dipanjanS

import os
import math
import numpy as np
import json
from tensorflow import keras
from tensorflow.keras.metrics import categorical_accuracy
import tensorflow as tf

os.environ['TF_ENABLE_AUTO_MIXED_PRECISION'] = '1'
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'

dipanjanS

# building sample data generator
sample_generator = train_datagen.flow_from_directory(directory=TRAIN_DIR,
                                                    classes=['brightpixel', 'narrowband', 
                                                             'narrowbanddrd', 'noise', 
                                                             'squarepulsednarrowband', 'squiggle',
                                                             'squigglesquarepulsednarrowband'],
                                                    target_size=IMG_DIMS,
                                                    batch_size=1,
                                                    class_mode='categorical',
                                                    interpolation='bicubic',

dipanjanS

# instantiating the image data generator objects
train_datagen = keras.preprocessing.image.ImageDataGenerator(rescale=1./255.,
                                                             zoom_range=0.05,
                                                             rotation_range=180,
                                                             vertical_flip=True,
                                                             horizontal_flip=True, 
                                                             fill_mode='reflect')

valid_datagen = keras.preprocessing.image.ImageDataGenerator(rescale=1./255.)
test_datagen = keras.preprocessing.image.ImageDataGenerator(rescale=1./255.)

dipanjanS

from keras.preprocessing.image import img_to_array, load_img
import glob
import matplotlib.pyplot as plt
%matplotlib inline

data_signal_files = glob.glob(TRAIN_DIR+'/*/*.png')
sample_files = [data_signal_files[idx] for idx in range(0, 5600, 800+15)]
fix, ax = plt.subplots(2, 4, figsize=(14, 6))

for idx, img in enumerate(sample_files):

dipanjanS

fig, ax = plt.subplots(2, 4, figsize=(14, 6))

for idx, img in enumerate(sample_files):
    id1 = 1 if idx > 3 else 0
    id2 = idx % 4
    img_arr = img_to_array(load_img(img))
    f = ax[id1, id2].imshow(img_arr / 255., aspect='auto')
    t = ax[id1, id2].set_title(img.split('/')[-1].split('.')[0].split('_')[-1], fontsize=10)
ax[1,3].set_axis_off()

dipanjanS

import cv2

img_arr_bgr = cv2.imread(sample_files[2])
img_arr_rgb = cv2.cvtColor(img_arr_bgr, cv2.COLOR_BGR2RGB)
print('Image Shape:', img_arr.shape)
fig, ax = plt.subplots(1, 2, figsize=(10, 4))
p1 = ax[0].imshow(img_arr_bgr, aspect='auto')
t1 = ax[0].set_title('BGR spectrogram')
p2 = ax[1].imshow(img_arr_rgb, aspect='auto')
t1 = ax[1].set_title('RGB spectrogram')

dipanjanS

from keras.preprocessing.image import img_to_array, load_img

img = load_img(sample_files[2])
img_arr = img_to_array(img)
print('Image Shape:', img_arr.shape)
fig, ax = plt.subplots(1, 2, figsize=(10, 4))
p1 = ax[0].imshow(np.array(img_arr, dtype='uint8'), aspect='auto')
t1 = ax[0].set_title('Image Spectrogram')
p2 = ax[1].imshow(img_arr / 255., aspect='auto')
t2 = ax[1].set_title('Image Spectrogram Normalized (0-1)')

dipanjanS

train_dir = os.path.join('./data/train')
data_signal_files = glob.glob(train_dir+'/*/*.png')
sample_files = [data_signal_files[idx] for idx in range(0, 5600, 800+10)]
sample_files

# Output
['./data/train/brightpixel/272_brightpixel.png',
 './data/train/squigglesquarepulsednarrowband/6438_squigglesquarepulsednarrowband.png',
 './data/train/narrowbanddrd/2012_narrowbanddrd.png',
 './data/train/squarepulsednarrowband/1396_squarepulsednarrowband.png',

dipanjanS

!sudo tree --dirsfirst --filelimit 10 './data/'

./data/
├── test
│   ├── brightpixel [100 entries exceeds filelimit, not opening dir]
│   ├── narrowband [100 entries exceeds filelimit, not opening dir]
│   ├── narrowbanddrd [100 entries exceeds filelimit, not opening dir]
│   ├── noise [100 entries exceeds filelimit, not opening dir]
│   ├── squarepulsednarrowband [100 entries exceeds filelimit, not opening dir]
│   ├── squiggle [100 entries exceeds filelimit, not opening dir]

dipanjanS

fig, ax = plt.subplots(4, 4, figsize=(12, 10))
id1 = 0
id2 = 0
for signal_type in df['signal_classification'].unique().tolist():
    data_ids = df[df['signal_classification']==signal_type].sample(n=2)['uuid'].tolist()
    data_files = [f for f in raw_signal_files if [x for x in data_ids if x in f]]

    for idx, data_file in enumerate(data_files):
        data = ibmseti.compamp.SimCompamp(open(data_file,'rb').read())
        spectrogram = data.get_spectrogram()