Initializing ``HIBOU``
======================
Now we can create another file that will get the data from ``data_reader`` and give them to ``hibou``. Let's start with the header::

    import data_reader

    import numpy as np
    import pandas pd
    import tensorflow as tf

    import hibou
    from hibou import filters, nn, stats, utils, config, preprocessing

Once this is done, we can define several constants::

    config.OUTPUT_DIR = "my_output_dir"
    TRAINING_SIZE = 3 * 336     # 3 weeks
    EVALUATION_SIZE = 12 * 336  # 12 weeks
    INPUTS_VAR = [
        config.Inputs.TEMPERATURE,
        config.Inputs.WIND_SPEED,
        config.Inputs.U,
        config.Inputs.V,
        config.Inputs.NET_RADIATION
        config.Inputs.DAILY_SIN,
        config.Inputs.DAILY_COS,
    ]
    TARGETS_VAR = [
        config.Targets.H,
        config.Targets.LE
    ]

    hibou.init()  # initialize directory for the output and logs

* The ``OUTPUT_DIR`` is the folder that will contain all the data, from the synchronized data to the graphs exported.

* ``TRAINING_SIZE`` and ``EVALUATION_SIZE`` are the sizes of training and evaluation sets.

* ``INPUTS_VAR`` are the inputs variables for the ANN.

* ``TARGETS_VAR`` are the targets variables for the ANN.

Some precisions on the training and evaluation sizes. These parameters describe the way the data will be splitted between training and evaluating data sets.

For instance, if ``TRAINING_SIZE`` is on 2 weeks and ``EVALUATION_SIZE`` on 6 weeks, the neural network will have two weeks of training then 6 weeks of evaluation and so on until the whole data are splitted::

          ┌──┬──┬──┬──┬──┬──┬──┬──┬──┬──┬──┬──┬──┬──┬──┬──┐
    weeks │ 1│ 2│ 3│ 4│ 5│ 6│ 7│ 8│ 9│10│11│12│13│14│15│16│
          └──┴──┴──┴──┴──┴──┴──┴──┴──┴──┴──┴──┴──┴──┴──┴──┘
          └─────┘━━━━━━━━━━━━━━━━━└─────┘━━━━━━━━━━━━━━━━━
          │train│   evaluation    │train│   evaluation    │