rohit-gupta

from keras.utils import Sequence

class BreitbartCommentsSequence(Sequence):
def __init__(self, comments, vocab_size, previous_words, batch_size):
    all_comments_text = ' '.join(comments)
    all_words = all_comments_text.split(" ")
    vocabulary = collections.Counter(all_words)
    self.vocabulary = [word for word, count in vocabulary.most_common(vocab_size)]
    self.previous_words = previous_words
    self.batch_size = batch_size

rohit-gupta

wget download-installer.cdn.mozilla.net/pub/firefox/releases/57.0b4/linux-x86_64/en-US/firefox-57.0b4.tar.bz2
wget http://download-installer.cdn.mozilla.net/pub/firefox/releases/57.0b4/SHA256SUMS
diff <(sha256sum firefox-57.0b4.tar.bz2 | awk 'BEGIN{FS=" "} {print $1}') <(grep "linux-x86_64/en-US/firefox-57.0b4.tar.bz2" SHA256SUMS | awk 'BEGIN{FS=" "} {print $1}')
tar xjf firefox-57.0b4.tar.bz2
sudo rm -r /opt/firefox
sudo mv firefox /opt/quantum
sudo mv /usr/bin/firefox /usr/bin/firefox-old
sudo ln -s /opt/quantum/firefox /usr/bin/firefox

rohit-gupta

from stopwords import remove_stopwords

dummy_string = ("""Mr. and Mrs. Dursley, of number four, Privet Drive, were """
                """proud to say that they were perfectly normal, thank you """
                """very much. They were the last people you'd expect to be """
                """involved in anything strange or mysterious, because they """
                """just didn't hold with such nonsense.""")

# Simple Word tokenizer
dummy_list = dummy_string.replace(",", " ").replace(".", " ").split(" ")

rohit-gupta

\usepackage{pgfplots}
\pgfplotsset{compat=1.8}


\begin{figure}
    \centering
    \begin{tikzpicture}
        \begin{axis}
        [   title  = Number of videos in each category in MSR-VTT,
            xbar,

rohit-gupta

def beam_search(captioning_model, prev_words_input, other_inputs, k):
    top_k_predictions = [copy.deepcopy(prev_words_input) for _ in range(k)]
    top_k_score = np.array([[0.0]*top_k_predictions[0].shape[0]]*k)

    # First Iteration 
    predictions = captioning_model.predict(other_inputs + [prev_words_input])
    for idx in range(prev_words_input.shape[0]):
        for version in range(k):
            top_k_predictions[version][idx][1] = np.argsort(predictions[idx])[-(version+1)]
            top_k_score[version][idx] = np.sort(predictions[idx])[-(version+1)]

rohit-gupta

ActivityNet 2018 Results: http://activity-net.org/challenges/2018/uploads/ActivityNet_Challenge_2018_Summary_and_Workshop_Papers.pdf

I3D: https://www.arxiv-vanity.com/papers/1705.07750/

3D CNNs: http://openaccess.thecvf.com/content_cvpr_2018/papers/Hara_Can_Spatiotemporal_3D_CVPR_2018_paper.pdf

AVA: http://openaccess.thecvf.com/content_cvpr_2018/papers/Gu_AVA_A_Video_CVPR_2018_paper.pdf

Kinetics: https://arxiv.org/pdf/1705.06950.pdf

rohit-gupta

How good is your GAN ? https://hal.inria.fr/hal-01850447/document
Low Shot visual recognition by hallucinating features https://arxiv.org/pdf/1606.02819.pdf
Google JFT-300M https://arxiv.org/pdf/1707.02968.pdf
Instagram Imagenet https://research.fb.com/wp-content/uploads/2018/05/exploring_the_limits_of_weakly_supervised_pretraining.pdf
Deep Layer Aggregation https://arxiv.org/pdf/1707.06484.pdf
CNNs fit random data https://arxiv.org/pdf/1611.03530.pdf

rohit-gupta

# Get equation text from LM
# Format: y = a + b.x, R^2 = r^2
equation_text = function(x, round_dig = 2) {
  lm_coef <- list(a = as.character(round(coef(x)[1], digits = round_dig)),
                  b = as.character(round(coef(x)[2], digits = round_dig)),
                  r2 = round(summary(x)$r.squared, digits = round_dig));
  lm_eq <- substitute(italic(y) == a + b %.% italic(x)*","~~italic(R)^2~"="~r2,lm_coef)
  as.character(as.expression(lm_eq));                 
}

rohit-gupta

# Basic Name Plot
tax_plot <- ggplot(State_Infrastructure, aes(x=tax_burden, y=percent_bad_roads, label=State))
tax_state_names_plot <- tax_plot + geom_text_repel() + geom_point(color = 'red') + theme_classic(base_size = 16)

# OUTPUT
tax_state_names_plot + ggtitle("Road Conditions vs Tax Burden") +
  scale_x_continuous(name = "State and Local Tax Burden") +
  scale_y_continuous(name = "Percent of roads not in good condition")

rohit-gupta

# Tax regression plot

# Linear Regression
tax_fit <- lm(percent_bad_roads ~ tax_burden, data = State_Infrastructure)

min_gap = 0.05
lab_coord_xmin = min(State_Infrastructure$tax_burden)
lab_coord_xmax = max(State_Infrastructure$tax_burden)
lab_coord_ymin = min(State_Infrastructure$percent_bad_roads) - min_gap 
lab_coord_height = 0.1