tilarids

tilarids

import tensorflow as tf
import numpy as np

trX = np.linspace(-1, 1, 1001, dtype=np.float32)
trY = 2 * trX + np.random.randn(*trX.shape).astype(np.float32) * 0.33

EPOCHS = 1000
CAP = 1000
queue_xy = tf.train.input_producer(tf.pack([trX, trY], axis=1), shuffle=False, capacity=CAP, num_epochs=EPOCHS)

tilarids

; A minimal Mach-o x32 executable for OS X El Capitan (with proper padding and symtable)
;
; Original (pre 10.10.5) version - https://gist.github.com/softboysxp/1084476
; $ nasm -O0 -f bin -o tiny_hello tiny_hello.asm
; $ chmod +x tiny_hello
; $ ./tiny_hello
; (returns 42)
; $

; c.f.

tilarids

tilarids

Policy Gradients (with & without TRPO).

More details and reproducing: https://github.com/tilarids/reinforcement_learning_playground
This specific commit was using to reproduce this: https://github.com/tilarids/reinforcement_learning_playground/commit/fd442e78ee4c93dfa38a3e83677b3d3cb3eefc90

tilarids

TRPO (described in http://arxiv.org/abs/1502.05477) with an additional neural network to predict value (used for advantage calculation).

More details and steps to reproduce: https://github.com/tilarids/reinforcement_learning_playground
Commit used to produce the result: https://github.com/tilarids/reinforcement_learning_playground/commit/df2b1c68735f31c6ed2b943a1e0309385b53cd0e

tilarids

tilarids

import os.path
import time

import numpy as np
import tensorflow as tf

from tensorflow.python.training import saver as saver_lib
from tensorflow.python.training import summary_io

# Basic model parameters as external flags.

tilarids

main
main.dSYM/

tilarids

main
main.dSYM/