def test_generate_internal_shape(N, expected):
if expected is Exception:
with pytest.raises(expected):
generate_internal_weights(N)
else:
W = generate_internal_weights(N)
assert W.shape == expected
def test_reproducibility_W():
seed0 = default_rng(78946312)
W0 = generate_internal_weights(N=100,
sr=1.2,
proba=0.4,
dist="uniform",
low=-1,
high=1,
seed=seed0).toarray()
seed1 = default_rng(78946312)
W1 = generate_internal_weights(N=100,
sr=1.2,
proba=0.4,
dist="uniform",
low=-1,
high=1,
seed=seed1).toarray()
seed2 = default_rng(6135435)
W2 = generate_internal_weights(N=100,
sr=1.2,
proba=0.4,
dist="uniform",
low=-1,
high=1,
seed=seed2).toarray()
assert_array_almost_equal(W0, W1)
assert_raises(AssertionError, assert_array_almost_equal, W0, W2)
def test_generate_internal_features(sr, proba):
W = generate_internal_weights(100, spectral_radius=sr,
proba=proba, seed=1234)
assert_almost_equal(max(abs(linalg.eig(W)[0])), sr, decimal=2)
assert_almost_equal(np.sum(W != 0.0) / W.size, proba, decimal=1)
def test_generate_internal_features(sr, proba):
W = generate_internal_weights(100,
sr=sr,
proba=proba,
seed=1234,
sparsity_type='dense')
assert_almost_equal(max(abs(linalg.eig(W)[0])), sr, decimal=2)
assert_almost_equal(np.sum(W != 0.0) / W.size, proba, decimal=1)
def test_generate_internal_features(sr, proba):
W = generate_internal_weights(100,
sr=sr,
proba=proba,
seed=1234,
sparsity_type="dense")
assert_allclose(max(abs(linalg.eig(W)[0])), sr)
assert_allclose(np.sum(W != 0.0) / W.size, proba)
def test_generate_internal_sparse(sr, proba):
W = generate_internal_weights(100, spectral_radius=sr,
proba=proba, sparsity_type="csr")
rho = max(abs(sparse.linalg.eigs(W, k=1, which='LM', return_eigenvectors=False)))
assert_almost_equal(rho, sr, decimal=2)
if sparse.issparse(W):
assert_almost_equal(np.sum(W.toarray() != 0.0) / W.toarray().size,
proba, decimal=1)
else:
assert_almost_equal(np.sum(W != 0.0) / W.size, proba, decimal=1)
def esn():
W = generate_internal_weights(N,
sr=SR,
proba=1.0,
dist="uniform",
seed=rng)
Win = generate_input_weights(N,
1,
input_scaling=IS,
proba=1,
input_bias=True,
seed=rng)
a = np.ones((1, N))
b = np.zeros((1, N))
return W, Win, a, b
def test_generate_internal_sparse(sr, proba):
W = generate_internal_weights(100,
sr=sr,
proba=proba,
sparsity_type="csr",
seed=42)
rho = max(
abs(
sparse.linalg.eigs(W,
k=1,
which="LM",
maxiter=20 * W.shape[0],
return_eigenvectors=False)))
assert_allclose(rho, sr)
if sparse.issparse(W):
assert_allclose(np.sum(W.toarray() != 0.0) / W.toarray().size, proba)
else:
assert_allclose(np.sum(W != 0.0) / W.size, proba)
def test_generate_internal_features_exception(sr, proba):
with pytest.raises(Exception):
generate_internal_weights(100, sr=sr, proba=proba)
