Skip to content

Instantly share code, notes, and snippets.

View reddragon's full-sized avatar
🤖
Too much to do, too little time.

Gaurav Menghani reddragon

🤖
Too much to do, too little time.
View GitHub Profile
View l2_lambda.md

Results on MNIST

Feed Forward model with two hidden layers (300, 60).

l2_lambda Accuracy@1 after 80k iters (Two Runs)
0.00 98.15, 98.04
0.01 98.31, 98.19
0.02 98.19, 98.15
0.04 97.93, 97.92
@reddragon
reddragon / transfer-learning.md
Created April 2, 2018 00:42
Transfer Learning Papers
View transfer-learning.md

How transferable are features in deep neural networks? - Yosinski et al.

  • Transfer Learning

    • Train on a base network, try to take that network and tweak it to work for a new target network.
    • Notes from CS231N.
  • Tries to figure out how much information can we transfer between networks trained on different datasets.

  • Quantifies the transferability by layer.

  • Hypothesis:

  • First few layers are general (Gabor Filters kind of features) and can adapt well.
@reddragon
reddragon / struct.cpp
Last active September 12, 2017 18:30
Set struct members inline
View struct.cpp
#include <iostream>
using namespace std;
struct Foo {
int a;
double b;
};
int main() {
const Foo f = {
@reddragon
reddragon / frozen-lake-nn.py
Created June 12, 2017 23:11
Frozen Lake NN Implementation
View frozen-lake-nn.py
import gym
import logging
import sys
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import cPickle as pickle
import os
@reddragon
reddragon / frozen-lake-iterative.py
Created June 11, 2017 17:12
Frozen Lake solved using the Q-Learning algorithm with an actual Q-value table
View frozen-lake-iterative.py
import gym
import logging
import sys
import numpy as np
from gym import wrappers
SEED = 0
NUM_EPISODES = 3000
# Hyperparams
@reddragon
reddragon / script.py
Created May 29, 2017 07:19
Get all of those graduation pics
View script.py
import os
import sys
import urllib2
def normalize_path(path):
if path[-1] == '/':
path = path[:-1]
return path
def get_dir_name(path):
@reddragon
reddragon / cart-pole-pg-v2.py
Created May 8, 2017 18:32
Slightly tweaked PG for CartPole #dogscience
View cart-pole-pg-v2.py
import gym
import logging
import sys
import numpy as np
from gym import wrappers
import torch
import torchvision
import torch.nn as nn
import torch.nn.functional as F
@reddragon
reddragon / cart-pole-pg.py
Created May 8, 2017 17:33
CartPole for the OpenAI gym using Policy Gradients
View cart-pole-pg.py
import gym
import logging
import sys
import numpy as np
from gym import wrappers
import torch
import torchvision
import torch.nn as nn
import torch.nn.functional as F
@reddragon
reddragon / pong-next20-data.py
Created April 30, 2017 06:43
Predicting whether there would be a goal in the next 20 steps in the ATARI Pong Game
View pong-next20-data.py
import gym
import logging
import sys
import numpy as np
from gym import wrappers
import torch
import torchvision
import torch.nn as nn
import torch.nn.functional as F
View mnist-pytorch.py
import torch
import torchvision
import torch.nn as nn
import torch.nn.functional as F
import torchvision.transforms as transforms
import matplotlib.pyplot as plt
import numpy as np
import torch.optim as optim
from torch.autograd import Variable