Initial MLFlow setup

Jenkinsfile-train

@@ -34,7 +34,7 @@ pipeline {
         }
         stage('Train model') {
             steps {
-                sh "python train_model.py with 'epochs=${params.EPOCHS}' 'batch_size=${params.BATCHSIZE}'"
+                sh "python train_model.py -e ${params.EPOCHS} -b ${params.BATCHSIZE}"
             }
         }
         stage('Archive model and evaluate it') {

requirements.txt

@@ -5,3 +5,4 @@ numpy~=1.22.3
 matplotlib==3.5.2
 sacred==0.8.2
 pymongo==4.1.1
+mlflow==1.25.1

train_model.py

@@ -1,12 +1,9 @@
-import argparse
-
-import numpy as np
-import pandas as pd
 import torch
-from sacred.observers import FileStorageObserver, MongoObserver
+import mlflow
+import argparse
 from torch import nn
-from torch.utils.data import DataLoader, Dataset
+from torch.utils.data import DataLoader
-from sacred import Experiment
+from urllib.parse import urlparse
 
 from model import PlantsDataset, MLP, train, test
 
@@ -15,8 +12,23 @@ default_epochs = 5
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
+mlflow.set_tracking_uri("http://172.17.0.1:5000")
+mlflow.set_experiment("s444409")
+
+
+def setup_args():
+    args_parser = argparse.ArgumentParser(prefix_chars='-')
+    args_parser.add_argument('-b', '--batchSize', type=int, default=default_batch_size)
+    args_parser.add_argument('-e', '--epochs', type=int, default=default_epochs)
+
+    return args_parser.parse_args()
+
+
+if __name__ == "__main__":
+    args = setup_args()
+    batch_size = args.batchSize
+    epochs = args.epochs
 
-def main(batch_size, epochs, _run):
     print(f"Using {device} device")
 
     plant_test = PlantsDataset('data/Plant_1_Generation_Data.csv.test')
@@ -35,37 +47,33 @@ def main(batch_size, epochs, _run):
 
     loss_fn = nn.MSELoss()
     optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)
-    for t in range(epochs):
+    
-        print(f"Epoch {t + 1}\n-------------------------------")
-        train(train_dataloader, model, loss_fn, optimizer)
-        last_loss = test(test_dataloader, model, loss_fn)
-        _run.log_scalar('training.loss', last_loss, t)
     print("Done!")
 
     torch.save(model.state_dict(), './model_out')
     print("Model saved in ./model_out file.")
+    with mlflow.start_run() as run:
+        print("MLflow run experiment_id: {0}".format(run.info.experiment_id))
+        print("MLflow run artifact_uri: {0}".format(run.info.artifact_uri))
 
+        mlflow.log_param("batch_size", batch_size)
+        mlflow.log_param("epochs", epochs)
+        for t in range(epochs):
+            print(f"Epoch {t + 1}\n-------------------------------")
+            train(train_dataloader, model, loss_fn, optimizer)
+            last_loss = test(test_dataloader, model, loss_fn)
+            mlflow.log_metric("rmse", last_loss)
 
-def setup_experiment():
+        with torch.no_grad():
-    ex = Experiment('Predict power output for a given time')
+            preds = model(plant_test.x_train)
-    ex.observers.append(FileStorageObserver('sacred_runs'))
+            signature = mlflow.models.signature.infer_signature(plant_test.x_train.numpy(), preds.numpy())
-    ex.observers.append(MongoObserver(url='mongodb://admin:IUM_2021@172.17.0.1:27017',
+        tracking_url_type_store = urlparse(mlflow.get_tracking_uri()).scheme
-                                      db_name='sacred'))
+        if tracking_url_type_store != "file":
-    return ex
+            mlflow.pytorch.log_model(
-
+                model,
-
+                "s444409-power-plant-model",
-ex = setup_experiment()
+                registered_model_name="s444409PowerPlant",
-
+                signature=signature
-
+            )
-@ex.config
+        else:
-def experiment_config():
+            mlflow.pytorch.log_model(model, "s444409-power-plant-model", signature=signature)
-    batch_size = 64
-    epochs = 5
-
-
-@ex.automain
-def run(batch_size, epochs, _run):
-    main(batch_size, epochs, _run)
-
-
-ex.add_artifact('model_out')