Feature/final exam

api/app.py

@@ -20,18 +20,56 @@
 def hello_world():
     return 'Digit Classification Model Deployment'
 
-@app.route("/predict",methods =['POST'])
-def predict():
-    model_path = "./models/svm_C : 0.1_gamma : 0.0001.joblib"
-    model = load(model_path)
+# @app.route("/predict",methods =['POST'])
+# def predict():
+#     model_path = "./models/svm_C : 0.1_gamma : 0.0001.joblib"
+#     model = load(model_path)
+#     data = request.get_json()
+#     data['image'] = np.array(data['image']).astype(float)
+
+#     image_data = data['image'].reshape(1, -1)
+
+#     _,prediction,_ = predict_and_eval(model , image_data,np.array([0]))
+#     return ({"Prediction": str(prediction[0])})
+
+# Global variables
+svm_model = None
+lr_model = None
+tree_model = None
+
+def load_model():
+    global svm_model, lr_model, tree_model
+
+    # Load SVM model
+    svm_model = load("./models/svm_C : 0.1_gamma : 0.0001.joblib")
+
+    # Load Logistic Regression model
+    lr_model = load("./models/M22AIE212_lr_lbfgs.joblib")
+
+    # Load Decision Tree model
+    tree_model = load("./models/decision_tree_criterion : entropy_max_depth : 20.joblib")
+
+# Call load_model() when the application starts
+load_model()
+
+@app.route('/predict/<model_type>', methods=['POST'])
+def predict(model_type):
+
+    if model_type == "svm" : 
+        model = svm_model
+    elif model_type == "decision_tree":
+        model = tree_model
+    elif model_type == "lr" :
+        model = lr_model
+
     data = request.get_json()
     data['image'] = np.array(data['image']).astype(float)
 
     image_data = data['image'].reshape(1, -1)
 
     _,prediction,_ = predict_and_eval(model , image_data,np.array([0]))
-    return ({"Prediction": str(prediction[0])})
+    return ({f"{model_type.upper()} Prediction": str(prediction[0])})
 
-if __name__ =="__main__" :
-    app.run(host = "0.0.0.0",port = 80)
+# if __name__ =="__main__" :
+#     app.run(host = "0.0.0.0",port = 80)
 

config.json

@@ -9,5 +9,10 @@
 
         "criterion" : ["gini","entropy"],
         "max_depth" : [10,15,20,30,40,50,60,70,80,90,100]
+    },
+
+    "lr" : {
+
+        "solver" : ["lbfgs", "liblinear","newton-cg", "newton-cholesky", "sag", "saga"]
     }
 }

docker/DependencyDockerfile

@@ -0,0 +1,4 @@
+FROM python:3.9.17
+WORKDIR /digits
+COPY . /digits/
+RUN pip3 install --no-cache-dir -r /digits/requirements.txt

docker/FinalDockerfile

@@ -0,0 +1,3 @@
+FROM base:latest
+RUN mkdir -p models
+CMD ["pytest"]

docker_push_helper.sh

@@ -0,0 +1,2 @@
+az acr build --file docker/DependencyDockerfile --registry sagarmlops23 --image base .
+az acr build --file docker/FinalDockerfile --registry sagarmlops23 --image digits .

exp.py

@@ -55,7 +55,7 @@
 parser.add_argument('-model',
             '--model_type',
             type = str,
-            default = 'decision_tree svm',
+            default = 'decision_tree svm lr',
             help = 'name of model')
 parser.add_argument('-config',
             '--config_path',
@@ -96,8 +96,15 @@
 param_grid = params_dict['decision_tree']
 list_of_all_param_combination_dt = ParameterGrid(param_grid)
 
+
+# Model 3 : Logistic Regression
+param_grid = params_dict['lr']
+list_of_all_param_combination_lr = ParameterGrid(param_grid)
+
+
 hparam_dict = {'svm' : list_of_all_param_combination_svm ,
-                    'decision_tree' : list_of_all_param_combination_dt }
+                    'decision_tree' : list_of_all_param_combination_dt,
+                    'lr'  : list_of_all_param_combination_lr }
 
 
 print("Iterating for different test and dev size : ")
@@ -147,22 +154,22 @@
 
     print(f"Average Dev Accuracy of {model_type}: ",sum(all_dev_accuracy)/max_run)
 
