abhi1868sharma’s gists

abhi1868sharma / positional_encoding_pytorch.py

Created November 28, 2021 12:53

	import torch

	from torch import nn

	import math


	class PositionalEncoding(nn.Module):
	"Implement the PE function."
	def __init__(self, d_model, max_len=5000):

abhi1868sharma / positional_encoding_numpy.py

Last active November 28, 2021 12:57

This gist creates a positional encoding of size x.size(1),512

	import numpy as np

	d_model=512

	def positional_embedding(pos,d_model=512):

	dpow=np.array([2i/d_model for i in range(d_model//2)]) # takes care of 10k^ 2i/d_model

	# defining the sin/cos denom
	sindenom=np.power(np.repeat(10000,d_model//2),dpow)

abhi1868sharma / treelite_vs_lightgbm.py

Last active August 22, 2020 14:45

	import treelite

	import treelite_runtime # runtime module

	import numpy as np

	import time

	dim = 100
	toolchain = 'gcc'

abhi1868sharma / treelite_vs_xgboost.py

Last active August 22, 2020 14:45

	import treelite

	import treelite_runtime # runtime module

	import numpy as np

	import time

	dim = 100
	toolchain = 'gcc'

abhi1868sharma / xgb_training.py

Created July 26, 2020 18:32

	import xgboost as xgb

	import numpy as np

	import random

	def xgb_data(training_rows=1000000,dim=100):

	"""
	Create training/validation data for xgboost

abhi1868sharma / lgb_training.py

Last active July 26, 2020 18:32

	import lightgbm as lgb

	import numpy as np

	import random

	from collections import Counter

	def lgb_data(training_rows=1000000,dim=100):

abhi1868sharma / xdeepfm-5

Created December 16, 2019 11:35

	model = xDeepFM(linear_feature_columns, dnn_feature_columns, dnn_hidden_units=(256, 256),\
	cin_layer_size=(128, 128), \
	cin_split_half=True, cin_activation='relu'\
	,l2_reg_linear=1e-05,\
	l2_reg_embedding=1e-05, l2_reg_dnn=0, l2_reg_cin=0, \
	init_std=0.0001,seed=1024, dnn_dropout=0,dnn_activation='relu', \
	dnn_use_bn=False, task='binary')
	#compiling the model
	model.compile("adam", "binary_crossentropy",metrics=['binary_crossentropy'], )
	# training the model

abhi1868sharma / xdeepfm-4

Created December 16, 2019 11:34

	# features to be used for dnn part of xdeepfm
	dnn_feature_columns = sparse_feature_columns + dense_feature_columns
	# features to be used for linear part of xdeepfm
	linear_feature_columns = sparse_feature_columns + dense_feature_columns
	feature_names = get_feature_names(linear_feature_columns + dnn_feature_columns)
	# creating train test splits
	train, test = train_test_split(data, test_size=0.2)
	train_model_input = {name:train[name].values for name in feature_names}
	test_model_input = {name:test[name].values for name in feature_names}

abhi1868sharma / xdeepfm-3

Created December 16, 2019 11:33

	# encoding function
	def encoding(data,feat,encoder):
	data[feat] = encoder.fit_transform(data[feat])
	# encoding for categorical features
	[encoding(data,feat,LabelEncoder()) for feat in sparse_features]
	# Using normalization for dense feature
	mms = MinMaxScaler(feature_range=(0,1))
	data[dense_features] = mms.fit_transform(data[dense_features])
	# creating a 4 bit embedding for every sparse feature
	sparse_feature_columns = [SparseFeat(feat, vocabulary_size=data[feat].nunique(),embedding_dim=4) \

abhi1868sharma / xdeepfm-2

Created December 16, 2019 11:33

	# reading criteo_sample data
	data = pd.read_csv('./criteo_sample.txt')
	# categorising the features into sparse/dense feature set
	sparse_features = ['C' + str(i) for i in range(1, 27)]
	dense_features = ['I'+str(i) for i in range(1, 14)]
	# data imputation for missing values
	data[sparse_features] = data[sparse_features].fillna('-1', )
	data[dense_features] = data[dense_features].fillna(0,)
	# creating target variable
	target = ['label']