Add tree learners to strategy evaluation directory

This version does not pass the automatic test.

strategy_evaluation/AbstractTreeLearner.py

@@ -0,0 +1,77 @@
+import numpy as np
+
+
+class AbstractTreeLearner:
+    LEAF = -1
+    NA = -1
+
+    def author(self):
+        return 'felixm' # replace tb34 with your Georgia Tech username
+
+    def create_node(self, factor, split_value, left, right):
+        return np.array([(factor, split_value, left, right), ],
+                        dtype='|i4, f4, i4,  i4')
+
+    def query_point(self, point):
+        node_index = 0
+        while self.rel_tree[node_index][0] != self.LEAF:
+            node = self.rel_tree[node_index]
+            split_factor = node[0]
+            split_value = node[1]
+            if point[split_factor] <= split_value:
+                # Recurse into left sub-tree.
+                node_index += node[2]
+            else:
+                node_index += node[3]
+        v = self.rel_tree[node_index][1]
+        return v
+
+    def query(self, points):
+        """
+        @summary: Estimate a set of test points given the model we built.
+        @param points: should be a numpy array with each row corresponding to a specific query.
+        @returns the estimated values according to the saved model.
+        """
+        query_point = lambda p: self.query_point(p)
+        r = np.apply_along_axis(query_point, 1, points)
+        return r
+
+    def build_tree(self, xs, y):
+        """
+        @summary: Build a decision tree from the training data.
+        @param dataX: X values of data to add
+        @param dataY: the Y training values
+        """
+        assert(xs.shape[0] == y.shape[0])
+        assert(xs.shape[0] > 0) # If this is 0 something went wrong.
+
+        if xs.shape[0] <= self.leaf_size:
+            value = np.mean(y)
+            if value < -0.3:
+                value = -1
+            elif value > 0.3:
+                value = 1
+            else:
+                value = 0
+            return self.create_node(self.LEAF, value, self.NA, self.NA)
+
+        if np.all(y[0] == y):
+            return self.create_node(self.LEAF, y[0], self.NA, self.NA)
+
+        i, split_value = self.get_i_and_split_value(xs, y)
+        select_l = xs[:, i] <= split_value
+        select_r = xs[:, i] > split_value
+        lt = self.build_tree(xs[select_l], y[select_l])
+        rt = self.build_tree(xs[select_r], y[select_r])
+        root = self.create_node(i, split_value, 1, lt.shape[0] + 1)
+        root = np.concatenate([root, lt, rt])
+        return root
+
+    def addEvidence(self, data_x, data_y):
+        """
+        @summary: Add training data to learner
+        @param dataX: X values of data to add
+        @param dataY: the Y training values
+        """
+        self.rel_tree = self.build_tree(data_x, data_y)
+

strategy_evaluation/BagLearner.py

@@ -0,0 +1,47 @@
+import numpy as np
+from AbstractTreeLearner import AbstractTreeLearner
+
+
+class BagLearner(AbstractTreeLearner):
+    def __init__(self, learner, bags=9, boost=False, verbose=False, kwargs={}):
+        self.learner = learner
+        self.verbose = verbose
+        self.bags = bags
+        self.learners = [learner(**kwargs) for _ in range(bags)]
+
+    def get_bag(self, data_x, data_y):
+        num_items = int(data_x.shape[0] * 0.5) # 50% of samples
+        bag_x, bag_y = [], []
+        for _ in range(num_items):
+            i = np.random.randint(0, data_x.shape[0])
+            bag_x.append(data_x[i,:])
+            bag_y.append(data_y[i])
+        return np.array(bag_x), np.array(bag_y)
+
+    def addEvidence(self, data_x, data_y):
+        """
+        @summary: Add training data to learner
+        @param dataX: X values of data to add
+        @param dataY: the Y training values
+        """
+        for learner in self.learners:
+            x, y = self.get_bag(data_x, data_y)
+            learner.addEvidence(x, y)
+
+    def query(self, points):
+        """
+        @summary: Estimate a set of test points given the model we built.
+        @param points: numpy array with each row corresponding to a query.
+        @returns the estimated values according to the saved model.
+        """
+        def to_discret(m):
+            print(m)
+            if m < -0.5:
+                return -1
+            elif m > 0.5:
+                return 1
+            return 0
+        m = np.mean([l.query(points) for l in self.learners], axis=0)
+        return m
+        # return np.apply_along_axis(to_discret, 1, m)
+

strategy_evaluation/ManualStrategy.py

@@ -36,18 +36,21 @@ class ManualStrategy:
             print(volume)
 
     def macd_strat(self, macd, orders):
+        """Strategy based on MACD cross."""
 
