kengz

The HEX color for Atom highlighter theme: close to Base16 but lighter.
bgdark: #2b303b
red: #e26a78
cyan: #78dbf9
green: #8bdb97
purple: #d59cf6
orange: #dbad83

kengz

# First create a Google Compute Engine (GCE) VM instance, then SSH into the VM.
# 1. update everything
sudo apt-get update

# 2. install these 4 things
sudo apt-get install python g++ make checkinstall

# 3. make a src directory and go into it
mkdir ~/src && cd $_

kengz

# Deploying NodeJS App on Google Cloud Platform

`authored Jul 3 2015 by kengz`

## Installation
- Account with Google Cloud Platform.
- Install [Google Cloud Platform SDK](https://cloud.google.com/sdk/) to use `cloud` in the terminal.
- Install [Docker](https://cloud.google.com/tutorials/install-docker) for the VM container.
- Locally, login for the first time: `gcloud auth login`

kengz

Global Hackers Missions
[ Feel free to contribute and sign your own contribution! Separate by “Insert > Horizontal line”. See example below. ]
[ Anyone can edit this document, so be responsible. Of course we back this up somewhere else too 

###

This page is for doers who dare to, and for builders who want to, make a dent in the universe.

Do's: Be open minded, be dare, believe. 	
Don’t be butt-hurt, don’t be evil.

kengz

Keybase proof

I hereby claim:

• I am kengz on github.
• I am kengz (https://keybase.io/kengz) on keybase.
• I have a public key ASCZnM1pXJ3J3XW9-Nk2h22qWrep6ra1BfxuN6YBZFTbggo

To claim this, I am signing this object:

kengz

    def post_body_init(self):
        '''Initializes the part of algorithm needing a body to exist first.'''
        self.body = self.agent.nanflat_body_a[0]  # single-body algo
        # create the extra replay memory for SIL
        memory_name = self.memory_spec['sil_replay_name']
        MemoryClass = getattr(memory, memory_name)
        self.body.replay_memory = MemoryClass(self.memory_spec, self, self.body)
        self.init_algorithm_params()
        self.init_nets()
        logger.info(util.self_desc(self))

kengz

    def train_shared(self):
        '''
        Trains the network when the actor and critic share parameters
        '''
        if self.to_train == 1:
            # onpolicy a2c update
            a2c_loss = super(SIL, self).train_shared()
            # offpolicy sil update with random minibatch
            total_sil_loss = torch.tensor(0.0)
            for _ in range(self.training_epoch):

kengz

    def calc_sil_policy_val_loss(self, batch):
        '''
        Calculate the SIL policy losses for actor and critic
        sil_policy_loss = -log_prob * max(R - v_pred, 0)
        sil_val_loss = norm(max(R - v_pred, 0)) / 2
        This is called on a randomly-sample batch from experience replay
        '''
        returns = math_util.calc_returns(batch, self.gamma)
        v_preds = self.calc_v(batch['states'])
        clipped_advs = torch.clamp(returns - v_preds, min=0.0)

kengz

See https://kengz.gitbooks.io/slm-lab/content/installation/installation.html for full instruction.

1. Clone SLM-Lab

git clone https://github.com/kengz/SLM-Lab.git
cd SLM-Lab

2. Install node, yarn, electron and orca for the environment and analysis module

kengz

# init singleton agent and env
self.env = make_env(self.spec)
body = Body(self.env, self.spec['agent'])
self.agent = Agent(self.spec, self.info_space, body=body)