hartikainen

import glob
import os
from pprint import pprint
import re
import shutil


RESULT_FILE_REGEXES = (
    "^result.json$",
    "^progress.csv$",

hartikainen

import os
import sys
from shutil import copyfile

import numpy as np
import pandas as pd


def get_result_path(trial_dir):
    return os.path.join(trial_dir, "progress.csv")

hartikainen

from collections import defaultdict
import time

import matplotlib.pyplot as plt
import numpy as np


def create_fields(observation_shape, action_shape):
    fields = {
        'observations': observation_shape,

hartikainen

import tempfile

import tensorflow as tf


class PicklableKerasModel(tf.keras.Model):
    def __getstate__(self):
        with tempfile.NamedTemporaryFile(suffix='.hdf5', delete=True) as fd:
            tf.keras.models.save_model(self, fd.name, overwrite=True)
            model_str = fd.read()

hartikainen

hartikainen

import numpy as np
import tensorflow as tf
import tensorflow_probability as tfp


class LearnableConditionalRealNVP(object):
    def __init__(self, input_shape, output_shape):
        self._input_shape = input_shape
        self._output_size = np.prod(output_shape)

hartikainen

class FloatLookupLayer(tf.keras.layers.Layer):

    def __init__(self, all_observation_pairs, **kwargs):
        self.all_observation_pairs = all_observation_pairs
        x = np.reshape(
            self.all_observation_pairs,
            (self.all_observation_pairs.shape[0],
             np.prod(self.all_observation_pairs.shape[1:3])))
        x = tf.reduce_join(tf.as_string(x), separator='|', axis=-1)
        x = tf.concat((x, ['ERROR']), axis=0)

hartikainen

# Base softlearning container that contains all softlearning requirements,
# but not the actual softlearning repo. Could be used for example when developing
# softlearning, in which case you would mount softlearning repo in to the container
# as a volume, and thus be able to modify code on the host, yet run things inside
# the container. You are encouraged to use docker-compose (docker-compose.dev.yml),
# which should allow you to setup your environment with a single one command.

FROM nvidia/cuda:9.0-runtime-ubuntu16.04

ARG MJKEY

hartikainen

import numpy as np
from sklearn.model_selection import train_test_split

def split_data(X, y, train_prop, val_prop, test_prop, N_dev=0, verbose=True):
    """ Split the dataset (X, y) into train, validation, and test sets.

    Also, possibly create a small dataset for development purposes. Note that
    the proportions should sum to 1.
    Arguments:
    X -- Feature matrix, shape of (N, D_in)