pvyas96/random_label_forest

1import streamlit as st2import numpy as np3from scipy.sparse import csr_matrix4import joblib5from sklearn.metrics import f1_score6from sklearn.decomposition import TruncatedSVD7from tqdm import tqdm8import tempfile9import os10 11st.set_page_config(page_title="RLF Model Tester", layout="wide")12st.title("Random Label Forest (RLF) - Test Dataset Evaluation")13 14# ----------------------- Parsing Function ----------------------- #15def parse_rcv1_sparse_file(filename):16    with open(filename, 'r') as f:17        header = f.readline()18        num_samples, num_features, num_labels = map(int, header.strip().split())19        20        feat_data, feat_rows, feat_cols = [], [], []21        label_data, label_rows, label_cols = [], [], []22 23        for row_idx, line in enumerate(f):24            line = line.strip()25            if not line:26                continue27            label_part, feature_part = line.split(' ', 1)28            labels = list(map(int, label_part.split(',')))29            features = feature_part.strip().split()30 31            for feat in features:32                fid, fval = feat.split(':')33                feat_rows.append(int(fid))34                feat_cols.append(row_idx)35                feat_data.append(float(fval))36 37            for lid in labels:38                label_rows.append(lid)39                label_cols.append(row_idx)40                label_data.append(1.0)41 42    X = csr_matrix((feat_data, (feat_rows, feat_cols)), shape=(num_features, num_samples), dtype=np.float32)43    Y = csr_matrix((label_data, (label_rows, label_cols)), shape=(num_labels, num_samples), dtype=np.uint8)44    return X.transpose(), Y.transpose()45 46# ----------------------- Prediction Functions ----------------------- #47def predict_scores_one_tree(x, tree):48    kmeans1 = tree['level1_kmeans']49    level2_dict = tree['level2_kmeans']50    tree_model = tree['model_tree']51 52    l1_cluster = kmeans1.predict(x.reshape(1, -1))[0]53 54    if l1_cluster not in level2_dict:55        return {}56 57    kmeans2 = level2_dict[l1_cluster]58    l2_cluster = kmeans2.predict(x.reshape(1, -1))[0]59 60    node_key = (l1_cluster, l2_cluster)61    if node_key not in tree_model:62        return {}63 64    classifiers = tree_model[node_key]['classifiers']65    label_ids = tree_model[node_key]['label_ids']66 67    preds = {}68    for clf, lid in zip(classifiers, label_ids):69        score = clf.decision_function(x.reshape(1, -1))[0]70        preds[lid] = score71    return preds72 73def predict_ensemble(X_test_reduced, ensemble):74    all_preds = []75    for i in range(X_test_reduced.shape[0]):76        instance_scores = {}77        x = X_test_reduced[i]78        for tree in ensemble:79            preds = predict_scores_one_tree(x, tree)80            for lid, score in preds.items():81                if lid not in instance_scores:82                    instance_scores[lid] = []83                instance_scores[lid].append(score)84        avg_scores = {lid: np.mean(scores) for lid, scores in instance_scores.items()}85        all_preds.append(avg_scores)86    return all_preds87 88def get_topk(pred_scores, k):89    top_labels = []90    for scores in pred_scores:91        if len(scores) == 0:92            top_labels.append([])93            continue94        sorted_labels = sorted(scores.items(), key=lambda x: -x[1])95        top_labels.append([lid for lid, _ in sorted_labels[:k]])96    return top_labels97 98def precision_at_k(preds_topk, Y_true, k):99    hits = 0100    total = len(preds_topk)101    for i, pred_labels in enumerate(preds_topk):102        true_labels = set(Y_true[i].nonzero()[1])103        hits += len(set(pred_labels) & true_labels)104    return hits / (total * k)105 106def evaluate_rforest(ensemble, X_test, Y_test, dim=300):107    svd = TruncatedSVD(n_components=dim, random_state=42)108    X_test_reduced = svd.fit_transform(X_test)109 110    pred_scores = predict_ensemble(X_test_reduced, ensemble)111 112    preds_at_1 = get_topk(pred_scores, k=1)113    preds_at_3 = get_topk(pred_scores, k=3)114 115    p1 = precision_at_k(preds_at_1, Y_test, k=1)116    p3 = precision_at_k(preds_at_3, Y_test, k=3)117 118    y_true = Y_test.toarray()119    y_pred_bin = np.zeros_like(y_true)120    for i, pred in enumerate(get_topk(pred_scores, k=5)):121        y_pred_bin[i, pred] = 1122 123    macro_f1 = f1_score(y_true, y_pred_bin, average='macro', zero_division=0)124 125    return p1, p3, macro_f1126 127# ----------------------- Load Pretrained Model ----------------------- #128@st.cache_resource129def load_model():130    return joblib.load("random_label_forest_model.pkl")131 132ensemble_model = load_model()133 134# ----------------------- User Interface ----------------------- #135uploaded_file = st.file_uploader("Upload Test Dataset (.txt in RCV1 format)", type=["txt"])136 137if uploaded_file:138    with tempfile.NamedTemporaryFile(delete=False, suffix=".txt") as tmp:139        tmp.write(uploaded_file.getvalue())140        test_file_path = tmp.name141 142    if st.button("Test Model"):143        st.info("Parsing dataset and evaluating. Please wait...")144        X_test, Y_test = parse_rcv1_sparse_file(test_file_path)145        p1, p3, macro_f1 = evaluate_rforest(ensemble_model, X_test, Y_test)146 147        st.success("Evaluation Completed!")148        st.metric("Precision@1", f"{p1:.4f}")149        st.metric("Precision@3", f"{p3:.4f}")150        st.metric("Macro F1 Score", f"{macro_f1:.4f}")151