         def strat(ser):
             m = macd.loc[ser.index]
-            prev_macd, prev_signal = m.iloc[0]
-            cur_macd, cur_signal = m.iloc[1]
+            prev_macd, prev_signal, _ = m.iloc[0]
+            cur_macd, cur_signal, _ = m.iloc[1]
             shares = 0
-            if cur_macd < -1 and prev_macd < prev_signal and cur_macd > cur_signal:
+            if cur_macd < -1 and prev_macd < prev_signal \
+                             and cur_macd > cur_signal:
                 if self.holding == 0:
                     shares = 1000
                 elif self.holding == -1000:
                     shares = 2000
-            elif cur_macd > 1 and prev_macd > prev_signal and cur_macd < cur_signal:
+            elif cur_macd > 1 and prev_macd > prev_signal \
+                              and cur_macd < cur_signal:
                 if self.holding == 0:
                     shares = -1000
                 elif self.holding == 1000:
@@ -58,6 +61,8 @@ class ManualStrategy:
         orders['Shares'] = orders['Shares'].rolling(2).apply(strat)
 
     def three_indicator_strat(self, macd, rsi, price_sma, orders):
+        """Strategy based on three indicators. Thresholds selected based on
+        scatter plots."""
         def strat(row):
             shares = 0
             _, _, macd_diff = macd.loc[row.name]
@@ -87,7 +92,7 @@ class ManualStrategy:
     def testPolicy(self, symbol="IBM",
                    sd=dt.datetime(2009, 1, 1),
                    ed=dt.datetime(2010, 1, 1),
-                   sv=10000):
+                   sv=10000, macd_strat=False):
 
         self.holding = 0
         df = util.get_data([symbol], pd.date_range(sd, ed))
@@ -102,7 +107,8 @@ class ManualStrategy:
         rsi = indicators.rsi(df, symbol)
         price_sma = indicators.price_sma(df, symbol, [8])
 
-        # self.macd_strat(macd, orders)
+        if macd_strat:
+            self.macd_strat(macd, orders)
+        else:
             self.three_indicator_strat(macd, rsi, price_sma, orders)
         return orders
-

strategy_evaluation/RTLearner.py

@@ -0,0 +1,30 @@
+import numpy as np
+from AbstractTreeLearner import AbstractTreeLearner
+
+
+class RTLearner(AbstractTreeLearner):
+
+    def __init__(self, leaf_size = 1, verbose = False):
+        self.leaf_size = leaf_size
+        self.verbose = verbose
+
+    def get_i_and_split_value(self, xs, y):
+        """
+        @summary: Pick a random i and split value.
+
+        Make sure that not all X are the same for i and also pick
+        different values to average the split_value from.
+        """
+        i = np.random.randint(0, xs.shape[1])
+        while np.all(xs[0,i] == xs[:,i]):
+            i = np.random.randint(0, xs.shape[1])
+
+        # I don't know about the performance of this, but at least it
+        # terminates reliably. If the two elements are the same something is
+        # wrong.
+        a = np.array(list(set(xs[:, i])))
+        r1, r2 = np.random.choice(a, size = 2, replace = False)
+        assert(r1 != r2)
+        split_value = (r1 + r2) / 2.0
+        return i, split_value
+

