multiexperiments code.

examples/scripts/replication/hyperparams/dmae_params.json

@@ -2,5 +2,5 @@
     "use_cov": false,
     "n_clusters": 10,
     "alpha": 100,
-    "dissimilarity": "kullback_leibler"
+    "dissimilarity": "euclidean"
 }

examples/scripts/replication/hyperparams/pretrain_params.json

@@ -2,6 +2,6 @@
     "iters": 100,
     "use_cov": false,
     "n_clusters": 10,
-    "epochs": 2,
+    "epochs": 500,
     "batch_size": 256
 }

examples/scripts/replication/hyperparams/train_params.json

@@ -1,4 +1,4 @@
 {
-    "epochs": 2,
+    "epochs": 300,
     "batch_size": 256
 }

examples/scripts/replication/logger.py

@@ -4,31 +4,34 @@
 class Logger:
     #TODOC
     def __init__(
-            self, models, metrics, 
-            datasets, dataset_name, 
+            self,
+            dataset_name, 
             evaluable_datasets = [
                 "pretrain", "clustering"
                 ],
+            metric_names = ["uACC", "NMI", "ARS"],
             save_path="./results/"
             ):
 
-        self.__models = models
-        self.__metrics = metrics
-        self.__datasets = datasets
         self.__dataset_name = dataset_name
         self.__evaluable_datasets = evaluable_datasets
+        self.__metric_names = metric_names
         self.__save_path = save_path
         self.__make_dataframes()
 
+    def add(self, models, metrics, datasets):
+        self.__models = models
+        self.__metrics = metrics
+        self.__datasets = datasets
+
     def __make_dataframes(self):
         #TODOC
         self.__dfs = {}
         for dataset in self.__evaluable_datasets:
-            metric_names = tuple(self.__metrics.keys())
             self.__dfs[dataset] = _pd.DataFrame(
                     columns=[
                         "iteration",
-                        *metric_names
+                        *self.__metric_names
                         ]
                     )
             self.__dfs[dataset].set_index("iteration", inplace=True)
@@ -44,7 +47,7 @@ def __call__(
         dataset.reset_gen()
         dataset = dataset()
 
-        y_true = self.__datasets["labels"]
+        y_true = self.__datasets["labels"].astype("int32")
         preds = model.predict(
                 dataset,
                 steps=self.__datasets["steps"]

examples/scripts/replication/main.py

@@ -17,13 +17,9 @@ def make_parser():
     parser.add_argument("--iters", type=int, default=5)
     return parser
 
-if __name__ == '__main__':
-    parser = make_parser()
-    args = parser.parse_args()
-    arguments = utils.json_arguments(args)
-
+def make_iteration(arguments, scorer, iteration):
     # make models
-    models = models.make_models(
+    cur_models = models.make_models(
             arguments["encoder_params"],
             arguments["decoder_params"],
             arguments["dmae_params"],
@@ -32,34 +28,42 @@ def make_parser():
             )
 
     # make datasets
-    datasets = datasets.make_datasets(
+    cur_datasets = datasets.make_datasets(
             **arguments["dataset_params"]
             )
 
     # make metrics
-    metrics = metrics.make_metrics()
+    cur_metrics = metrics.make_metrics()
 
     # logger
-    scorer = logger.Logger(
-            models, metrics, 
-            datasets, "mnist"
-            )
-
+    scorer.add(cur_models, cur_metrics, cur_datasets)
+
     # Pretrain
     arguments["pretrain_params"]\
-            ["steps_per_epoch"] = datasets["steps"]
+            ["steps_per_epoch"] = cur_datasets["steps"]
     arguments["train_params"]\
-            ["steps_per_epoch"] = datasets["steps"]
+            ["steps_per_epoch"] = cur_datasets["steps"]
 
