fix: incorporate smdebug_ruleconfigs pkg until availability in PyPI (#291)

Author: knakad

setup.py

@@ -102,4 +102,5 @@ def read_version():
     install_requires=required_packages,
     extras_require=extras,
     entry_points={"console_scripts": ["sagemaker=sagemaker.cli.main:main"]},
+    include_package_data=True,  # TODO-reinvent-2019 [knakad]: Remove after rule_configs is in PyPI
 )

src/sagemaker/debugger.py

@@ -21,8 +21,7 @@
 """
 from __future__ import absolute_import
 
-# TODO-reinvent-2019 [knakad]: Uncomment this once PyPI integration is complete post-re:Invent-2019
-# import smdebug_rulesconfig as rule_configs  # noqa: F401 # pylint: disable=unused-import
+import smdebug_rulesconfig as rule_configs  # noqa: F401 # pylint: disable=unused-import
 
 
 RULES_ECR_REPO_NAME = "sagemaker-debugger-rules"

src/smdebug_rulesconfig/__init__.py

@@ -0,0 +1,22 @@
+from __future__ import absolute_import
+
+from .builtin_rules import vanishing_gradient  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import all_zero  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import check_input_images  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import similar_across_runs  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import weight_update_ratio  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import exploding_tensor  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import unchanged_tensor  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import loss_not_decreasing  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import dead_relu  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import confusion  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import class_imbalance  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import overfit  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import tree_depth  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import tensor_variance  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import overtraining  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import poor_weight_initialization  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import saturated_activation  # noqa: F401 # pylint: disable=unused-import
+from .builtin_rules import nlp_sequence_ratio  # noqa: F401 # pylint: disable=unused-import
+
+from ._collections import get_collection  # noqa: F401 # pylint: disable=unused-import

src/smdebug_rulesconfig/_collections.py

@@ -0,0 +1,7 @@
+from __future__ import absolute_import
+
+from ._utils import _get_collection_config
+
+
+def get_collection(collection_name):
+    return _get_collection_config(collection_name)

src/smdebug_rulesconfig/_constants.py

@@ -0,0 +1,5 @@
+from __future__ import absolute_import
+
+RULE_CONFIG_FILE = "rule_config_jsons/ruleConfigs.json"
+RULE_GROUPS_CONFIG_FILE = "rule_config_jsons/ruleGroups.json"
+COLLECTION_CONFIG_FILE = "rule_config_jsons/collections.json"

src/smdebug_rulesconfig/_ruleGroups.py

@@ -0,0 +1,30 @@
+from __future__ import absolute_import
+
+from enum import Enum
+from ._utils import _get_config_for_group
+
+
+class MXNET(Enum):
+    SIMPLE = ["VanishingGradient", "LossNotDecreasing", "WeightUpdateRatio"]
+    ALL = []
+
+
+class TENSORFLOW(Enum):
+    SIMPLE = ["VanishingGradient", "LossNotDecreasing", "WeightUpdateRatio"]
+    ALL = []
+
+
+class PYTORCH(Enum):
+    SIMPLE = ["VanishingGradient", "LossNotDecreasing", "WeightUpdateRatio"]
+    ALL = []
+
+
+class XGBOOST(Enum):
+    SIMPLE = ["TreeDepth", "ClassImbalance"]
+    ALL = []
+
+
+def get_rule_groups(ruleEnum):
+    ruleEnumVal = ruleEnum.value
+    rules_config = _get_config_for_group(ruleEnumVal)
+    return rules_config

src/smdebug_rulesconfig/_utils.py

@@ -0,0 +1,60 @@
+from __future__ import absolute_import
+
+import json
+import os
+from ._constants import RULE_CONFIG_FILE, RULE_GROUPS_CONFIG_FILE, COLLECTION_CONFIG_FILE
+
+
+def _get_rule_config(rule_name):
+    rule_config = None
+    print(rule_name)
+    config_file_path = os.path.dirname(os.path.abspath(__file__)) + "/" + RULE_CONFIG_FILE
+    print(config_file_path)
+
+    if os.path.exists(config_file_path):
+        with open(config_file_path) as json_data:
+            configs = json.load(json_data)
+            if rule_name in configs:
+                rule_config = configs[rule_name]
+    return rule_config
+
+
+def _get_rule_list(framework, type):
+    rules_list = []
+
+    config_file_path = os.path.dirname(os.path.abspath(__file__)) + "/" + RULE_GROUPS_CONFIG_FILE
+
+    if os.path.exists(config_file_path):
+        with open(config_file_path) as json_data:
+            configs = json.load(json_data)
+            if framework in configs:
+                if type in configs[framework]:
+                    rules_list = configs[framework][type]
+    return rules_list
+
+
+def _get_config_for_group(rules):
+    rules_config = []
+
+    config_file_path = os.path.dirname(os.path.abspath(__file__)) + "/" + RULE_CONFIG_FILE
+
+    if os.path.exists(config_file_path):
+        with open(config_file_path) as json_data:
+            configs = json.load(json_data)
+            for rule_name in rules:
+                if rule_name in configs:
+                    rules_config.append(configs[rule_name])
+    return rules_config
+
+
+def _get_collection_config(collection_name):
+    coll_config = None
+
+    config_file_path = os.path.dirname(os.path.abspath(__file__)) + "/" + COLLECTION_CONFIG_FILE
+
+    if os.path.exists(config_file_path):
+        with open(config_file_path) as json_data:
+            configs = json.load(json_data)
+            if collection_name in configs:
+                coll_config = configs[collection_name]
+    return coll_config