-print("Confusion Matrix Between SVM  Decision Tree predictions for Test Data : ")
-y_svm_pred = pd.Series(predictions['svm'], name='Production(SVM)')
-y_dt_pred= pd.Series(predictions['decision_tree'], name='Candidate(DT)')
-df_confusion = pd.crosstab(y_svm_pred, y_dt_pred)
-print(df_confusion)
+# print("Confusion Matrix Between SVM  Decision Tree predictions for Test Data : ")
+# y_svm_pred = pd.Series(predictions['svm'], name='Production(SVM)')
+# y_dt_pred= pd.Series(predictions['decision_tree'], name='Candidate(DT)')
+# df_confusion = pd.crosstab(y_svm_pred, y_dt_pred)
+# print(df_confusion)
 
-# SVM test predictions Comparison with actuals
-y_check_svm = y_svm_pred == actuals['svm']
+# # SVM test predictions Comparison with actuals
+# y_check_svm = y_svm_pred == actuals['svm']
 
-# Decision tree test Comparison with actuals
-y_check_dt = y_dt_pred == actuals['decision_tree']
+# # Decision tree test Comparison with actuals
+# y_check_dt = y_dt_pred == actuals['decision_tree']
 
-y_svm = pd.Series(y_check_svm, name='Production(SVM)')
-y_dt = pd.Series(y_check_dt, name='Candidate(DT)')
-df_confusion = pd.crosstab(y_svm, y_dt)
-print(df_confusion)
+# y_svm = pd.Series(y_check_svm, name='Production(SVM)')
+# y_dt = pd.Series(y_check_dt, name='Candidate(DT)')
+# df_confusion = pd.crosstab(y_svm, y_dt)
+# print(df_confusion)
 
-print("F1 Score Production (SVM)" , f1_score(actuals['svm'], y_svm_pred, average='macro'))
-print("F1 Score Candidate (DT)" , f1_score(actuals['decision_tree'], y_dt_pred, average='macro'))
+# print("F1 Score Production (SVM)" , f1_score(actuals['svm'], y_svm_pred, average='macro'))
+# print("F1 Score Candidate (DT)" , f1_score(actuals['decision_tree'], y_dt_pred, average='macro'))

test_utils.py

@@ -2,6 +2,7 @@
 import os
 from sklearn.model_selection import ParameterGrid
 import numpy as np
+from sklearn.linear_model import LogisticRegression
 np.random.seed(42)
 # Creating models directory
 if os.path.exists("models"):
@@ -26,6 +27,14 @@ def dummy_hyper_parameters():
 
     return gamma_list , C_ranges,h_params_combinations
 
+def dummy_hyper_lr_parameters():
+
+    solver = ["lbfgs", "liblinear","newton-cg", "newton-cholesky", "sag", "saga"]
+    param_grid = {'solver': solver}
+    h_params_combinations =ParameterGrid(param_grid)
+
+    return h_params_combinations
+
 def create_dummy_data() :
 
     X , y = read_digits()
@@ -78,7 +87,105 @@ def test_get_root() :
     assert response.get_data() == b'Digit Classification Model Deployment'
 
