def create_quantiles(quantiles_dir):
"""Creates quantiles directory if it doesn't yet exist."""
input_fn = get_test_input_fn()
tfl.save_quantiles_for_keypoints(input_fn=input_fn,
save_dir=quantiles_dir,
feature_columns=create_feature_columns(),
num_steps=None)
def create_quantiles(quantiles_dir):
"""Creates quantiles directory if it doesn't yet exist."""
batch_size = 10000
input_fn = get_train_input_fn(batch_size=batch_size,
num_epochs=1,
shuffle=False)
# Reads until input is exhausted, 10000 at a time.
tfl.save_quantiles_for_keypoints(input_fn=input_fn,
save_dir=quantiles_dir,
feature_columns=create_feature_columns(),
num_steps=None)
def create_quantiles(quantiles_dir):
batch_size = 10000 #设置批次
#创建输入函数
input_fn = get_input_fn(traindir, batch_size, num_epochs=1, shuffle=False)
tfl.save_quantiles_for_keypoints( #默认保存1000个校准关键点
input_fn=input_fn,
save_dir=quantiles_dir, #默认会建立一个文件目录
feature_columns=create_feature_columns(),
num_steps=None)
estimator.train(input_fn=input_fn())
print("Train iter:", time.time())
print(f"Finished {iter_cnt}/{iters}.")
evaluation = estimator.evaluate(input_fn=input_fn())
print("Eval:", time.time())
print(f"Current average_loss: {evaluation['average_loss']}")
window_size = 21
num_chrom_samples = 4096
sample_intervals = 16
iterations = 32
quantiles_path = os.path.join(quantiles_dir, str(window_size))
if os.path.exists(quantiles_path):
shutil.rmtree(quantiles_path)
tfl.save_quantiles_for_keypoints(
input_fn = get_input_fn(num_chrom_samples, sample_intervals, window_size),
save_dir=quantiles_path,
feature_columns=create_feature_columns(window_size),
num_steps=None)
print("Starting run")
config = tf.estimator.RunConfig().replace(model_dir=model_dir)
model = create_calibrated_rtl(window_size, config)
train(model, iterations, window_size, num_chrom_samples, sample_intervals)
