alexlimh

import argparse
import collections
import functools
import gc
import os
import pathlib
import resource
import sys
import warnings

alexlimh

ATARI_DEFAULTS = dict(
    # General
    steps=2e8,
    eval_every=1e5,
    log_every=1e4,
    prefill=5e4,
    dataset_size=4e6,
    pretrain=0,

    # Environment

alexlimh

import argparse
import collections
import functools
import gc
import os
import pathlib
import resource
import sys
import warnings

alexlimh

import argparse
import collections
import functools
import gc
import os
import pathlib
import resource
import sys
import warnings

alexlimh

Running Minecraft in Headless Linux System

Prerequisite

1. xpra

   sudo apt-get install xpra

2. jdk 1.8

alexlimh

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import collections

from dm_control import mujoco
from dm_control.rl import control
from dm_control.suite import base
from dm_control.suite import common

alexlimh

#!/bin/bash

#SBATCH --mem=50G
#SBATCH -c 8
#SBATCH --gres=gpu:1
#SBATCH -p gpu


CKPT=/checkpoint/minghan/${SLURM_JOB_ID}
touch $CKPT

alexlimh

<?xml version="1.0" encoding="UTF-8" standalone="no" ?>
<Mission xmlns="http://ProjectMalmo.microsoft.com">
    <About>
        <Summary>$(ENV_NAME)</Summary>
    </About>

    <ModSettings>
         <MsPerTick>50</MsPerTick>
          <!--<PrioritiseOffscreenRendering>true</PrioritiseOffscreenRendering>-->
    </ModSettings>

alexlimh

# 1. retrieve top-k passages for each query
# 2. merge the top-k passages and get the union
# 3. calculate the dot-product on the union for each query
# 3.5. save the dot-products into .pkl or .npy file
# 4. put the score into this function and get mutual information
def get_mi(self, logits, tau=1.0): # logits are the dot-product scores, size: QxP, where Q is the number of queries, P is the number of retrieved passages
    # tau is the temperature coef that controls the sharpness of the distribution
    # numeric stability
    max_logit = np.max(logits, axis=-1, keepdims=True)
    probs = np.exp((logits - max_logit)/tau)/np.sum(np.exp((logits - max_logit)/tau), -1, keepdims=True)    

alexlimh

#!/usr/bin/env python3
# Copyright (c) Facebook, Inc. and its affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.

"""
 Command line tool to get dense results and validate them
"""