Ravish Chawla ravishchawla

## laughter_is_the_best_medicine.md

      
              1 file
            
          
              0 forks
            
          
              0 comments
            
          
              0 stars
            
          
                ravishchawla
                / laughter_is_the_best_medicine.md
            
            
              Last active
              June 8, 2020 22:50
            
          
    Sources

Early findings of the phrase - King James 17.22

https://www.kingjamesbibleonline.org/Proverbs-17-22/
https://qr.ae/pNKpyP


Health Benefits of Laughter

https://caregiver.com/articles/laughter-is-the-best-medicine/


University Studies that researched the benefits of laughter between two groups of people

https://www.mcgill.ca/oss/article/health-and-nutrition-trotter-public-science-symposium/laughter-best-medicine
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2762283/


https://www.health.harvard.edu/blog/laugh-and-be-thankful-its-good-for-the-heart-20101124839


## quora_model_results.csv

          
            Model Configuration
             Number of Epochs
             Training F1 Score
             Validation F1 Score

            
              2 LSTM Layers with 64 Hidden Units
               5 epochs
               0.637
               0.540

            
              2 LSTM Layers with 128 Hidden Units
               5 epochs
               0.625
               0.567

            
              2 LSTM Layers with 128 Hidden Units
               15 epochs
               0.661
               0.558

            
              1 LSTM Layer and a Self Attention Layer
               5 epochs
               0.771
               0.601

## quora_model_training.py
def train(nn_model, nn_optimizer, nn_criterion, data_loader, val_loader = None, num_epochs = 5, print_ratio = 0.1, verbose=True):
    for epoch in range(num_epochs):
        # Enable Training for the model
        nn_model.train()
        running_loss = 0;

        for ite, (x, y, l) in enumerate(data_loader):
            init_time = time.time();

            # Convert our tensors to GPU tensors

## quora_torch_model.py
class Model(nn.Module):
    def __init__(self, embedding_matrix, hidden_unit = 64):
        super(Model, self).__init__();
        vocab_size = embeddings_tensor.shape[0];
        embedding_dim = embeddings_tensor.shape[1];

        self.embedding_layer = nn.Embedding(vocab_size, embedding_dim);
        self.embedding_layer.weight = nn.Parameter(embeddings_tensor);
        self.embedding_layer.weight.requires_grad = True;


## quora_torch_model
class Model(nn.Module):
    def __init__(self, embedding_matrix, hidden_unit = 64):
        super(Model, self).__init__();
        vocab_size = embeddings_tensor.shape[0];
        embedding_dim = embeddings_tensor.shape[1];

        self.embedding_layer = nn.Embedding(vocab_size, embedding_dim);
        self.embedding_layer.weight = nn.Parameter(embeddings_tensor);
        self.embedding_layer.weight.requires_grad = True;


## data_augmentation.py
nearest_syns = NearestNeighbors(n_neighbors=total_syns+1).fit(embeddings_matrix);
neighbours_mat = nearest_syns.kneighbors(embeddings_matrix[1:top_k])[1];
synonyms = {x[0]: x[1:] for x in neighbours_mat};

def augment_sentence(encoded_sentence, prob = 0.5):
    for posit in range(len(encoded_sentence)):
        if random.random() > prob:
            try:
                syns = synonyms[encoded_sentence[posit]];
                rand_syn = np.random.choice(syns);

## data_cleaning_methods.csv

          
            Data Cleaning Procedure
             Coverage of Vocabulary
             Coverage of Dataset

            
              Raw Data (all records)
               0.18
               0.71

            
              Raw Data (on 10% sample)
               0.08
               0.71

            
              Lower Casing all words (on 10% sample)
               0.10
               0.87

            
              Removing and Replacing Non-Alpha Numeric Characters (on 10% sample)
               0.11
               0.98

            
              Replacing Contractions with Full words (on 10% sample)
               0.11
               0.98

            
              All methods (all records)
               0.27
               0.98

## check_coverage.py
def check_coverage(text, embeddings_dict):
    known_words, unknown_words = {}, {};
    total_known, total_unknown = 0, 0;

    for sentence in text:
        for word in sentence.split(' '):
            if word in known_words:
                total_known = total_known + 1;
            elif word in embeddings_dict:
                known_words[word] = embeddings_dict[word];