 
-def test_post_predict():
+# def test_post_predict():
+
+#     X,y = read_digits()
+#     X_reshape = preprocess_data(X)
+
+#     ## Digit 0 test
+#     indices_digit_0 = np.where(y == 0)[0]
+#     index_digit_0 = np.random.choice(indices_digit_0)
+#     image_data_0 = list(X_reshape[index_digit_0].astype(str))
+
+#     response = app.test_client().post("/predict", json={"image":image_data_0})
+
+#     ## Status Code Check
+#     assert response.status_code == 200 
+#     assert response.get_json()['Prediction'] == '0'
+
+#     ## Digit 1 test
+#     indices_digit_1 = np.where(y == 1)[0]
+#     index_digit_1 = np.random.choice(indices_digit_1)
+#     image_data_1 = list(X_reshape[index_digit_1].astype(str))
+#     response = app.test_client().post("/predict", json={"image":image_data_1})
+#     assert response.get_json()['Prediction'] == '1'
+
+#     ## Digit 2 test
+#     indices_digit_2 = np.where(y == 2)[0]
+#     index_digit_2 = np.random.choice(indices_digit_2)
+#     image_data_2 = list(X_reshape[index_digit_2].astype(str))
+#     response = app.test_client().post("/predict", json={"image":image_data_2})
+#     assert response.get_json()['Prediction'] == '2'
+
+#     ## Digit 3 test
+#     indices_digit_3 = np.where(y == 3)[0]
+#     index_digit_3 = np.random.choice(indices_digit_3)
+#     image_data_3 = list(X_reshape[index_digit_3].astype(str))
+#     response = app.test_client().post("/predict", json={"image":image_data_3})
+#     assert response.get_json()['Prediction'] == '3'
+
+#     ## Digit 4 test
+#     indices_digit_4 = np.where(y == 4)[0]
+#     index_digit_4 = np.random.choice(indices_digit_4)
+#     image_data_4 = list(X_reshape[index_digit_4].astype(str))
+#     response = app.test_client().post("/predict", json={"image":image_data_4})
+#     assert response.get_json()['Prediction'] == '4'
+
+#     ## Digit 5 test
+#     indices_digit_5 = np.where(y == 5)[0]
+#     index_digit_5 = np.random.choice(indices_digit_5)
+#     image_data_5 = list(X_reshape[index_digit_5].astype(str))
+#     response = app.test_client().post("/predict", json={"image":image_data_5})
+#     assert response.get_json()['Prediction'] == '5'
+
+#     ## Digit 6 test
+#     indices_digit_6 = np.where(y == 6)[0]
+#     index_digit_6 = np.random.choice(indices_digit_6)
+#     image_data_6 = list(X_reshape[index_digit_6].astype(str))
+#     response = app.test_client().post("/predict", json={"image":image_data_6})
+#     assert response.get_json()['Prediction'] == '6'
+
+#     ## Digit 7 test
+#     indices_digit_7 = np.where(y == 7)[0]
+#     index_digit_7 = np.random.choice(indices_digit_7)
+#     image_data_7 = list(X_reshape[index_digit_7].astype(str))
+#     response = app.test_client().post("/predict", json={"image":image_data_7})
+#     assert response.get_json()['Prediction'] == '7'
+
+#     ## Digit 8 test
+#     indices_digit_8 = np.where(y == 8)[0]
+#     index_digit_8 = np.random.choice(indices_digit_8)
+#     image_data_8 = list(X_reshape[index_digit_8].astype(str))
+#     response = app.test_client().post("/predict", json={"image":image_data_8})
+#     assert response.get_json()['Prediction'] == '8'
+
+#     ## Digit 9 test
+#     indices_digit_9 = np.where(y == 9)[0]
+#     index_digit_9 = np.random.choice(indices_digit_9)
+#     image_data_9 = list(X_reshape[index_digit_9].astype(str))
+#     response = app.test_client().post("/predict", json={"image":image_data_9})
+#     assert response.get_json()['Prediction'] == '9'
+
+def test_model_lr_loading() :
+    X_train,y_train,X_dev,y_dev = create_dummy_data()
+    list_of_all_param_combination = dummy_hyper_lr_parameters() 
+    _ ,best_model_path , _ = tune_hparams(X_train,y_train,X_dev,y_dev,list_of_all_param_combination,model_type = 'lr')
+
+    loaded_model = load(best_model_path)
+    assert isinstance(loaded_model, LogisticRegression), f"Model loaded from {best_model_path} is not a Logistic Regression model"
+
+
+
+def test_model_lr_loading_solver() :
+    X_train,y_train,X_dev,y_dev = create_dummy_data()
+    list_of_all_param_combination = dummy_hyper_lr_parameters() 
+    _ ,best_model_path , _ = tune_hparams(X_train,y_train,X_dev,y_dev,list_of_all_param_combination,model_type = 'lr')
+
+    loaded_model = load(best_model_path)
+
+    assert loaded_model.solver == best_model_path.split("_")[-1].split(".")[0].split(":")[-1].strip(), f"Solver name in the model file name does not match the solver used in the loaded model ({loaded_model.solver})"
+
+def test_post_predict_svm():
 
     X,y = read_digits()
     X_reshape = preprocess_data(X)
@@ -88,71 +195,37 @@ def test_post_predict():
     index_digit_0 = np.random.choice(indices_digit_0)
     image_data_0 = list(X_reshape[index_digit_0].astype(str))
 
-    response = app.test_client().post("/predict", json={"image":image_data_0})
+    response = app.test_client().post("/predict/svm", json={"image":image_data_0})
 
