/" + system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + ".txt"
+        f_train = os.path.join(DATA_PATH, "variants", system + "_train.txt")
+        f_test = os.path.join(DATA_PATH, "variants", system + "_test.txt")
+        f_pop = os.path.join(DATA_PATH, "variants", system + "_pop.txt")
+        f_eval = os.path.join(DATA_PATH, "avatar", "train_data", system + "_eval.txt")
+        if strategy is None:
+            f_gan = os.path.join(DATA_PATH, "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + ".txt")
+        else:
+            f_gan = os.path.join(DATA_PATH, "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + ".txt")
 
 
     train = readVariantFile(f_train, unique=False)

avatar/generalization.py

@@ -9,6 +9,10 @@
 from pm4py.evaluation.replay_fitness import factory as replay_factory
 from pm4py.evaluation.precision import factory as precision_factory
 
+from conf.settings import DATA_PATH
+
+WORK_PATH = os.path.abspath(os.getcwd())
+
 def readFile(f_name1, f_name2, unique=False):
     traces = []
 
@@ -93,12 +97,19 @@ def convertToCsv(traces, to_path):
     pn = args.pn
     strategy = args.strategy
 
-    train_file = "data/variants/" + system + "_train.txt"
-    gen_file = "data/avatar/variants/" + system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + ".txt"
-    csv_file = "data/avatar/variants/" + system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + "_generalization.csv"
-    xes_file = "data/avatar/variants/" + system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + "_generalization.xes"
+    if DATA_PATH is None:
+        train_file = os.path.join(WORK_PATH, "data", "variants", system + "_train.txt")
+        gen_file = os.path.join(WORK_PATH, "data", "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + ".txt")
+        csv_file = os.path.join(WORK_PATH, "data", "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + "_generalization.csv")
+        xes_file = os.path.join(WORK_PATH, "data", "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + "_generalization.xes")
+        pn_file = os.path.join(WORK_PATH, "data", "pns", system, pn)
+    else:
+        train_file = os.path.join(DATA_PATH, "variants", system + "_train.txt")
+        gen_file = os.path.join(DATA_PATH, "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + ".txt")
+        csv_file = os.path.join(DATA_PATH, "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + "_generalization.csv")
+        xes_file = os.path.join(DATA_PATH, "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + "_" + strategy + "_generalization.xes")
+        pn_file = os.path.join(DATA_PATH, "pns", system, pn)
 
-    pn_file = os.path.join('data/pns', system, pn)
 
     """ READ FILES AND CONVERT TO XES """
     traces = readFile(train_file,gen_file, unique=True)

