Merge pull request #8 from milank94/model-training-update

Model training update

Makefile

@@ -9,3 +9,11 @@ run-eda:
 process-data:
 	@echo "Processing data"
 	@python data_processing/main.py
+
+train-model:
+	@echo "Training model"
+	@python model_training/main.py
+
+evaluate-model:
+	@echo "Evaluating model"
+	@python model_evaluation/model_evaluation.py

environment.yml

@@ -1,11 +1,14 @@
 name: motor-fault-classification
 channels:
+  - defaults
   - conda-forge
 dependencies:
+  - pip>=8.1.2
   - python=3.8.8
   - numpy=1.20.2
   - pandas=1.2.3
-  - keras=2.4.3
-  - matplotlib=3.4.1
   - scikit-learn=0.24.2
+  - matplotlib=3.4.1
   - scipy=1.6.3
+  - pip:
+    - tensorflow>=0.12.0rc1

model_evaluation/model_eval_config.py

@@ -0,0 +1,8 @@
+"""
+Any config specific to model training step goes here.
+"""
+
+from pathlib import Path
+
+# Output data path
+OUTPUT_DATA_DIR = Path('./output')

model_evaluation/model_evaluation.py

@@ -0,0 +1,39 @@
+from pathlib import Path
+import numpy as np
+import pandas as pd
+from sklearn.metrics import accuracy_score
+from tensorflow.keras.models import load_model
+from data_acquisition.main import get_save_data
+from data_processing.main import get_save_train_test_data
+import model_eval_config as config
+
+# Load the data
+dataset = get_save_data()
+
+# Process the data and train/test split
+train_test_data = get_save_train_test_data(dataset)
+
+# Prepare the data
+X_test = np.array(train_test_data.X_test)
+y_test = np.array(train_test_data.y_test)
+
+# Loading the model and checking accuracy on the test data
+model_path = config.OUTPUT_DATA_DIR / Path('best_model.pkl')
+model = load_model(model_path)
+test_preds = np.argmax(model.predict(X_test), axis=-1)
+print(accuracy_score(y_test, test_preds))
+
+# Comparing the actual values versus the predicted values
+data_dict = {
+    0: 'normal',
+    1: 'horizontal misalignment',
+    2: 'imbalance',
+    3: 'vertical misalignment',
+    4: 'overhang',
+    5: 'underhang'
+    }
+results = pd.DataFrame([y_test, test_preds]).T
+results.columns = ['Actual', 'Prediction']
+results.applymap(lambda x: data_dict[x])
+
+print(results)

model_training/main.py

@@ -1,34 +1,26 @@
-import pandas as pd
+from pathlib import Path
 import numpy as np
 import matplotlib.pyplot as plt
 import logging
-import os
-from data_acquisition.main import get_data
-from data_processing.main import get_train_test_data
+from data_acquisition.main import get_save_data
+from data_processing.main import get_save_train_test_data
 import model_training_config as config
 
-from keras.models import Sequential
-from keras.layers import Dense
-from keras.layers import LSTM
-
-from keras.models import load_model
-from keras.callbacks import ModelCheckpoint
-
-from sklearn.metrics import accuracy_score
-
-path_parent = os.path.dirname(os.getcwd())
-os.chdir(path_parent)
+from tensorflow.keras.models import Sequential
+from tensorflow.keras.layers import Dense
+from tensorflow.keras.layers import LSTM
+from tensorflow.keras.callbacks import ModelCheckpoint
 
 logging.basicConfig(level=logging.INFO)
 
-# Model Development
+config.OUTPUT_DATA_DIR.mkdir(parents=True, exist_ok=True)
 
+# Model Development
 # Load the data
-dataset = get_data()
+dataset = get_save_data()
 
 # Process the data and train/test split
-train_test_data = get_train_test_data(dataset)
-
+train_test_data = get_save_train_test_data(dataset)
 
 # Prepare the data
 X_train = np.array(train_test_data.X_train)
@@ -41,24 +33,25 @@
 model = Sequential()
 model.add(LSTM(config.LSTM_UNITS, input_shape=(X_train.shape[1], X_train.shape[2])))
 model.add(Dense(config.OUTPUT_SIZE, activation=config.ACTIVATION))
-model.summary()
-
+print(model.summary())
 
 # Training the model
-chk = ModelCheckpoint('best_model.pkl', monitor=config.MONITOR, save_best_only=True, mode='auto', verbose=1)
+model_path = config.OUTPUT_DATA_DIR / Path('best_model.pkl')
+chk = ModelCheckpoint(model_path, monitor=config.MONITOR, save_best_only=True, mode='auto', verbose=1)
 model.compile(loss=config.LOSS_FUNCTION, optimizer=config.OPTIMIZER, metrics=['accuracy'])
 hist = model.fit(
     X_train,
     y_train,
     epochs=config.EPOCHS,
     batch_size=int(X_train.shape[0]),
     callbacks=[chk],
-    validation_split=config.VAL_SPLOT
+    validation_split=config.VAL_SPLIT
     )
 
 # Model Validation
 # Plotting training and validation accuracy per epoch
-fig, axs = plt.subplots(nrows=1, figsize=(11, 9))
+_, axs = plt.subplots(nrows=1, figsize=(11, 9))
+file_location = config.OUTPUT_DATA_DIR / Path('plots/model_accuracy.png')
 plt.rcParams['font.size'] = '14'
 
 for label in (axs.get_xticklabels() + axs.get_yticklabels()):
@@ -71,16 +64,4 @@
 axs.set_ylabel('Accuracy', fontsize=14)
 axs.set_xlabel('Epoch', fontsize=14)
 plt.legend(['train', 'val'], loc='upper left')
-plt.show()
-
-# Loading the model and checking accuracy on the test data
-model = load_model('best_model.pkl')
-
-test_preds = model.predict_classes(X_test)
-accuracy_score(y_test, test_preds)
-
-# Comparing the actual values versus the predicted values
-data_dict = {0: 'normal', 1: 'horizontal misalignment', 2: 'imbalance'}
-results = pd.DataFrame([y_test, test_preds]).T
-results.columns = ['Actual', 'Prediction']
-results.applymap(lambda x: data_dict[x])
+plt.savefig(file_location)

model_training/model_training_config.py

@@ -2,21 +2,19 @@
 Any config specific to model training step goes here.
 """
 
+from pathlib import Path
+
 # General configuration
 MONITOR = 'val_loss'
 VAL_SPLIT = 0.2
 LSTM_UNITS = 100
 OPTIMIZER = 'adam'
 EPOCHS = 100
 
-# +
-# Binary classification
-# OUTPUT_SIZE = 1
-# ACTIVATION = 'sigmoid'
-# LOSS_FUNCTION = 'binary_crossentropy'
-# -
-
 # Multi-class classification
-OUTPUT_SIZE = 3
+OUTPUT_SIZE = 6
 ACTIVATION = 'softmax'
 LOSS_FUNCTION = 'sparse_categorical_crossentropy'
+
+# Output data path
+OUTPUT_DATA_DIR = Path('./output')