strategy_evaluation/StrategyLearner.py

@@ -1,88 +1,94 @@
-"""
-Template for implementing StrategyLearner  (c) 2016 Tucker Balch
-
-Copyright 2018, Georgia Institute of Technology (Georgia Tech)
-Atlanta, Georgia 30332
-All Rights Reserved
-
-Template code for CS 4646/7646
-
-Georgia Tech asserts copyright ownership of this template and all derivative
-works, including solutions to the projects assigned in this course. Students
-and other users of this template code are advised not to share it with others
-or to make it available on publicly viewable websites including repositories
-such as github and gitlab.  This copyright statement should not be removed
-or edited.
-
-We do grant permission to share solutions privately with non-students such
-as potential employers. However, sharing with other current or future
-students of CS 7646 is prohibited and subject to being investigated as a
-GT honor code violation.
-
------do not edit anything above this line---
-
-Student Name: Tucker Balch (replace with your name)
-GT User ID: tb34 (replace with your User ID)
-GT ID: 900897987 (replace with your GT ID)
-"""
-
 import datetime as dt
 import pandas as pd
-import util as ut
+import util
+import indicators
+from BagLearner import BagLearner
+from RTLearner import RTLearner
+
 
 class StrategyLearner(object):
 
-    # constructor
-    def __init__(self, verbose = False, impact=0.0, commission=0.0):
+    def __init__(self, verbose=False, impact=0.0, commission=0.0, testing=False):
         self.verbose = verbose
         self.impact = impact
         self.commission = commission
+        self.testing = testing
 
-    # this method should create a QLearner, and train it for trading
-    def addEvidence(self, symbol = "IBM", \
-        sd=dt.datetime(2008,1,1), \
-        ed=dt.datetime(2009,1,1), \
-        sv = 10000):
-
-        # add your code to do learning here
-
-        # example usage of the old backward compatible util function
-        syms=[symbol]
-        dates = pd.date_range(sd, ed)
-        prices_all = ut.get_data(syms, dates)  # automatically adds SPY
-        prices = prices_all[syms]  # only portfolio symbols
-        # prices_SPY = prices_all['SPY']  # only SPY, for comparison later
-        if self.verbose: print(prices)
-
-        # example use with new colname
-        volume_all = ut.get_data(syms, dates, colname = "Volume")  # automatically adds SPY
+    def _get_volume(self):
+        """For reference."""
+        volume_all = ut.get_data(syms, dates, colname="Volume")
         volume = volume_all[syms]  # only portfolio symbols
         # volume_SPY = volume_all['SPY']  # only SPY, for comparison later
-        if self.verbose: print(volume)
+        if self.verbose:
+            print(volume)
 
-    # this method should use the existing policy and test it against new data
-    def testPolicy(self, symbol = "IBM", \
-        sd=dt.datetime(2009,1,1), \
-        ed=dt.datetime(2010,1,1), \
-        sv = 10000):
+    def _add_indicators(self, df, symbol):
+        """Add indicators for learning to DataFrame."""
+        df.drop(columns=["SPY"], inplace=True)
+        indicators.macd(df, symbol)
+        indicators.rsi(df, symbol)
+        indicators.price_sma(df, symbol, [8])
+        indicators.price_delta(df, symbol, 3)
+        df.dropna(inplace=True)
 
