def test_loading():
import tensorflow as tf
checkpoint_path = "examples/iris_tensorflow/model_files/model.ckpt"
const_feed_dict_pkl = "examples/iris_tensorflow/model_files/const_feed_dict.pkl"
# dict of variables
# input = list
a = TensorFlowModel(input_nodes="inputs",
target_nodes="probas",
checkpoint_path=checkpoint_path)
o = a.predict_on_batch(np.ones((3, 4)))
assert o.shape == (3, 3)
# input = dict
a = TensorFlowModel(input_nodes={"out_name": "inputs"},
target_nodes="probas",
checkpoint_path=checkpoint_path)
with pytest.raises(AssertionError):
o = a.predict_on_batch(np.ones((3, 4)))
o = a.predict_on_batch({"out_name": np.ones((3, 4))})
assert o.shape == (3, 3)
# input = list
a = TensorFlowModel(input_nodes=["inputs"],
target_nodes="probas",
checkpoint_path=checkpoint_path)
with pytest.raises(AssertionError):
o = a.predict_on_batch(np.ones((3, 4)))
o = a.predict_on_batch([np.ones((3, 4))])
assert o.shape == (3, 3)
# output = dict
a = TensorFlowModel(input_nodes="inputs",
target_nodes={"out_name": "probas"},
checkpoint_path=checkpoint_path)
o = a.predict_on_batch(np.ones((3, 4)))
assert isinstance(o, dict)
assert list(o.keys()) == ["out_name"]
assert o['out_name'].shape == (3, 3)
# output = list
a = TensorFlowModel(input_nodes="inputs",
target_nodes=["probas"],
checkpoint_path=checkpoint_path)
o = a.predict_on_batch(np.ones((3, 4)))
assert isinstance(o, list)
assert len(o) == 1
assert o[0].shape == (3, 3)
# test with the extra input
a = TensorFlowModel(input_nodes="inputs",
target_nodes="probas",
checkpoint_path=checkpoint_path,
const_feed_dict_pkl=const_feed_dict_pkl)
o = a.predict_on_batch(np.ones((3, 4)))
assert o.shape == (3, 3)
def test_grad_tens_generation():
import tensorflow as tf
checkpoint_path = "examples/iris_tensorflow/model_files/model.ckpt"
a = TensorFlowModel(input_nodes="inputs",
target_nodes="probas",
checkpoint_path=checkpoint_path)
fwd_values = a.predict_on_batch(np.ones((3, 4)))
assert np.all(
a.get_grad_tens(fwd_values,
DummySlice()[:,
0:1], "min")[0, :] == np.array([1, 0, 0]))
assert np.all(
a.get_grad_tens(fwd_values,
DummySlice()[:,
0:2], "min")[0, :] == np.array([1, 0, 0]))
assert np.all(
a.get_grad_tens(fwd_values,
DummySlice()[:,
0:2], "max")[0, :] == np.array([0, 1, 0]))
assert np.all(
a.get_grad_tens(fwd_values,
DummySlice()[0:2], "max")[0, :] == a.get_grad_tens(
fwd_values,
DummySlice()[:, 0:2], "max")[0, :])
def test_tf_model():
tf.reset_default_graph()
input_nodes = "inputs"
target_nodes = "preds"
meta_graph = "model_files/model.tf.meta"
# meta_graph = 'model_files/model.tf-modified.meta'
checkpoint = "model_files/model.tf"
index = "model_files/model.tf.index"
pkl_file = "model_files/const_feed_dict.pkl"
from kipoi.model import TensorFlowModel
m = TensorFlowModel(input_nodes="inputs",
target_nodes="preds",
meta_graph=meta_graph,
checkpoint=checkpoint,
const_feed_dict_pkl=pkl_file)
ops = tf.get_default_graph().get_operations()
# TODO - modify the
out = tf.train.export_meta_graph(
filename='model_files/model.tf-modified.meta', as_text=True)
ops[0].outputs[0].shape[0] = None
pops = [
op.outputs[0] for op in ops
if op.type == "Placeholder" and op.name.startswith("Placeholder")
]
m.input_ops # view shapes of the data
m.target_ops
from concise.preprocessing import encodeDNA
x = encodeDNA(["T" * m.input_ops.shape[1].value] * 2).astype("float32")
out = m.predict_on_batch(x)