import gradio as gr import numpy as np from skimage.feature import hog import pickle import cv2 loading_errors = {} def load_pickle_file(file_name): try: with open(file_name, "rb") as f: loaded_file = pickle.load(f) return loaded_file except pickle.UnpicklingError: error_msg = f"{file_name} is not a valid pickle. Run `git lfs pull`." print(error_msg, flush=True) loading_errors[file_name] = error_msg return None except Exception as exc: error_msg = f"Could not load {file_name}: {exc}" print(error_msg, flush=True) loading_errors[file_name] = error_msg return None # def build_knn_from_pca_package(package): # if package is None: # return None # required_keys = {"best_k", "weights", "X_train", "y_train", "pca_mean", "pca_components"} # missing_keys = required_keys.difference(package) # if missing_keys: # print(f"knn_pca.pkl is missing keys: {sorted(missing_keys)}") # return None # knn = MultimodeKNN(k=package["best_k"], weights=package["weights"]) # knn.fit(package["X_train"], package["y_train"]) # return knn gnb_package = load_pickle_file("gnb_digit6_package.pkl") pca_multi = load_pickle_file("pcaMulti.pkl") rf_model = load_pickle_file("rf_hog_model.pkl") #pca_KNN = load_pickle_file("knn_pca.pkl") if pca_multi is not None: pca_multi = pca_multi['pca'] if gnb_package is not None: gnb_binary_model = gnb_package["model"] pca = gnb_package["pca"] best_weight = gnb_package["best_weight"] hog_params = gnb_package["hog_params"] else: gnb_binary_model = None pca = None best_weight = None hog_params = None print("Loading Decision Tree...") dt_package = load_pickle_file("dt.pkl") if dt_package is not None: dt_binary_model = dt_package["model"] else: dt_binary_model = None print("Loading Logistic Regression...") logreg_binary_model = load_pickle_file("custom_logreg_hog.pkl") svm_package = load_pickle_file("svm2.pkl") if svm_package is not None: svm_binary_model = svm_package["model"] else: svm_binary_model = None svm_scaler_dict = load_pickle_file("svmScaler.pkl") svm_scaler = svm_scaler_dict.get('scaler', svm_scaler_dict.get('model')) knn_model_binary = load_pickle_file('knn_final_model.pkl') knn_multi = load_pickle_file("knn_hog_pca.pkl") gnb_multimode_model = load_pickle_file('gnbMulti.pkl') if gnb_multimode_model is not None: gnb_multimode_model = gnb_multimode_model["model"] dt_multimode_model = load_pickle_file('dt_Multi.pkl') if dt_multimode_model is not None: dt_multimode_model = dt_multimode_model['model'] svm_multimode_model = load_pickle_file('svm_hog.pkl') svm_multimode_scaler = load_pickle_file('scaler_hog.pkl') logreg_multimode_model = load_pickle_file('multiLog2_hog.pkl') def preprocess_image(image, model_choice, mode_choice): #image = cv2.resize(image, (28, 28)) image = np.array(image) image = image / 255.0 image_hog = hog( image, **hog_params ) if mode_choice == "Binary": if model_choice == "GNB" or model_choice == "KNN": image_hog = image_hog.reshape(1, -1) image_pca = pca.transform(image_hog) return image_pca elif model_choice == "SVM": image_r = image.reshape(1,-1) scaled_img = svm_scaler.transform(image_r) scaled_img = scaled_img.reshape(28,28) svm_img_hog = hog(scaled_img, **hog_params) return svm_img_hog.reshape(1, -1) else: image_hog = image_hog.reshape(1, -1) return image_hog else: if model_choice == "GNB": image_hog = image_hog.reshape(1, -1) image_pca = pca_multi.transform(image_hog) return image_pca elif model_choice == "DT" or model_choice == "Logistic Regression" or model_choice == "SVM": image_hog = image_hog.reshape(1, -1) return image_hog elif model_choice == "KNN": image_hog = image_hog.reshape(1, -1) image_pca = pca_multi.transform(image_hog) return image_pca def classify_binary_gnb(image): processed_image = preprocess_image(image, "GNB", "Binary") prediction = gnb_binary_model.predict( processed_image, class_weights={0: 1.0, 1: best_weight} ) if prediction[0] == 1: return "GNB Binary: Is 6" else: return "GNB Binary: Not 6" def classify_binary_dt(image): processed_image = preprocess_image(image, "DT", "Binary") prediction = dt_binary_model.predict(processed_image) if prediction[0] == 1: return "Decision Tree Binary: Is 6" else: return "Decision Tree Binary: Not 6" def classify_binary_svm(image): processed_image = preprocess_image(image, "SVM", "Binary") #scaled = svm_scaler.transform(processed_image) prediction = svm_binary_model.predict(processed_image) if prediction[0] == 1: