tensorflow-qnd 0.0.8

Quick and Distributed TensorFlow command framework

# tensorflow-qnd

[](https://badge.fury.io/py/tensorflow-qnd)
[](setup.py)
[](https://travis-ci.org/raviqqe/tensorflow-qnd)
[](https://unlicense.org)

Quick and Distributed TensorFlow command framework

tensorflow-qnd is a TensorFlow framework to create commands to train and
evaluate models on multiple computers.
The framework is built on top of
`tensorflow.contrib.learn.python.learn.learn_runner` and relevant modules.
While made to be used on multiple computers in a cluster, this library is also
useful to exploit multiple GPUs on a single machine.


## Installation

Python 3.5+ and TensorFlow 0.12+ are required.

```
pip3 install --user --upgrade tensorflow-qnd
```


## Usage

See [documentation](https://raviqqe.github.io/tensorflow-qnd/qnd).


## Examples

```python
import logging

import qnd
import tensorflow as tf


logging.getLogger().setLevel(logging.INFO)


def read_file(filename_queue):
_, serialized = tf.TFRecordReader().read(filename_queue)

scalar_feature = lambda dtype: tf.FixedLenFeature([], dtype)

features = tf.parse_single_example(serialized, {
"image_raw": scalar_feature(tf.string),
"label": scalar_feature(tf.int64),
})

image = tf.decode_raw(features["image_raw"], tf.uint8)
image.set_shape([28**2])

return tf.to_float(image) / 255 - 0.5, features["label"]


def minimize(loss):
return tf.contrib.layers.optimize_loss(
loss,
tf.contrib.framework.get_global_step(),
0.001,
"Adam")


def model(image, number):
h = tf.contrib.layers.fully_connected(image, 64)
h = tf.contrib.layers.fully_connected(h, 10, activation_fn=None)

loss = tf.reduce_mean(
tf.nn.sparse_softmax_cross_entropy_with_logits(h, number))
predictions = tf.argmax(h, axis=1)

return predictions, loss, minimize(loss), {
"accuracy": tf.contrib.metrics.streaming_accuracy(predictions,
number)[1],
}


run = qnd.def_run()


def main():
run(model, read_file)


if __name__ == "__main__":
main()
```

With the code above, you can create a command with the following interface.

```
usage: mnist_simple.py [-h] [--output_dir OUTPUT_DIR]
[--train_steps TRAIN_STEPS] [--eval_steps EVAL_STEPS]
[--min_eval_frequency MIN_EVAL_FREQUENCY] --master_host
MASTER_HOST --ps_hosts PS_HOSTS
[--worker_hosts WORKER_HOSTS] --task_type TASK_TYPE
[--task_index TASK_INDEX] [--num_cores NUM_CORES]
[--log_device_placement]
[--save_summary_steps SAVE_SUMMARY_STEPS]
[--save_checkpoints_steps SAVE_CHECKPOINTS_STEPS]
[--batch_size BATCH_SIZE]
[--batch_queue_capacity BATCH_QUEUE_CAPACITY]
--train_file TRAIN_FILE
[--filename_queue_capacity FILENAME_QUEUE_CAPACITY]
--eval_file EVAL_FILE

optional arguments:
-h, --help show this help message and exit
--output_dir OUTPUT_DIR
Directory where checkpoint and event files are stored
(default: output)
--train_steps TRAIN_STEPS
Maximum number of train steps (default: None)
--eval_steps EVAL_STEPS
Maximum number of eval steps (default: None)
--min_eval_frequency MIN_EVAL_FREQUENCY
Minimum evaluation frequency in number of model
savings (default: 1)
--master_host MASTER_HOST
HOSTNAME:PORT pair of a master host (default: None)
--ps_hosts PS_HOSTS Comma-separated list of $hostname:$port pairs of ps
hosts (default: [])
--worker_hosts WORKER_HOSTS
Comma-separated list of $hostname:$port pairs of
worker hosts (default: [])
--task_type TASK_TYPE
Must be in ['master', 'ps', 'worker'] (aka job)
(default: None)
--task_index TASK_INDEX
Task index within a job (default: 0)
--num_cores NUM_CORES
Number of CPU cores used. 0 means use of a default
value. (default: 0)
--log_device_placement
If specified, log device placement information
(default: False)
--save_summary_steps SAVE_SUMMARY_STEPS
Number of steps every time of which summary is saved
(default: 100)
--save_checkpoints_steps SAVE_CHECKPOINTS_STEPS
Number of steps every time of which a model is saved
(default: None)
--batch_size BATCH_SIZE
Mini-batch size (default: 64)
--batch_queue_capacity BATCH_QUEUE_CAPACITY
Batch queue capacity (default: 1024)
--train_file TRAIN_FILE
File path of train data file(s). A glob is available.
(e.g. train/*.tfrecords) (default: None)
--filename_queue_capacity FILENAME_QUEUE_CAPACITY
Capacity of filename queues of train and eval data
(default: 32)
--eval_file EVAL_FILE
File path of eval data file(s). A glob is available.
(e.g. eval/*.tfrecords) (default: None)
```

See also [examples](examples) directory.


## Caveats

### Necessary update of a global step variable

As done in [all examples](examples), you must get a global step variable
with `tf.contrib.framework.get_global_step()` and update (increment) it in each
training step.


### Use streaming metrics for `eval_metric_ops`

When non-streaming ones such as `tf.contrib.metrics.accuracy` are used in a
return value `eval_metric_ops` of your `model_fn` or as arguments of
`ModelFnOps`, their values will be ones of the last batch in every evaluation
step.


## License

[The Unlicense](https://unlicense.org)


## References

- [Distributed TensorFlow | TensorFlow](https://www.tensorflow.org/how_tos/distributed/)