Function test_basic

tests/test_encoder.py:169–273

()

def test_basic():
    # np.set_printoptions(precision=3)
    np.set_printoptions(formatter={'float_kind': _fmt_float})

    nqueries = 20
    X, Q = _load_digits_X_Q(nqueries)
    # X, _ = load_digits(return_X_y=True)
    # Q = X[-nqueries:]
    # X = X[:-nqueries]

    # print "X.shape", X.shape
    # print "X nbytes", X.nbytes

    # ------------------------------------------------ squared l2

    enc = bolt.Encoder(accuracy='low', reduction=bolt.Reductions.SQUARED_EUCLIDEAN)
    enc.fit(X)

    l2_corrs = np.empty(nqueries)
    for i, q in enumerate(Q):
        l2_true = _dists_sq(X, q).astype(np.int)
        l2_bolt = enc.transform(q)
        l2_corrs[i] = corr(l2_true, l2_bolt)[0]

    mean_l2 = np.mean(l2_corrs)
    std_l2 = np.std(l2_corrs)
    assert mean_l2 > .95
    print "squared l2 dist correlation: {} +/- {}".format(mean_l2, std_l2)

    # ------------------------------------------------ dot product

    enc = bolt.Encoder(accuracy='low', reduction=bolt.Reductions.DOT_PRODUCT)
    enc.fit(X)

    dot_corrs = np.empty(nqueries)
    for i, q in enumerate(Q):
        dots_true = np.dot(X, q)
        dots_bolt = enc.transform(q)
        dot_corrs[i] = corr(dots_true, dots_bolt)[0]

    mean_dot = np.mean(dot_corrs)
    std_dot = np.std(dot_corrs)
    print "dot product correlation: {} +/- {}".format(mean_dot, std_dot)

    # ------------------------------------------------ l2 knn

    enc = bolt.Encoder(accuracy='low', reduction='l2')
    enc.fit(X)

    k_bolt = 10  # tell bolt to search for true knn
    k_true = 10  # compute this many true neighbors
    true_knn = _knn(X, Q, k_true)
    bolt_knn = [enc.knn(q, k_bolt) for q in Q]

    contained = np.empty((nqueries, k_bolt), dtype=np.bool)
    for i in range(nqueries):
        true_neighbors = true_knn[i]
        bolt_neighbors = bolt_knn[i]