add mlflow

README.md

@@ -12,14 +12,19 @@ and prints a short summary of the dataset as well as its subsets.
 ### Zadanie 2
 add Jenkinsfiles and mock data preprocessing
 
-### Zadanie 5
-added lab4 file with new python script and updated Dockerfile.
+### Zadanie 4
+added lab4 directory with new python script and updated Dockerfile.
 The container downloads the dataset and installs software needed,
 then trains and evaluates model on the dataset.
 Loss and accuracy are saved to test_eval.txt file.
 
+### Zadanie 5
+added lab5 directory with scripts
+
 ### Zadanie 6
 added create, train, eval directories in lab5
 
+### Zadanie 7
+updated contents of lab5/train directory
 
 ium01.ipynb is a notebook used to develop previously mentioned scripts.

lab5/eval/Jenkinsfile

@@ -55,6 +55,7 @@ pipeline {
                 stage('archive artifact') {
                     steps {
                         archiveArtifacts 'eval.csv'
+                        archiveArtifacts 'plot.png'
                     }
                 }
             }

lab5/eval/eval.py

@@ -1,9 +1,6 @@
 import csv
+import matplotlib.pyplot as plt
 import pandas as pd
-import seaborn as sns
-import sys
-import tensorflow
-from tensorflow.keras import layers
 from tensorflow.keras.models import load_model
 
 
@@ -20,3 +17,11 @@ with open('eval.csv', 'a', newline='') as fp:
     wr = csv.writer(fp, dialect='excel')
     wr.writerow(results)
 
+metrics = pd.read_csv('eval.csv', header=None, names=['loss', 'accuracy'])
+
+fig = plt.figure()
+plt.plot(metrics.accuracy)
+plt.ylabel('Accuracy')
+plt.xlabel('Build no.')
+
+fig.savefig('plot.png')

lab5/train/MLproject

@@ -0,0 +1,8 @@
+name: s470607
+docker_env:
+    image: kubakonieczny/ium:train-sacred-mlflow
+entry_points:
+    main:
+        parameters:
+            learning_rate: (type: float, default: 0.0003)
+        comment: "python3 train.py {learning_rate}"

lab5/train/requirements.txt

@@ -3,3 +3,4 @@ pandas
 tensorflow
 keras==2.3.1
 sacred
+mlflow

lab5/train/train.py

@@ -1,4 +1,5 @@
 from datetime import datetime
+import mlflow
 import pandas as pd
 from sacred import Experiment
 from sacred.observers import FileStorageObserver, MongoObserver
@@ -19,34 +20,37 @@ def my_config():
 
 @ex.capture
 def prepare_train_model(learning_rate, _run):
-    _run.info["prepare_model"] = str(datetime.now())
+    with mlflow.start_run():
+        _run.info["prepare_model"] = str(datetime.now())
 
-    X_train = pd.read_csv('train.csv')
-    X_valid = pd.read_csv('valid.csv')
+        X_train = pd.read_csv('train.csv')
+        X_valid = pd.read_csv('valid.csv')
 
-    Y_train = X_train.pop('stabf')
-    Y_train = pd.get_dummies(Y_train)
+        Y_train = X_train.pop('stabf')
+        Y_train = pd.get_dummies(Y_train)
 
-    Y_valid = X_valid.pop('stabf')
-    Y_valid = pd.get_dummies(Y_valid)
+        Y_valid = X_valid.pop('stabf')
+        Y_valid = pd.get_dummies(Y_valid)
 
-    model = tensorflow.keras.Sequential([
-        layers.Input(shape=(12,)),
-        layers.Dense(32),
-        layers.Dense(16),
-        layers.Dense(2, activation='softmax')
-    ])
+        model = tensorflow.keras.Sequential([
+            layers.Input(shape=(12,)),
+            layers.Dense(32),
+            layers.Dense(16),
+            layers.Dense(2, activation='softmax')
+        ])
 
-    model.compile(
-        loss=tensorflow.keras.losses.BinaryCrossentropy(),
-        optimizer=tensorflow.keras.optimizers.Adam(learning_rate=learning_rate),
-        metrics=[tensorflow.keras.metrics.BinaryAccuracy()])
+        model.compile(
+            loss=tensorflow.keras.losses.BinaryCrossentropy(),
+            optimizer=tensorflow.keras.optimizers.Adam(learning_rate=learning_rate),
+            metrics=[tensorflow.keras.metrics.BinaryAccuracy()])
 
-    history = model.fit(X_train, Y_train, epochs=2, validation_data=(X_valid, Y_valid))
+        history = model.fit(X_train, Y_train, epochs=2, validation_data=(X_valid, Y_valid))
 
-    model.save('grid-stability-dense.h5')
+        model.save('grid-stability-dense.h5')
 
-    _run.info['history'] = str(history.history['loss'][-1])
+        _run.info['history'] = str(history.history['loss'][-1])
+        mlflow.log_metric('loss', history.history['loss'][-1])
+        mlflow.log_param('learning_rate', learning_rate)
 
 
 @ex.main