vishal-keshav

Logging in C++

When writing a high performance C++ program, sometimes it become crucial to log what is happening while program execution is in progress. There are several logging libraries like glog (from Google) and log(from Boost), but using those for a simple program can be counter productive as it unneccesarily increases the executable size and takes more time to setup.

I use the below 5 lines code snippet to log in C++ program. Variadic templates are being used to create such a minimilistic logging system.

template <typename T>
void log_it(T t) {
 cout &lt;&lt; t &lt;&lt; endl;

vishal-keshav

import tensorflow as tf
from tensorflow.python.platform import gfile
import numpy as np

from imagenet_classes import class_names
from scipy.misc import imread, imresize

dir_name = 'mobilenet_v1_1.0_224'
with tf.Graph().as_default() as graph:
	with tf.Session() as sess:

vishal-keshav

"""
A well-documented module to export the tensorflow trained graph as protobuf and
tflite.

Background on protobuf and tflite model format:
Protocol buffer format(frozen) and tflite format contains exactly the same graph
and same weights associated to that graph, the only difference is that tflite
format is understood by tflite interpreter while protobuf format is understood
by tensorflow.

vishal-keshav

"""
Testing the brand new datasets from tensorflow community for experimenting on
ImageNet2012 dataset.

We identify several problems while working with ImageNet dataset:
1. The dataset is not easy to download. Credentials (email) of some well known
organization/university is required to get the dowanload link.

2. The huge size if dataset, namely "ILSVRC2012_img_train.tar" -> 138Gb
and "ILSVRC2012_img_val.tar" -> 7Gb

vishal-keshav

import sys
import tensorflow as tf
from tensorflow.python.tools import freeze_graph
from tensorflow.python.tools import optimize_for_inference_lib

def convert_frozen_to_inference(model_path = "generated_model",
                frozen_file = "generated_model.pb", inputs = ["input"],
                outputs = ["output"], out_file = "generated_model_opt.pb"):
    frozen_graph = tf.GraphDef()
    with tf.gfile.Open(model_path + "/" + frozen_file) as f:

vishal-keshav

import tensorflow as tf
import numpy as np

def preprocess(img):
	# Apply any preprocessing on the input
	return img

def inference_from_pb(pb_file = "model.pb", img, inputs = ['input'], outputs = ['output']):
	img = preporcess(img)
	with tf.gfile.GFile(pb_file, "rb") as f:

vishal-keshav

def get_model(model_name, input_tensor, is_train, pretrained):
	....

def optimisation(label, logits, param_dict):
	....

def get_data_provider(dataset_name, dataset_path, param_dict):
	....

def test_distributed_training(model, dataset, param_dict):

vishal-keshav

#include "stdio.h"
int main (void) {
    int a=10, b=0, c=10;
    char* bits ="TFy!QJu ROo TNn(ROo)SLq SLq ULo+UHs UJq TNn*RPn/QPbEWS_JSWQAIJO^NBELPeHBFHT}TnALVlBLOFAkHFOuFETpHCStHAUFAgcEAelclcn^r^r\\tZvYxXyT|S~Pn SPm SOn TNn ULo0ULo#ULo-WHq!WFs XDt!";
    a = bits[b];
    while (a != 0) {
        a = bits[b];
        b++;
        while (a > 64) {
            a--;

vishal-keshav

This snippet I often use to output a bunch of sample images in jupyter notebook.

%matplotlib inline
from matplotlib import pyplot as plt
_, axs = plt.subplots(1, 10, figsize=(32, 32))
axs = axs.flatten()
for img, ax in zip(example_images, axs):
    ax.imshow(img)
plt.show()
print(example_labels)

vishal-keshav

class Solution(object):
    def convert(self, s, numRows):
        if numRows == 1: return s
        arr_of_arr = [[] for i in range(numRows)]
        running_index, direction = 0, 1
        for c in s:
            arr_of_arr[running_index].append(c)
            running_index += direction
            if running_index>=numRows:
                running_index, direction = numRows-2, -1