#+TITLE: Tensorflow experiments
#+AUTHOR: Amin Kasrou Aouam
#+PROPERTY: header-args :session poetry-session
* Experiments

#+begin_src elisp :results silent
(pyvenv-activate "~/.cache/pypoetry/virtualenvs/locimend-hM_4JND0-py3.8/")
#+end_src

In this notebook we'll extract knowledge from our generated dataset. First, let's import our dependencies:

#+begin_src python
from tensorflow_io import genome
#+end_src

#+RESULTS:
: 2021-05-06 20:41:53.592058: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /nix/store/9ilyrqidrjbqvmnn8ykjc7lygdd86g7q-gcc-10.2.0-lib/lib:
: 2021-05-06 20:41:53.592101: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.


Tensorflow I/O is an extension that contains a module for genome parsing, we'll use it to import the sequences contained in our FASTQ files:

#+begin_src python :results silent
def parse_data(filepath):
    HVR = genome.read_fastq(filename=filepath)
    return HVR.sequences, HVR.raw_quality
#+end_src

Let's import both the immuneSIM generated HVR dataset and the CuReSim processed one, which contains sequencing errors (mostly indels):

#+begin_src python
original_HVR, _ = parse_data("../data/HVR.fastq")
processed_HVR, _ = parse_data("../data/CuReSim-HVR.fastq")
print(original_HVR)
print(processed_HVR)
#+end_src

#+RESULTS:
#+begin_example
tf.Tensor(
[b'TGTGCCAGCAGCTTAACCATCGGACGCAGTACTTCGGGCCAGGCACGCGGCTCCTGG'
 b'TGTGCCAGCAGCTTAACCATCGGACGCAGTACTTCGGGCCAGGCACGCGGCTCCTGG'
 b'TGTGCCAGCAGCTTAACCATCGGACGCAGTACTTCGGGCCAGGCACGCGGCTCCTGG'
 b'TGTGCCAGCAGCTTAACCATCGGACGCAGTACTTCGGGCCAGGCACGCGGCTCCTGG'
 b'TGTGCCAGCAGCTTAACCATCGGACGCAGTACTTCGGGCCAGGCACGCGGCTCCTGG'
 b'TGTGCCAGCAGCTTAACCATCGGACGCAGTACTTCGGGCCAGGCACGCGGCTCCTGG'
 b'TGTGCCAGCAGCTTAACCATCGGACGCAGTACTTCGGGCCAGGCACGCGGCTCCTGG'
 b'TGTGCCAGCAGCTTAACCATCGGACGCAGTACTTCGGGCCAGGCACGCGGCTCCTGG'
 b'TGTGCCAGCAGCTTAACCATCGGACGCAGTACTTCGGGCCAGGCACGCGGCTCCTGG'
 b'TGTGCCAGCAGCTTAACCATCGGACGCAGTACTTCGGGCCAGGCACGCGGCTCCTGG'], shape=(10,), dtype=string)
tf.Tensor(
[b'GCGCCAGCAGCTATTGGATATGGACTAGCTACTC'
 b'TGTGCCAGCAGTGATGTGGTGACATGGGTGCGTAGCAATCAGCCAGCATG'
 b'GCGCCAGCAGCTTGGATAGGACTAGCTACTT'
 b'TGTGCCAGCAGTGAATGGGTGACAGGGTGCGTAGCATCAGCCCCAGCATTT'
 b'TTGCGCAGCAGCTTGGATAGGACTAGCTACTT'
 b'TGTGCCAGCAGTGAATGGGGACAGGGGCGTAGCAATCAGCCCCAGCATTT'
 b'TTGCGCCAGCAGCTTGGATAGGACTAGCTACTT'
 b'TGTGCAGCAGTGAATGGGGACAGGGGCGTAGCAATCAGCCCCAGCATTT'
 b'TGCGCCAGCAGCTTGGATAGGACTAGCTACTT'
 b'TGTGCCAGCAGTGAATGGGGACAGGGGCGTAGCAATCAGCCCAGCATTT'
 b'TTGCGCCAGCAGCTTGGATAGGACTAGCTACTT'
 b'TGTGCCAGCAGTGAATGGGGACAGGGGCGTAGCAATCAGCCCCAGCATTT'
 b'TGCGCCAGCAGCTTGGATAGGACTAGCTACTT'
 b'TGTGCCAGCAGTGAATGGGGACAGGGGCGTAGCAATCAGCCCCAGCATTT'
 b'TGCGCCAGCAGCTTGGATAGGACTAGCTACTT'
 b'TGTGCCAGCAGTGAATGGGGACAGGGGCGTAGCAATCAGCCCCAGCATTT'
 b'TGCGCCAGCAGCTTGGATAGGACTAGCTACTT'
 b'TGTGCCAGCAGTGAATGGGGACAGGGGCGTAGCAATCAGCCCCAGCATTT'
 b'TGCGCCAGCAGCTTGGATAGGACTAGCTACTT'
 b'TGTGCCAGCAGTGAATGGGGACAGGGGCGTAGCAATCAGCCCCAGCATTT'], shape=(20,), dtype=string)
#+end_example