def test_embedding_function_exact_weighted():
"""
Creates embedding function and makes sure that for known X it produces exact Y.
:return:
"""
data = load_iris()
X = data.data
labels = data.target
# Embedding function requires unique arrays.
# Well, embedding fuction can protect from it, but still we can get some confusion in mean square error, if it
# "chooses" different sample
un_array, un_idx = np.unique(X, axis=0, return_index=True)
X = X[un_idx, :]
dTSNE = dynamic_tsne.DynamicTSNE(perplexity=20)
# Small dataset. Iterations are very fast, we can afford more
y = dTSNE.fit(X,
verbose=2,
optimizer_kwargs={
'momentum': 0.8,
'n_iter': 3000
},
random_seed=1)
for power in [1, 0.5, 2]:
embedder = dTSNE.generate_embedding_function(
embedding_function_type='weighted-inverse-distance',
function_kwargs={'power': power})
y2 = embedder(X, verbose=2)
assert np.mean(np.sum((y - y2)**2,
axis=1)) == 0, "Embedding was not exact"
def test_random_seed():
"""
Sets random seed and tests that results are the same.
:return:
"""
data = load_iris()
X = data.data
dTSNE = dynamic_tsne.DynamicTSNE(perplexity=20)
# Small dataset. Iterations are very fast, we can afford more
y = dTSNE.fit(X,
optimizer_kwargs={
'momentum': 0.8,
'n_iter': 3000
},
random_seed=1)
y2 = dTSNE.fit(X,
optimizer_kwargs={
'momentum': 0.8,
'n_iter': 3000
},
random_seed=1)
assert (y == y2).all(), "Same random seed - different results"
def test_init():
"""
Test - initialize and do not crash.
:return:
"""
dynamic_tsne.DynamicTSNE(perplexity=20)
if __name__ == "__main__":
data = load_digits()
X = data.images.reshape((-1, 64)) # 8x8 image to 64-length vector
labels = data.target
# Embedding function requires unique arrays.
# Well, embedding fuction can protect from it, but still we can get some confusion in mean square error, if it
# "chooses" different sample
temp = np.ascontiguousarray(X).view(
np.dtype((np.void, X.dtype.itemsize * X.shape[1])))
_, un_idx = np.unique(temp, return_index=True)
X = X[un_idx, :]
labels = labels[un_idx]
dTSNE = dynamic_tsne.DynamicTSNE(perplexity=20)
# Small dataset. Iterations are very fast, we can afford more
y = dTSNE.fit(X,
verbose=2,
optimizer_kwargs={
'momentum': 0.8,
'n_iter': 1000
},
random_seed=1)
embedder2 = dTSNE.generate_embedding_function(
embedding_function_type='weighted-inverse-distance')
embedder3 = dTSNE.generate_embedding_function(
embedding_function_type='weighted-inverse-distance',
function_kwargs={'power': 20})
embedder4 = dTSNE.generate_embedding_function(
embedding_function_type='linear')