avatar/relgan/real/real_gan/real_train.py

@@ -40,63 +40,8 @@ def real_train(generator, discriminator, oracle_loader, config, prob_discriminat
     gen_file = os.path.join(sample_dir, 'generator.txt')
     gen_text_file = os.path.join(sample_dir, 'generator_text.txt')
     csv_file = os.path.join(log_dir, 'experiment-log-rmcgan.csv')
-    data_file = os.path.join(data_dir, '{}.txt'.format(dataset))
-    if dataset == 'image_coco':
-        test_file = os.path.join(data_dir, 'testdata/test_coco.txt')
-    elif dataset == 'emnlp_news':
-        test_file = os.path.join(data_dir, 'testdata/test_emnlp.txt')
-    elif dataset == 'logic1':
-        test_file = os.path.join(data_dir, 'logic1.txt')
-    elif dataset == 'pb_system_1_5':
-        test_file = os.path.join(data_dir, 'pb_system_1_5.txt')
-    elif dataset == 'pb_system_2_4':
-        test_file = os.path.join(data_dir, 'pb_system_2_4.txt')
-    elif dataset == 'pb_system_3_6':
-        test_file = os.path.join(data_dir, 'pb_system_3_6.txt')
-    elif dataset == 'pb_system_4_1':
-        test_file = os.path.join(data_dir, 'pb_system_4_1.txt')
-    elif dataset == 'pb_system_4_1_10':
-        test_file = os.path.join(data_dir, 'pb_system_4_1_10.txt')
-    elif dataset == 'pb_system_4_1_30':
-        test_file = os.path.join(data_dir, 'pb_system_4_1_30.txt')
-    elif dataset == 'pb_system_4_1_50':
-        test_file = os.path.join(data_dir, 'pb_system_4_1_50.txt')
-    elif dataset == 'pb_system_5_3':
-        test_file = os.path.join(data_dir, 'pb_system_5_3.txt')
-
-    elif dataset == 'pb_system_5_3_b':
-        test_file = os.path.join(data_dir, 'pb_system_5_3_b.txt')
-    elif dataset == 'pb_system_1_5_b':
-        test_file = os.path.join(data_dir, 'pb_system_1_5_b.txt')
-    elif dataset == 'pb_system_2_4_b':
-        test_file = os.path.join(data_dir, 'pb_system_2_4_b.txt')
-    elif dataset == 'pb_system_3_6_b':
-        test_file = os.path.join(data_dir, 'pb_system_3_6_b.txt')
-    elif dataset == 'pb_system_4_1_b':
-        test_file = os.path.join(data_dir, 'pb_system_4_1_b.txt')
-
-    elif dataset == 'pa_system_1_5':
-        test_file = os.path.join(data_dir, 'pa_system_1_5.txt')
-    elif dataset == 'pa_system_2_3':
-        test_file = os.path.join(data_dir, 'pa_system_2_3.txt')
-    elif dataset == 'pa_system_4_3':
-        test_file = os.path.join(data_dir, 'pa_system_4_3.txt')
-    elif dataset == 'pa_system_5_5':
-        test_file = os.path.join(data_dir, 'pa_system_5_5.txt')
-    elif dataset == 'pa_system_6_3':
-        test_file = os.path.join(data_dir, 'pa_system_6_3.txt')
-    elif dataset == 'pa_system_7_7':
-        test_file = os.path.join(data_dir, 'pa_system_7_7.txt')
-    elif dataset == 'pa_system_8_3':
-        test_file = os.path.join(data_dir, 'pa_system_8_3.txt')
-    elif dataset == 'pa_system_9_1':
-        test_file = os.path.join(data_dir, 'pa_system_9_1.txt')
-    elif dataset == 'pa_system_10_2':
-        test_file = os.path.join(data_dir, 'pa_system_10_2.txt')
-    elif dataset == 'pa_system_11_3':
-        test_file = os.path.join(data_dir, 'pa_system_11_3.txt')
-    else:
-        raise NotImplementedError('Unknown dataset!')
+    data_file = os.path.join(data_dir, "..", "train_data", '{}.txt'.format(dataset))
+    test_file = os.path.join(data_dir, "..", "train_data", '{}.txt'.format(dataset))
 
     # create necessary directories
     if not os.path.exists(data_dir):

avatar/relgan/run.py

@@ -6,6 +6,9 @@
 from avatar.relgan.utils.text_process import text_precess
 from avatar.relgan.utils.utils import pp
 import avatar.relgan.models as models
+from conf.settings import DATA_PATH
+
+WORK_PATH = os.path.abspath(os.getcwd())
 
 def main(given_args=None):
     parser = argparse.ArgumentParser(description='Train and run a RelGAN')
@@ -60,7 +63,15 @@ def main(given_args=None):
     parser.add_argument('--gen-emb-dim', default=32, type=int, help="generator embedding dimension")
     parser.add_argument('--dis-emb-dim', default=64, type=int, help="TOTAL discriminator embedding dimension")
     parser.add_argument('--num-rep', default=64, type=int, help="number of discriminator embedded representations")
-    parser.add_argument('--data-dir', default='/data/julian/data/relgan/data', type=str,
+
+    if DATA_PATH is None:
+        def_dir = os.path.join(WORK_PATH, "data", "avatar", "sgans")
+    else:
+        def_dir = os.path.join(DATA_PATH, "avatar", "sgans")
+
+    parser.add_argument('--data-dir',
+                        default=def_dir,
+                        type=str,
                         help='Where data data is stored')
 
     if given_args is None:
@@ -72,12 +83,17 @@ def main(given_args=None):
 
     print(config)
 