src/smdebug_rulesconfig/_version.py

@@ -0,0 +1,3 @@
+from __future__ import absolute_import
+
+__version__ = "0.1.2"

src/smdebug_rulesconfig/builtin_rules.py

@@ -0,0 +1,92 @@
+from __future__ import absolute_import
+from ._utils import _get_rule_config
+
+
+def vanishing_gradient():
+    rule_config = _get_rule_config("VanishingGradient")
+    return rule_config
+
+
+def similar_across_runs():
+    rule_config = _get_rule_config("SimilarAcrossRuns")
+    return rule_config
+
+
+def weight_update_ratio():
+    rule_config = _get_rule_config("WeightUpdateRatio")
+    return rule_config
+
+
+def all_zero():
+    rule_config = _get_rule_config("AllZero")
+    return rule_config
+
+
+def exploding_tensor():
+    rule_config = _get_rule_config("ExplodingTensor")
+    return rule_config
+
+
+def unchanged_tensor():
+    rule_config = _get_rule_config("UnchangedTensor")
+    return rule_config
+
+
+def loss_not_decreasing():
+    rule_config = _get_rule_config("LossNotDecreasing")
+    return rule_config
+
+
+def check_input_images():
+    rule_config = _get_rule_config("CheckInputImages")
+    return rule_config
+
+
+def dead_relu():
+    rule_config = _get_rule_config("DeadRelu")
+    return rule_config
+
+
+def confusion():
+    rule_config = _get_rule_config("Confusion")
+    return rule_config
+
+
+def tree_depth():
+    rule_config = _get_rule_config("TreeDepth")
+    return rule_config
+
+
+def class_imbalance():
+    rule_config = _get_rule_config("ClassImbalance")
+    return rule_config
+
+
+def overfit():
+    rule_config = _get_rule_config("Overfit")
+    return rule_config
+
+
+def tensor_variance():
+    rule_config = _get_rule_config("TensorVariance")
+    return rule_config
+
+
+def overtraining():
+    rule_config = _get_rule_config("Overtraining")
+    return rule_config
+
+
+def poor_weight_initialization():
+    rule_config = _get_rule_config("PoorWeightInitialization")
+    return rule_config
+
+
+def saturated_activation():
+    rule_config = _get_rule_config("SaturatedActivation")
+    return rule_config
+
+
+def nlp_sequence_ratio():
+    rule_config = _get_rule_config("NLPSequenceRatio")
+    return rule_config

src/smdebug_rulesconfig/rule_config_jsons/collections.json

@@ -0,0 +1,42 @@
+{
+  "weights": {
+    "CollectionName": "weights",
+    "CollectionParameters": {
+          "save_interval": "500"
+    }
+  },
+  "gradients": {
+    "CollectionName": "gradients",
+    "CollectionParameters": {
+          "save_interval": "500"
+    }
+  },
+  "losses": {
+    "CollectionName": "losses",
+    "CollectionParameters": {
+        "save_interval": "500"
+    }
+  },
+  "input_image": {
+    "CollectionName": "input_image",
+    "CollectionParameters": {
+      "include_regex": ".*hybridsequential0_input_0",
+      "save_interval" : "500"
+    }
+  },
+  "relu_output": {
+    "include_regex": ".*relu_output",
+    "save_interval": "500"
+  },
+  "relu_input": {
+    "include_regex": ".*relu_input",
+    "save_interval": "500"
+  },
+  "input_sequence": {
+    "CollectionName": "input_sequence",
+    "CollectionParameters": {
+      "include_regex": ".*embedding0_input_0",
+      "save_interval" : "500"
+    }
+  }
+}