-        # here we build a fake set of trades
-        # your code should return the same sort of data
-        dates = pd.date_range(sd, ed)
-        prices_all = ut.get_data([symbol], dates)  # automatically adds SPY
-        trades = prices_all[[symbol,]]  # only portfolio symbols
-        # trades_SPY = prices_all['SPY']  # only SPY, for comparison later
-        trades.values[:,:] = 0 # set them all to nothing
-        trades.values[0,:] = 1000 # add a BUY at the start
-        trades.values[40,:] = -1000 # add a SELL
-        trades.values[41,:] = 1000 # add a BUY
-        trades.values[60,:] = -2000 # go short from long
-        trades.values[61,:] = 2000 # go long from short
-        trades.values[-1,:] = -1000 #exit on the last day
-        if self.verbose: print(type(trades)) # it better be a DataFrame!
-        if self.verbose: print(trades)
-        if self.verbose: print(prices_all)
-        return trades
+    def addEvidence(self, symbol="IBM",
+                    sd=dt.datetime(2008, 1, 1),
+                    ed=dt.datetime(2009, 1, 1),
+                    sv=10000):
+
+        self.indicators = ['macd_diff', 'rsi', 'price_sma_8']
+        df = util.get_data([symbol], pd.date_range(sd, ed))
+        self._add_indicators(df, symbol)
+
+        def classify_y(row):
+            if row > 0.1:
+                return 1
+            elif row < -0.1:
+                return -1
+            return 0
+
+        self.learner = RTLearner(leaf_size = 7)
+        # self.learner = BagLearner(RTLearner, 5, {'leaf_size': 5})
+        data_x = df[self.indicators].to_numpy()
+        y = df['pct_3'].apply(classify_y)
+        self.learner.addEvidence(data_x, y.to_numpy())
+        return y
+
+    def strat(self, data_y, orders):
+        self.holding = 0
+
+        def strat(row):
+            y = int(data_y.loc[row.name][0])
+            shares = 0
+            if self.holding == 0 and y == 1:
+                shares = 1000
+            elif self.holding == -1000 and y == 1:
+                shares = 2000
+            elif self.holding == 0 and y == -1:
+                shares = -1000
+            elif self.holding == 1000 and y == -1:
+                shares = -2000
+            self.holding += shares
+            return shares
+
+        orders["Shares"] = orders.apply(strat, axis=1)
+
+    def testPolicy(self, symbol="IBM",
+                   sd=dt.datetime(2009, 1, 1),
+                   ed=dt.datetime(2010, 1, 1),
+                   sv=10000):
+        df = util.get_data([symbol], pd.date_range(sd, ed))
+        self._add_indicators(df, symbol)
+        data_x = df[self.indicators].to_numpy()
+        data_y = pd.DataFrame(index=df.index, data=self.learner.query(data_x))
+
+        orders = pd.DataFrame(index=df.index)
+        orders["Symbol"] = symbol
+        orders["Order"] = ""
+        orders["Shares"] = 0
+        self.strat(data_y, orders)
+        if self.testing:
+            return orders
+        else:
+            return orders[["Shares"]]
 
-if __name__=="__main__":
-    print("One does not simply think up a strategy")

strategy_evaluation/experiment1.py

@@ -9,6 +9,7 @@ import matplotlib.pyplot as plt
 from matplotlib.widgets import MultiCursor
 from BenchmarkStrategy import BenchmarkStrategy
 from ManualStrategy import ManualStrategy
+from StrategyLearner import StrategyLearner
 
 
 def plot_indicators(symbol, df):
@@ -16,7 +17,6 @@ def plot_indicators(symbol, df):
 
     price_sma = indicators.price_sma(df, symbol, [8])
     bb = indicators.bollinger_band(df, symbol)
-    sma = indicators.sma(df, symbol, [8])
     rsi = indicators.rsi(df, symbol)
     macd = indicators.macd(df, symbol).copy()
 
@@ -57,41 +57,81 @@ def visualize_correlations(symbol, df):
     sys.exit(0)
 
 
-def experiment1():
-    symbol = "JPM"
-    start_value = 10000
-    sd = dt.datetime(2008, 1, 1)  # in-sample
-    ed = dt.datetime(2009, 12, 31)  # in-sample
-    # sd = dt.datetime(2010, 1, 1)  # out-sample
-    # ed = dt.datetime(2011, 12, 31)  # out-sample
+def compare_manual_strategies(symbol, sv, sd, ed):
 
     df = util.get_data([symbol], pd.date_range(sd, ed))
     df.drop(columns=["SPY"], inplace=True)
 
-    # visualize_correlations(symbol, df)
-    # plot_indicators(symbol, df)
-
     bs = BenchmarkStrategy()
-    orders = bs.testPolicy(symbol, sd, ed, start_value)
-    df["Benchmark"] = marketsim.compute_portvals(orders, start_value)
+    orders = bs.testPolicy(symbol, sd, ed, sv)
+    df["Benchmark"] = marketsim.compute_portvals(orders, sv)
     df["Orders Benchmark"] = orders["Shares"]
 