     train.pretrain(
-            models, datasets,
+            cur_models, cur_datasets,
             arguments["pretrain_params"],
-            scorer, iteration=1,
+            scorer, iteration=iteration,
             dissimilarity=arguments["dmae_params"]["dissimilarity"]
             )
     train.train(
-            models, datasets,
+            cur_models, cur_datasets,
             arguments["train_params"],
-            scorer, iteration=1
+            scorer, iteration=iteration
             )
+
+if __name__ == '__main__':
+    parser = make_parser()
+    args = parser.parse_args()
+    arguments = utils.json_arguments(args)
+    scorer = logger.Logger(
+            arguments["dataset_params"]["dataset_name"]
+            )
+
+    for iteration in range(args.iters):
+        make_iteration(arguments, scorer, iteration)
     scorer.save()

examples/scripts/replication/results/mnist/clustering.csv

@@ -1,4 +1,5 @@
 iteration,uACC,NMI,ARS
-1,0.19112007783882784,0.1706133870001527,-0.17376299376086046
-mean,0.19112007783882784,0.1706133870001527,-0.17376299376086046
-std,0.0,0.0,0.0
+0,0.3249198717948718,0.2974064119317825,0.16677801889403499
+1,0.27691449175824173,0.25669355378788655,0.12515683728665603
+mean,0.3009171817765568,0.2770499828598345,0.14596742809034552
+std,0.024002690018315037,0.020356429071947985,0.02081059080368948

examples/scripts/replication/results/mnist/pretrain.csv

@@ -1,4 +1,5 @@
 iteration,uACC,NMI,ARS
-1,0.34450835622710624,0.3124251856194935,0.16642473909353817
-mean,0.34450835622710624,0.3124251856194935,0.16642473909353817
-std,0.0,0.0,0.0
+0,0.37102220695970695,0.3299324743100253,0.2076916216553968
+1,0.32822516025641024,0.26351401699037746,0.13510127504054154
+mean,0.3496236836080586,0.29672324565020136,0.1713964483479692
+std,0.021398523351648352,0.033209228659823925,0.03629517330742764

requirements_docker_gpu.txt

@@ -1,8 +1,9 @@
 absl-py==0.11.0
 astunparse==1.6.3
-cachetools==4.2.0
+cachetools==4.2.1
 certifi==2020.12.5
 chardet==4.0.0
+cycler==0.10.0
 flatbuffers==1.12
 gast==0.3.3
 google-auth==1.24.0
@@ -11,17 +12,24 @@ google-pasta==0.2.0
 grpcio==1.32.0
 h5py==2.10.0
 idna==2.10
+importlib-metadata==3.4.0
 joblib==1.0.0
 Keras-Preprocessing==1.1.2
+kiwisolver==1.3.1
 Markdown==3.3.3
-numpy==1.19.5
-pandas==1.1.5
 matplotlib==3.3.4
+numpy==1.19.5
 oauthlib==3.1.0
 opt-einsum==3.3.0
+pandas==1.1.5
+Pillow==8.1.0
 protobuf==3.14.0
 pyasn1==0.4.8
 pyasn1-modules==0.2.8
+pyparsing==2.4.7
+python-dateutil==2.8.1
+python-ternary==1.0.7
+pytz==2021.1
 requests==2.25.1
 requests-oauthlib==1.3.0
 rsa==4.7
@@ -30,14 +38,14 @@ scipy==1.5.4
 six==1.15.0
 tensorboard==2.4.1
 tensorboard-plugin-wit==1.8.0
-tensorflow-gpu==2.4.1
-tensorflow-addons==0.12.0
+tensorflow-addons==0.12.1
 tensorflow-estimator==2.4.0
+tensorflow-gpu==2.4.1
 termcolor==1.1.0
 threadpoolctl==2.1.0
 typeguard==2.10.0
 typing-extensions==3.7.4.3
-urllib3==1.26.2
+urllib3==1.26.3
 Werkzeug==1.0.1
 wrapt==1.12.1
-python-ternary==1.0.7
+zipp==3.4.0