tldrafael

# Install cuda-10.0 because the newest one (for now the version 10.2) mismatchs the tensorflow 1.14.0 requirements
wget https://developer.nvidia.com/compute/cuda/10.0/Prod/local_installers/cuda-repo-ubuntu1804-10-0-local-10.0.130-410.48_1.0-1_amd6
 
dpkg -i cuda-repo-ubuntu1804-10-0-local-10.0.130-410.48_1.0-1_amd6

apt-key add /var/cuda-repo-10-0-local-10.0.130-410.48/7fa2af80.pub

apt-get install cuda

# This cuda version install nvidia v410 drivers. Remove it.

tldrafael

from line_profiler import LineProfiler
def do_profile(follow=[]):
    def inner(func):
        def profiled_func(*args, **kwargs):
            try:
                profiler = LineProfiler()
                profiler.add_function(func)
                for f in follow:
                    profiler.add_function(f)
                profiler.enable_by_count()

tldrafael

# https://arxiv.org/pdf/1602.07360.pdf
#
from keras import backend as K
from keras.layers import Input, Convolution2D, MaxPooling2D, Activation, concatenate
from keras.layers import GlobalAveragePooling2D
from keras.models import Model


class SqueezeNet:
    def __init__(self, input_shape, n_classes):

tldrafael

from glob import iglob

def return_impaths(dpath):
  impaths = [list(iglob('{}/*.{}'.format(dpath, e))) for e in ['png', 'jpg', 'jpeg']]
  impaths = sum(impaths, [])
  return impaths

tldrafael

import numpy as np

class inputGen:
    def __init__(self, batch_size, X, y, shuffle=True):
        self.batch_size = batch_size
        self.X = X
        self.y = y
        self.cursor = 0
        self.n_samples = X.shape[0]
        self.ids_sequence = np.arange(X.shape[0])

tldrafael

import rlcompleter
import pdb
pdb.Pdb.complete = rlcompleter.Completer(locals()).complete
pdb.set_trace()

# Oneline to insert into code
import rlcompleter; import pdb; pdb.Pdb.complete = rlcompleter.Completer(locals()).complete; pdb.set_trace()

tldrafael

import argparse

def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument('-e', type=str, help='Environment variable', required=True, dest='var1',
                        choices=['staging', 'production'])
    parser.add_argument('--hypertuning', help='Run hypertuning process', action='store_true', dest='var2')
    return parser.parse_args()

tldrafael

# Easy to go tools
  - http://www.robots.ox.ac.uk/~vgg/software/via/via-1.0.6.html
  - https://github.com/virajmavani/semi-auto-image-annotation-tool
  - https://labelbox.com/ -> Free until 2500 images per year.

# More references
  - https://github.com/jsbroks/awesome-dataset-tools
  - https://www.datasetlist.com/tools/

tldrafael

import matplotlib.pyplot as plt

fig, axs = plt.subplots(4, figsize=(12, 12))

for i in range(4):
    axs[i].imshow(sample_masks[:, :, i])
    axs[i].axis('off')

tldrafael

# Code snippet to train lightgbm models quickly, using the number of iterations based on CV

import lightgbm as lgb
lgb_dftrain = lgb.Dataset(features_train, target_train)

params = {'boosting_type': 'gbdt',
          'objective': 'binary',
          'enable_bundle': True,
          'max_conflict_rate': 0,
          'max_depth': 20,