Function eval_encoder

experiments/python/main.py:317–483

(dataset, encoder, dist_func_true=None, dist_func_enc=None,
                 eval_dists=True, verbosity=1, plot=False, smaller_better=True)

# ================================================================ Main

def eval_encoder(dataset, encoder, dist_func_true=None, dist_func_enc=None,
                 eval_dists=True, verbosity=1, plot=False, smaller_better=True):

    X = dataset.X_test
    queries = dataset.Q
    true_nn = dataset.true_nn

    if true_nn is not None:
        print "eval encoder(): got true_nn with shape: ", true_nn.shape

    queries = queries[:1000]  # TODO rm for tables; fine for plots
    print "queries.shape", queries.shape

    need_true_dists = eval_dists or plot or true_nn is None

    if len(queries.shape) == 1:
        queries = [queries]

    if dist_func_true is None:
        dist_func_true = encoder.dists_true
    if dist_func_enc is None:
        dist_func_enc = encoder.dists_enc

    t0 = time.time()

    # performance metrics
    RECALL_Rs = [1, 5, 10, 50, 100, 500, 1000]
    recall_counts = np.zeros(len(RECALL_Rs))
    fracs_below_max = []
    if eval_dists:
        all_corrs = []
        all_rel_errs = []
        all_errs = []
        total_dist = 0.

    if need_true_dists:
        X = X[:10000]  # limit to 10k points because otherwise it takes forever
        queries = queries[:256, :]

    print "encoding X..."
    X_enc = encoder.encode_X(X)
    print "trying queries..."
    for i, q in enumerate(queries):

        if i % 100 == 0:
            print "trying query {}...".format(i)

        q_enc = encoder.encode_q(q)
        encoder.fit_query(q)
        if need_true_dists:
            all_true_dists = dist_func_true(X, q)

        all_enc_dists = dist_func_enc(X_enc, q_enc)

        # ------------------------ begin analysis / reporting code

        # find true knn
        if need_true_dists: