library version updates (#174)

Signed-off-by: s0nicboOm <[email protected]>

numaprom/default-configs/numalogic_config.yaml

@@ -11,7 +11,8 @@ model:
       - 16
     dropout_p: 0.25
 trainer:
-  max_epochs: 30
+  pltrainer_conf:
+    max_epochs: 30
 preprocess:
   - name: "StandardScaler"
 threshold:

numaprom/udf/inference.py

@@ -3,10 +3,10 @@
 from typing import Final
 
 from numalogic.config import NumalogicConf
-from numalogic.models.autoencoder import AutoencoderTrainer
 from numalogic.registry import ArtifactData, RedisRegistry, LocalLRUCache
 from numalogic.tools.data import StreamingDataset
 from numalogic.tools.exceptions import RedisRegistryError
+from numalogic.tools.trainer import TimeseriesTrainer
 from orjson import orjson
 from pynumaflow.mapper import Datum
 from torch.utils.data import DataLoader
@@ -32,7 +32,7 @@ def _run_inference(
     stream_data = payload.get_stream_array()
     stream_loader = DataLoader(StreamingDataset(stream_data, numalogic_conf.model.conf["seq_len"]))
 
-    trainer = AutoencoderTrainer()
+    trainer = TimeseriesTrainer()
     try:
         recon_err = trainer.predict(model, dataloaders=stream_loader)
     except Exception as err:

numaprom/udsink/train.py

@@ -4,7 +4,7 @@
 import numpy as np
 import pandas as pd
 from numalogic.config import PreprocessFactory, ModelInfo, ThresholdFactory, ModelFactory
-from numalogic.models.autoencoder import AutoencoderTrainer
+from numalogic.models.autoencoder import TimeseriesTrainer
 from numalogic.registry import RedisRegistry
 from numalogic.tools.data import StreamingDataset
 from numalogic.tools.exceptions import RedisRegistryError
@@ -43,7 +43,7 @@ def _train_model(uuid, x, model_cfg, trainer_cfg):
     model = model_factory.get_instance(model_cfg)
     dataset = StreamingDataset(x, model.seq_len)
 
-    trainer = AutoencoderTrainer(**trainer_cfg)
+    trainer = TimeseriesTrainer(**trainer_cfg.pltrainer_conf)
     trainer.fit(model, train_dataloaders=DataLoader(dataset, batch_size=64))
 
     LOGGER.debug(
@@ -121,7 +121,7 @@ def train(datums: list[Datum]) -> Responses:
             payload,
             metric_config,
             {"namespace": payload.composite_keys["namespace"]},
-            hours=metric_config.train_hours,
+            hours=metric_config.numalogic_conf.trainer.train_hours,
         )
         train_df = clean_data(train_df)
 

numaprom/udsink/train_rollout.py

@@ -6,7 +6,7 @@
 from numalogic.tools.exceptions import InvalidDataShapeError
 import pandas as pd
 from numalogic.config import PreprocessFactory, ModelInfo, ThresholdFactory, ModelFactory
-from numalogic.models.autoencoder import AutoencoderTrainer
+from numalogic.models.autoencoder import TimeseriesTrainer
 from numalogic.registry import RedisRegistry
 from numalogic.tools.data import StreamingDataset
 from numalogic.tools.exceptions import RedisRegistryError
@@ -57,7 +57,7 @@ def _train_model(uuid, x, model_cfg, trainer_cfg):
     model = model_factory.get_instance(model_cfg)
     dataset = StreamingDataset(x, model.seq_len)
 
-    trainer = AutoencoderTrainer(**trainer_cfg)
+    trainer = TimeseriesTrainer(**trainer_cfg.pltrainer_conf)
     trainer.fit(model, train_dataloaders=DataLoader(dataset, batch_size=64))
 
     LOGGER.debug(
@@ -145,7 +145,7 @@ def train_rollout(datums: Iterator[Datum]) -> Responses:
             metric_config,
             {"namespace": payload.composite_keys["namespace"]},
             return_labels=[hash_label],
-            hours=metric_config.train_hours,
+            hours=metric_config.numalogic_conf.trainer.train_hours,
         )
         try:
             train_df = clean_data(train_df, hash_label)

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "numalogic-prometheus"
-version = "0.6.0"
+version = "0.6.1dev0"
 description = "ML inference on numaflow using numalogic on Prometheus metrics"
 authors = ["Numalogic developers"]
 packages = [{ include = "numaprom" }]
@@ -19,14 +19,14 @@ repository = "https://github.com/numaproj/numalogic-prometheus"
 
 [tool.poetry.dependencies]
 python = "~3.10"
-redis = {extras = ["hiredis"], version = "^4.5" }
+redis = {extras = ["hiredis"], version = "^5.0.1" }
 pynumaflow = "~0.5"
-numalogic = {version = "~0.5", extras = ["redis"]}
+numalogic = {version = "0.6.1.dev5", extras = ["redis"]}
 orjson = "^3.8.4"
 omegaconf = "^2.3.0"
 watchdog = "^3.0.0"
 loguru = "^0.7.0"
-prometheus-client = "^0.17"
+prometheus-client = "^0.19"
 
 [tool.poetry.group.dev]
 optional = true

tests/resources/configs/numalogic_config.yaml

@@ -11,7 +11,8 @@ model:
       - 16
     dropout_p: 0.25
 trainer:
-  max_epochs: 5
+  pltrainer_conf:
+    max_epochs: 5
 preprocess:
   - name: "LogTransformer"
     stateful: false

tests/udf/test_inference.py

@@ -3,7 +3,7 @@
 from unittest.mock import patch, Mock
 
 from freezegun import freeze_time
-from numalogic.models.autoencoder import AutoencoderTrainer
+from numalogic.models.autoencoder import TimeseriesTrainer
 from numalogic.registry import RedisRegistry
 from orjson import orjson
 from pynumaflow.mapper import Messages
@@ -52,7 +52,7 @@ def test_inference(self):
 
     @freeze_time("2022-02-20 12:00:00")
     @patch.object(RedisRegistry, "load", Mock(return_value=return_mock_lstmae()))
-    @patch.object(AutoencoderTrainer, "predict", Mock(side_effect=RuntimeError))
+    @patch.object(TimeseriesTrainer, "predict", Mock(side_effect=RuntimeError))
     def test_inference_err(self):
         for msg in self.inference_input:
             _out = inference([""], get_datum(msg.value))