rayheberer

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt

from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
from sklearn.ensemble import GradientBoostingRegressor

from collections import defaultdict

rayheberer

import numpy as np
import tensorflow as tf


def dense(x, weights, bias, activation=tf.identity, **activation_kwargs):
    """Dense layer."""
    z = tf.matmul(x, weights) + bias
    return activation(z, **activationn_kwargs)

rayheberer

import numpy as np
import tensorflow as tf


def dense(x, weights, bias, activation=tf.identity, **activation_kwargs):
    """Dense layer."""
    z = tf.matmul(x, weights) + bias
    return activation(z, **activation_kwargs)

rayheberer

import numpy as np
import tensorflow as tf


def dense(x, weights, bias, activation=tf.identity, **activation_kwargs):
    """Dense layer."""
    z = tf.matmul(x, weights) + bias
    return activation(z, **activation_kwargs)

rayheberer

df_lagged = df.copy()
trailing_window_size = 10

for window in range(1, trailing_window_size + 1):
    shifted = df.shift(window)
    shifted.columns = [x + "_lag" + str(window) for x in df.columns]
    
    df_lagged = pd.concat((df_lagged, shifted), axis=1)
df_lagged = df_lagged.dropna()
df_lagged.head()

rayheberer

import numpy as np
import pandas as pd
import pandas_datareader.data as web

import datetime

start = datetime.datetime(2012, 1, 1)
end = datetime.datetime(2019, 1, 1)

df = web.DataReader("TSLA", 'yahoo', start, end)

rayheberer

class AtariNet(object):
    # ...
    # ...
    def _build_optimization(self):
        # ...
        # ...
        self.advantage = tf.subtract(
            self.returns,
            tf.squeeze(tf.stop_gradient(self.value_estimate)),
            name="advantage")

rayheberer

class A2CAtari(base_agent.BaseAgent):
    # ...
    # ...
    def _get_batch(self, terminal):
        # ...
        # ...
        # calculate discounted rewards
        raw_rewards = list(self.reward_buffer)
        if terminal:
            value = 0

rayheberer

class A2CAtari(base_agent.BaseAgent):
    # ...
    # ...
    def _sample_action(self,
                       screen_features,
                       minimap_features,
                       flat_features,
                       available_actions):
        """Sample actions and arguments from policy output layers."""
        screen_features = np.expand_dims(screen_features, 0)

rayheberer

class AtariNet(object):
    # ...
    # ...
    def _build(self):
        # ...
        # ...
        # action function identifier policy
        self.function_policy = tf.squeeze(tf.layers.dense(
            inputs=self.state_representation,
            units=NUM_ACTIONS,