     ms = ManualStrategy()
-    orders = ms.testPolicy(symbol, sd, ed, start_value)
-    df["Manual"] = marketsim.compute_portvals(orders, start_value)
-    df["Orders Manual"] = orders["Shares"]
-    df["Holding Manual"] = orders["Shares"].cumsum()
+    orders = ms.testPolicy(symbol, sd, ed, sv, macd_strat=True)
+    df["MACD Strat"] = marketsim.compute_portvals(orders, sv)
+    df["Orders MACD"] = orders["Shares"]
+    # df["Holding Manual"] = orders["Shares"].cumsum()
+
+    orders = ms.testPolicy(symbol, sd, ed, sv)
+    df["Three Strat"] = marketsim.compute_portvals(orders, sv)
+    df["Orders Three"] = orders["Shares"]
 
     fig, ax = plt.subplots(3, sharex=True)
     df[[symbol]].plot(ax=ax[0])
-    df[["Benchmark", "Manual"]].plot(ax=ax[1])
-    df[["Orders Benchmark", "Orders Manual"]].plot(ax=ax[2])
+    df[["Benchmark", "MACD Strat", "Three Strat"]].plot(ax=ax[1])
+    df[["Orders Benchmark", "Orders MACD", "Orders Three"]].plot(ax=ax[2])
 
     for a in ax:
         a.grid()
-    multi = MultiCursor(fig.canvas, ax, color='r', lw=0.5)
+    MultiCursor(fig.canvas, ax, color='r', lw=0.5)
+
+    # plt.show()
+    fig.set_size_inches(10, 8, forward=True)
+    plt.savefig('figure_1.png', dpi=fig.dpi)
+
+
+def experiment1():
+    symbol = "JPM"
+    sv = 10000
+    sd = dt.datetime(2008, 1, 1)  # in-sample
+    ed = dt.datetime(2009, 12, 31)  # in-sample
+    sd_out = dt.datetime(2010, 1, 1)  # out-sample
+    ed_out = dt.datetime(2011, 12, 31)  # out-sample
+
+    df = util.get_data([symbol], pd.date_range(sd, ed_out))
+    df.drop(columns=["SPY"], inplace=True)
+
+    # visualize_correlations(symbol, df)
+    # plot_indicators(symbol, df)
+    # compare_manual_strategies(symbol, sv, sd, ed)
+
+    bs = BenchmarkStrategy()
+    orders = bs.testPolicy(symbol, sd_out, ed_out, sv)
+    df["Benchmark"] = marketsim.compute_portvals(orders, sv)
+    df["Orders Benchmark"] = orders["Shares"]
+
+    sl = StrategyLearner(testing=True)
+    sl.addEvidence(symbol, sd, ed, sv)
+    orders = sl.testPolicy(symbol, sd_out, ed_out, sv)
+    df["SL"] = marketsim.compute_portvals(orders, sv)
+    df["Orders SL"] = orders["Shares"]
+
+    fig, ax = plt.subplots(3, sharex=True)
+    df[[symbol]].plot(ax=ax[0])
+    df[["Benchmark", "SL"]].plot(ax=ax[1])
+    df[["Orders Benchmark", "Orders SL"]].plot(ax=ax[2])
+
+    for a in ax:
+        a.grid()
+    MultiCursor(fig.canvas, ax, color='r', lw=0.5)
     plt.show()
-    # plt.savefig('figure_1.png')
+
+    # For debugging the classification learner:
+    # df["y_train"] = sl.addEvidence(symbol, sd, ed, sv)
+    # df["y_query"] = sl.testPolicy(symbol, sd, ed, sv)
+    # df[["y_train", "y_query"]].plot(ax=ax[1])
+
 
 
 if __name__ == "__main__":

strategy_evaluation/indicators.py

@@ -73,7 +73,7 @@ def rsi(df, symbol, period=14):
                            (avg_loss / period))))
         return rsi
 
-    key = f"rsi"
+    key = "rsi"
     # Add one to get 'period' price changes (first change is nan).
     period += 1
     df[key] = df[symbol].rolling(period).apply(rsi)
@@ -91,13 +91,6 @@ def macd(df, symbol):
     return df[[k1, k2, k3]]
 
 
-def price_delta(df, symbol, period=1):
-    """Calculate delta between previous day and today."""
-    k = f"diff_{period}"
-    df[k] = df[symbol].diff(periods=period)
-    return df[k]
-
-
 def price_delta(df, symbol, period=1):
     """Calculate percentage change for period."""
     k = f"pct_{period}"