-    data_file = os.path.join(args.data_dir, '{}.txt'.format(args.dataset))
+    #data_file = os.path.join(args.data_dir, "..", "train_data", '{}.txt'.format(args.dataset))
+
+    if DATA_PATH is None:
+        seq_vocab_file = os.path.join(WORK_PATH, "data", "variants", str(args.dataset) + "_train.txt")
+    else:
+        seq_vocab_file = os.path.join(DATA_PATH, "variants", str(args.dataset) + "_train.txt")
 
-    if args.dataset == 'pb_system_4_1_10':
+    if args.dataset == 'pb_system_4_1_10' or '_s10' in args.dataset:
         args.batch_size = 32
 
-    seq_len, vocab_size = text_precess(data_file)
+    seq_len, vocab_size = text_precess(seq_vocab_file)
     config['seq_len'] = seq_len
     config['vocab_size'] = vocab_size
     print('seq_len: %d, vocab_size: %d' % (seq_len, vocab_size))

avatar/sampling.py

@@ -6,6 +6,9 @@
 from avatar.util.LoadRelgan import LoadRelgan
 from avatar.util.MHGAN import MHGAN
 from avatar.util.util import writeToFile, readTraces
+from conf.settings import DATA_PATH
+
+WORK_PATH = os.path.abspath(os.getcwd())
 
 if __name__ == "__main__":
     np.random.seed(seed=1234)
@@ -43,7 +46,7 @@
     mh_count = int(args.mh_count)
     mh_patience = int(args.mh_patience)
     mh_k = int(args.mh_count)
-    mh_maxiter = int(args.mh_mi)
+    mh_maxiter = int(args.mh_maxiter)
 
     os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
 
@@ -52,16 +55,24 @@
         print("****** SAMPLE FOR SUFFIX ", suffix, " ******")
         relgan = LoadRelgan(system=system, suffix=suffix, job=job)
 
-        f_out = "data/avatar/variants/" + system + "_relgan_" + str(suffix) + "_j" + str(job) + "_naive.txt"
+        if DATA_PATH is None:
+            f_out = os.path.join(WORK_PATH, "data", "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + "_naive.txt")
+        else:
+            f_out = os.path.join(DATA_PATH, "avatar", "variants",system + "_relgan_" + str(suffix) + "_j" + str(job) + "_naive.txt")
+
         print("Start NAIVE SAMPLING")
         gen_samples = relgan.generate(n_samples=n_samples)
         print("Generated samples - shape:", gen_samples.shape)
         print("Writing to file", f_out)
         writeToFile(relgan, f_out, gen_samples)
 
     elif strategy == "mh":
-        eval_path = "data/avatar/train_data/" + system + "_eval.txt"
-        f_out = "data/avatar/variants/" + system + "_relgan_" + str(suffix) + "_j" + str(job) + "_mh.txt"
+        if DATA_PATH is None:
+            eval_path = os.path.join(WORK_PATH, "data", "avatar", "train_data", system + "_eval.txt")
+            f_out = os.path.join(WORK_PATH, "data", "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + "_mh.txt")
+        else:
+            eval_path = os.path.join(DATA_PATH, "avatar", "train_data", system + "_eval.txt")
+            f_out = os.path.join(DATA_PATH, "avatar", "variants", system + "_relgan_" + str(suffix) + "_j" + str(job) + "_mh.txt")
 
         tf.reset_default_graph()
         print("****** SAMPLE FOR SUFFIX ", suffix, " ******")