## swagger_jira_v2.json
{
    "openapi":"3.0.1",
    "info":{
       "title":"The Jira Cloud platform REST API",
       "description":"Jira Cloud platform REST API documentation",
       "termsOfService":"http://atlassian.com/terms/",
       "contact":{
          "email":"ecosystem@atlassian.com"
       },
       "license":{

## rl_hyperparams.csv

          
            Agent
             Hyperparameters

            
              Replay Buffer Size
               1e5

            
              Minibatch Size
               128

            
              Discount Rate
               0.99

            
              TAU
               1e-3

            
              Actor Learning Rate
               1e-4

            
              Critic Learning Rate
               1e-4

            
              L2 Weight Decay
               1e-6


              Actor Model
               Hyperparameters
Model Configuration	Number of Epochs	Training F1 Score	Validation F1 Score
2 LSTM Layers with 64 Hidden Units	5 epochs	0.637	0.540
2 LSTM Layers with 128 Hidden Units	5 epochs	0.625	0.567
2 LSTM Layers with 128 Hidden Units	15 epochs	0.661	0.558
1 LSTM Layer and a Self Attention Layer	5 epochs	0.771	0.601
	def train(nn_model, nn_optimizer, nn_criterion, data_loader, val_loader = None, num_epochs = 5, print_ratio = 0.1, verbose=True):
	for epoch in range(num_epochs):
	# Enable Training for the model
	nn_model.train()
	running_loss = 0;

	for ite, (x, y, l) in enumerate(data_loader):
	init_time = time.time();

	# Convert our tensors to GPU tensors
	class Model(nn.Module):
	def __init__(self, embedding_matrix, hidden_unit = 64):
	super(Model, self).__init__();
	vocab_size = embeddings_tensor.shape[0];
	embedding_dim = embeddings_tensor.shape[1];

	self.embedding_layer = nn.Embedding(vocab_size, embedding_dim);
	self.embedding_layer.weight = nn.Parameter(embeddings_tensor);
	self.embedding_layer.weight.requires_grad = True;
	nearest_syns = NearestNeighbors(n_neighbors=total_syns+1).fit(embeddings_matrix);
	neighbours_mat = nearest_syns.kneighbors(embeddings_matrix[1:top_k])[1];
	synonyms = {x[0]: x[1:] for x in neighbours_mat};

	def augment_sentence(encoded_sentence, prob = 0.5):
	for posit in range(len(encoded_sentence)):
	if random.random() > prob:
	try:
	syns = synonyms[encoded_sentence[posit]];
	rand_syn = np.random.choice(syns);
Data Cleaning Procedure	Coverage of Vocabulary	Coverage of Dataset
Raw Data (all records)	0.18	0.71
Raw Data (on 10% sample)	0.08	0.71
Lower Casing all words (on 10% sample)	0.10	0.87
Removing and Replacing Non-Alpha Numeric Characters (on 10% sample)	0.11	0.98
Replacing Contractions with Full words (on 10% sample)	0.11	0.98
All methods (all records)	0.27	0.98
	def check_coverage(text, embeddings_dict):
	known_words, unknown_words = {}, {};
	total_known, total_unknown = 0, 0;

	for sentence in text:
	for word in sentence.split(' '):
	if word in known_words:
	total_known = total_known + 1;
	elif word in embeddings_dict:
	known_words[word] = embeddings_dict[word];
	{
	"openapi":"3.0.1",
	"info":{
	"title":"The Jira Cloud platform REST API",
	"description":"Jira Cloud platform REST API documentation",
	"termsOfService":"http://atlassian.com/terms/",
	"contact":{
	"email":"ecosystem@atlassian.com"
	},
	"license":{
	Agent	Hyperparameters
	Replay Buffer Size	1e5
	Minibatch Size	128
	Discount Rate	0.99
	TAU	1e-3
	Actor Learning Rate	1e-4
	Critic Learning Rate	1e-4
	L2 Weight Decay	1e-6

	Actor Model	Hyperparameters