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

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

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
#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
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
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
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
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
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
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
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