     ## Status Code Check
     assert response.status_code == 200 
-    assert response.get_json()['Prediction'] == '0'
-
-    ## Digit 1 test
-    indices_digit_1 = np.where(y == 1)[0]
-    index_digit_1 = np.random.choice(indices_digit_1)
-    image_data_1 = list(X_reshape[index_digit_1].astype(str))
-    response = app.test_client().post("/predict", json={"image":image_data_1})
-    assert response.get_json()['Prediction'] == '1'
-
-    ## Digit 2 test
-    indices_digit_2 = np.where(y == 2)[0]
-    index_digit_2 = np.random.choice(indices_digit_2)
-    image_data_2 = list(X_reshape[index_digit_2].astype(str))
-    response = app.test_client().post("/predict", json={"image":image_data_2})
-    assert response.get_json()['Prediction'] == '2'
-
-    ## Digit 3 test
-    indices_digit_3 = np.where(y == 3)[0]
-    index_digit_3 = np.random.choice(indices_digit_3)
-    image_data_3 = list(X_reshape[index_digit_3].astype(str))
-    response = app.test_client().post("/predict", json={"image":image_data_3})
-    assert response.get_json()['Prediction'] == '3'
-
-    ## Digit 4 test
-    indices_digit_4 = np.where(y == 4)[0]
-    index_digit_4 = np.random.choice(indices_digit_4)
-    image_data_4 = list(X_reshape[index_digit_4].astype(str))
-    response = app.test_client().post("/predict", json={"image":image_data_4})
-    assert response.get_json()['Prediction'] == '4'
-
-    ## Digit 5 test
-    indices_digit_5 = np.where(y == 5)[0]
-    index_digit_5 = np.random.choice(indices_digit_5)
-    image_data_5 = list(X_reshape[index_digit_5].astype(str))
-    response = app.test_client().post("/predict", json={"image":image_data_5})
-    assert response.get_json()['Prediction'] == '5'
-
-    ## Digit 6 test
-    indices_digit_6 = np.where(y == 6)[0]
-    index_digit_6 = np.random.choice(indices_digit_6)
-    image_data_6 = list(X_reshape[index_digit_6].astype(str))
-    response = app.test_client().post("/predict", json={"image":image_data_6})
-    assert response.get_json()['Prediction'] == '6'
-
-    ## Digit 7 test
-    indices_digit_7 = np.where(y == 7)[0]
-    index_digit_7 = np.random.choice(indices_digit_7)
-    image_data_7 = list(X_reshape[index_digit_7].astype(str))
-    response = app.test_client().post("/predict", json={"image":image_data_7})
-    assert response.get_json()['Prediction'] == '7'
-
-    ## Digit 8 test
-    indices_digit_8 = np.where(y == 8)[0]
-    index_digit_8 = np.random.choice(indices_digit_8)
-    image_data_8 = list(X_reshape[index_digit_8].astype(str))
-    response = app.test_client().post("/predict", json={"image":image_data_8})
-    assert response.get_json()['Prediction'] == '8'
-
-    ## Digit 9 test
-    indices_digit_9 = np.where(y == 9)[0]
-    index_digit_9 = np.random.choice(indices_digit_9)
-    image_data_9 = list(X_reshape[index_digit_9].astype(str))
-    response = app.test_client().post("/predict", json={"image":image_data_9})
-    assert response.get_json()['Prediction'] == '9'
+
+def test_post_predict_lr():
+
+    X,y = read_digits()
+    X_reshape = preprocess_data(X)
+
+    ## Digit 0 test
+    indices_digit_0 = np.where(y == 0)[0]
+    index_digit_0 = np.random.choice(indices_digit_0)
+    image_data_0 = list(X_reshape[index_digit_0].astype(str))
+
+    response = app.test_client().post("/predict/lr", json={"image":image_data_0})
+
+    ## Status Code Check
+    assert response.status_code == 200 
+
+def test_post_predict_decision_tree():
+
+    X,y = read_digits()
+    X_reshape = preprocess_data(X)
+
+    ## Digit 0 test
+    indices_digit_0 = np.where(y == 0)[0]
+    index_digit_0 = np.random.choice(indices_digit_0)
+    image_data_0 = list(X_reshape[index_digit_0].astype(str))
+
+    response = app.test_client().post("/predict/decision_tree", json={"image":image_data_0})
+
+    ## Status Code Check
+    assert response.status_code == 200