return "SVM Binary: Is 6" else: return "SVM Binary: Not 6" def classify_binary_logreg(image): processed_image = preprocess_image(image, "Logistic Regression", "Binary") prediction = logreg_binary_model.predict(processed_image) if prediction[0] == 1: return "Logistic Regression Binary: Is 6" else: return "Logistic Regression Binary: Not 6" def classify_knn(image): if knn_model_binary is None: return "KNN Binary model placeholder: model is not inserted yet." processed_image = preprocess_image(image, "KNN", "Binary") prediction = knn_model_binary.predict(processed_image) if prediction[0] == 1: return "KNN Binary: Is 6" else: return "KNN Binary: Not 6" def classify_multimode_gnb(image): if gnb_multimode_model is None or pca_multi is None: return "GNB Multimode model could not be loaded. If this is a Git LFS file, run `git lfs pull` first." processed_image = preprocess_image(image, "GNB", "Multimode") prediction = gnb_multimode_model.predict(processed_image) return "GNB Multimode prediction: " + str(prediction[0]) def classify_multimode_dt(image): if dt_multimode_model is None: return "Decision Tree Multimode model could not be loaded. If this is a Git LFS file, run `git lfs pull` first." processed_image = preprocess_image(image, "DT", "Multimode") prediction = dt_multimode_model.predict(processed_image) return "Decision Tree Multimode prediction: " + str(prediction[0]) def classify_multimode_svm(image): if svm_multimode_model is None: if 'svm_hog.pkl' in loading_errors: return f"ERROR: {loading_errors['svm_hog.pkl']}" else: return "SVM Multimode model placeholder: model is not inserted yet." processed_image = preprocess_image(image, "SVM", "Multimode") scaled = svm_multimode_scaler.transform(processed_image) prediction = svm_multimode_model.predict(scaled) return "SVM Multimode prediction: " + str(prediction[0]) def classify_multimode_logreg(image): if logreg_multimode_model is None: return "Logistic Regression Multimode model placeholder: model is not inserted yet." processed_image = preprocess_image(image, "Logistic Regression", "Multimode") image_scaled = svm_multimode_scaler.transform(processed_image) prediction = logreg_multimode_model.predict(image_scaled) return "Logistic Regression Multimode prediction: " + str(prediction[0]) def classify_multimode_knn(image): if knn_multi is None: return "KNN Multimode model could not be loaded. If this is a Git LFS file, run `git lfs pull` first." processed_image = preprocess_image(image, "KNN", "Multimode") prediction = knn_multi.predict(processed_image) return "KNN Multimode prediction: " + str(prediction[0]) def classify_rf(image): processed_image = preprocess_image(image , "DT" , "Multimode") prediction = rf_model.predict(processed_image) return "RF Prediction: " + str(prediction[0]) def predict_with_selected_model(image, model_choice, mode_choice): if image is None: return "Please upload an image." if mode_choice == "Binary": if model_choice == "GNB": return classify_binary_gnb(image) elif model_choice == "DT": return classify_binary_dt(image) elif model_choice == "SVM": return classify_binary_svm(image) elif model_choice == "Logistic Regression": return classify_binary_logreg(image) elif model_choice == "KNN": return classify_knn(image) else: return "Unknown binary model selected." elif mode_choice == "Multimode": if model_choice == "GNB": return classify_multimode_gnb(image) elif model_choice == "DT": return classify_multimode_dt(image) elif model_choice == "SVM": return classify_multimode_svm(image) elif model_choice == "Logistic Regression": return classify_multimode_logreg(image) elif model_choice == "KNN": return classify_multimode_knn(image) elif model_choice == "Random Forest": return classify_rf(image) else: return "Unknown multimode model selected." else: return "Unknown mode selected." demo = gr.Interface( fn=predict_with_selected_model, inputs=[ gr.Image( image_mode="L", type="numpy", label="Upload Digit Image" ), gr.Radio( choices=["GNB", "DT", "SVM", "Logistic Regression", "KNN" , "Random Forest"], value="GNB", label="Choose Model Type" ), gr.Radio( choices=["Binary", "Multimode"], value="Binary", label="Choose Classification Mode" ) ], outputs=gr.Textbox(label="Prediction"), title="MNIST Classifier", description="Upload an image, select the model type, choose binary or multimode, and get a prediction." ) if __name__ == "__main__": demo.launch()