#!/usr/bin/env python
# coding: utf-8

# # Zipline algorithm analysis example in pyfolio
# 
# Here's an example where we run an algorithm with zipline, then produce tear sheets for that algorithm.

# ## Imports
# Import pyfolio and zipline, and ingest the pricing data for backtesting.

# In[1]:


import sys
sys.path.append('/Users/george/Desktop/pyfolio/')


# In[ ]:


import pyfolio as pf
get_ipython().run_line_magic('matplotlib', 'inline')

# silence warnings
import warnings
warnings.filterwarnings('ignore')

import zipline
get_ipython().run_line_magic('load_ext', 'zipline')
get_ipython().system('zipline ingest')


# ## Run our zipline algorithm
# This algorithm can also be adjusted to execute a modified, or completely different, trading strategy.

# In[2]:


get_ipython().run_cell_magic('zipline', '--start 2004-1-1 --end 2010-1-1 -o results.pickle', '\n# Zipline trading algorithm\n# Taken from zipline.examples.olmar\n\nimport numpy as np\n\nfrom zipline.finance import commission, slippage\n\nSTOCKS = [\'AMD\', \'CERN\', \'COST\', \'DELL\', \'GPS\', \'INTC\', \'MMM\']\n\n\n# On-Line Portfolio Moving Average Reversion\n\n# More info can be found in the corresponding paper:\n# http://icml.cc/2012/papers/168.pdf\ndef initialize(algo, eps=1, window_length=5):\n    algo.stocks = STOCKS\n    algo.sids = [algo.symbol(symbol) for symbol in algo.stocks]\n    algo.m = len(algo.stocks)\n    algo.price = {}\n    algo.b_t = np.ones(algo.m) / algo.m\n    algo.eps = eps\n    algo.window_length = window_length\n\n    algo.set_commission(commission.PerShare(cost=0))\n    algo.set_slippage(slippage.FixedSlippage(spread=0))\n\n\ndef handle_data(algo, data):\n    m = algo.m\n\n    x_tilde = np.zeros(m)\n    b = np.zeros(m)\n\n    # find relative moving average price for each asset\n    mavgs = data.history(algo.sids, \'price\', algo.window_length, \'1d\').mean()\n    for i, sid in enumerate(algo.sids):\n        price = data.current(sid, "price")\n        # Relative mean deviation\n        x_tilde[i] = mavgs[sid] / price\n\n    ###########################\n    # Inside of OLMAR (algo 2)\n    x_bar = x_tilde.mean()\n\n    # market relative deviation\n    mark_rel_dev = x_tilde - x_bar\n\n    # Expected return with current portfolio\n    exp_return = np.dot(algo.b_t, x_tilde)\n    weight = algo.eps - exp_return\n    variability = (np.linalg.norm(mark_rel_dev)) ** 2\n\n    # test for divide-by-zero case\n    if variability == 0.0:\n        step_size = 0\n    else:\n        step_size = max(0, weight / variability)\n\n    b = algo.b_t + step_size * mark_rel_dev\n    b_norm = simplex_projection(b)\n    np.testing.assert_almost_equal(b_norm.sum(), 1)\n\n    rebalance_portfolio(algo, data, b_norm)\n\n    # update portfolio\n    algo.b_t = b_norm\n\n\ndef rebalance_portfolio(algo, data, desired_port):\n    # rebalance portfolio\n    for i, sid in enumerate(algo.sids):\n        algo.order_target_percent(sid, desired_port[i])\n\n\ndef simplex_projection(v, b=1):\n    """Projection vectors to the simplex domain\n    Implemented according to the paper: Efficient projections onto the\n    l1-ball for learning in high dimensions, John Duchi, et al. ICML 2008.\n    Implementation Time: 2011 June 17 by Bin@libin AT pmail.ntu.edu.sg\n    Optimization Problem: min_{w}\\| w - v \\|_{2}^{2}\n    s.t. sum_{i=1}^{m}=z, w_{i}\\geq 0\n    Input: A vector v \\in R^{m}, and a scalar z > 0 (default=1)\n    Output: Projection vector w\n    :Example:\n    >>> proj = simplex_projection([.4 ,.3, -.4, .5])\n    >>> print(proj)\n    array([ 0.33333333, 0.23333333, 0. , 0.43333333])\n    >>> print(proj.sum())\n    1.0\n    Original matlab implementation: John Duchi (jduchi@cs.berkeley.edu)\n    Python-port: Copyright 2013 by Thomas Wiecki (thomas.wiecki@gmail.com).\n    """\n\n    v = np.asarray(v)\n    p = len(v)\n\n    # Sort v into u in descending order\n    v = (v > 0) * v\n    u = np.sort(v)[::-1]\n    sv = np.cumsum(u)\n\n    rho = np.where(u > (sv - b) / np.arange(1, p + 1))[0][-1]\n    theta = np.max([0, (sv[rho] - b) / (rho + 1)])\n    w = (v - theta)\n    w[w < 0] = 0\n    return w\n')


# ## Extract metrics
# Get the returns, positions, and transactions from the zipline backtest object.

# In[3]:


import pandas as pd

results = pd.read_pickle('results.pickle')
returns, positions, transactions = pf.utils.extract_rets_pos_txn_from_zipline(results)


# ## Single plot example
# Make one plot of the top 5 drawdown periods.

# In[4]:


pf.plot_drawdown_periods(returns, top=5).set_xlabel('Date')


# ## Full tear sheet example
# Create a full tear sheet for our algorithm. As an example, set the live start date to something arbitrary.

# In[5]:


pf.create_full_tear_sheet(returns, positions=positions, transactions=transactions,
                          live_start_date='2009-10-22', round_trips=True)


# ## Suppressing symbol output
# 
# When sharing tear sheets it might be undesirable to display which symbols where used by a strategy. To suppress these in the tear sheet you can pass `hide_positions=True`.

# In[6]:


pf.create_full_tear_sheet(returns, positions=positions, transactions=transactions,
                          live_start_date='2009-10-22', hide_positions=True)


# In[ ]: