cmap2 = sns.color_palette('Reds_r', n_colors=len(gammas)*2)
ax.plot(months, cum_rets(e_perf), alpha=alpha,
label='ETE', lw=lw, c='steelblue')
ax.plot(months, cum_rets(mv_perf), alpha=alpha,
label='MV', lw=lw, c='forestgreen')
for i, gamma in enumerate(reversed(gammas[1:])):
ax.plot(months, cum_rets(capm[gamma]), alpha=alpha,
label=f'CAPM $\\gamma={gamma}$', lw=lw, c=cmap2[i])
fig.autofmt_xdate()
plt.ylabel('Cumulative Return')
plt.xlabel('Time')
plt.legend()
plt.tight_layout()
eu.save_fig(f'../plots/portfolio_comparison', dpi=300)
plt.show()
eu.latex_figure(f'../plots/portfolio_comparison')
# %%
tbl = pd.DataFrame({
'ETE': e_perf,
'MV': mv_perf,
'CAPM 1': capm[1],
'CAPM 10': capm[10],
}, index=months[1:])
tbl = (tbl.mean()*4).to_frame().join((tbl.std()*np.sqrt(4)).to_frame(),
rsuffix='vol')
tbl.columns = 'Return Volatility'.split()
tbl['Sharpe'] = tbl['Return']/tbl['Volatility']
def plot_temporal_vol_and_matrices(assets):
mod = TransferEntropy(assets=assets)
periods = [
('7/1/2011', '8/1/2011'),
('8/1/2011', '9/1/2011'),
('9/1/2011', '10/1/2011'),
('10/1/2011', '11/1/2011'),
('11/1/2011', '12/1/2011'),
('12/1/2011', '1/1/2012'),
('1/1/2012', '2/1/2012'),
]
dates = [pd.to_datetime(p[0]) for p in periods]
xlabs = [date.strftime('%b %Y') for date in dates]
df = mod.data
df = df**2
df = df.resample('M').sum()
df = (df.mean(axis=1) * 12)**0.5
df = df[(df.index >= pd.to_datetime('6/1/2011'))
& (df.index <= pd.to_datetime(periods[-1][-1]))]
fig, ax = plt.subplots(1, 1, figsize=(20, 3))
ax.plot(df, c='steelblue')
tick = mtick.StrMethodFormatter('{x:.0%}')
ax.yaxis.set_major_formatter(tick)
ax.set_xlabel('Date')
ax.set_ylabel('Annualized Volatility')
eu.save_fig('temporal_vol', dir='../figures')
plt.show()
# %%
# Find maximum transfer entropy for all periods.
te_max, ete_max, ete_min = 0, 0, 0
for i, period in enumerate(periods):
mod.set_timeperiod(*period)
mod.compute_transfer_entropy()
te_max = max(np.max(mod.te.values), te_max)
mod.compute_effective_transfer_entropy(sims=3)
ete_max = max(np.max(mod.ete.values), ete_max)
ete_min = min(np.min(mod.ete.values), ete_min)
# %%
fig, axes = plt.subplots(3, len(periods), figsize=(20, 10),
sharex=True, sharey=True)
for i, period in enumerate(periods):
mod.set_timeperiod(*period)
mod.plot_corr(method='pearson', ax=axes[0, i],
cbar=False, labels=False)
mod.compute_transfer_entropy()
mod.plot_te(ax=axes[1, i], cbar=False, labels=False,
vmin=0, vmax=te_max)
mod.compute_effective_transfer_entropy(sims=5)
mod.plot_te(ax=axes[2, i], te='ete', cbar=False, labels=False,
vmin=ete_min, vmax=0.75*te_max)
axes[2, i].set_xlabel(xlabs[i])
axes[0, 0].set_ylabel('Correlation')
axes[1, 0].set_ylabel('Transfer Entropy')
axes[2, 0].set_ylabel('Effective Transfer Entropy')
plt.tight_layout()
# eu.save_fig('temporal_matrices', dir='../figures')
plt.show()
def make_algo_comparison_tables(year=2018, fdir='../figures', save=True):
# %%
ess_train = CompareModels('essential', period='weekly', subset='train')
ess_cv = CompareModels(stats='essential', period='weekly', subset='cv')
ess_test = CompareModels(stats='essential', period='weekly', subset='test')
ess_thresh = CompareModels(
'essential_below_thresh',period='weekly', subset='train')
non_ess = CompareModels('non_essential', period='weekly', subset='train')
# %%
# Essential stats.
fig, axes=plt.subplots(1, 2, figsize=(18, 12))
ess_train.plot_color_table(ax=axes[0], fontsize=12, prec=2)
ess_cv.plot_color_table(ax=axes[1], fontsize=12, prec=2)
plt.tight_layout()
if save:
eu.save_fig('essential_MAE_color_table', dir=fdir)
else:
plt.show()
ess_train.plot_color_table(figsize=(9, 12), fontsize=12, prec=2)
if save:
eu.save_fig('essential_train_MAE_table', dir=fdir)
else:
plt.show()
ess_cv.plot_color_table(figsize=(9, 12), fontsize=12, prec=2)
if save:
eu.save_fig('essential_cv_MAE_table', dir=fdir)
cap = 'Mean Absolute Error (MAE) values for all essential stats '\
'and studied algorithms.'
eu.latex_figure(
fids=['essential_train_MAE_table', 'essential_cv_MAE_table'],
dir=fdir,
caption=cap,
subcaptions=['Training MAE.', 'Cross-Validation MAE.'],
width=0.98,
)
else:
plt.show()
ess_thresh.plot_color_table(figsize=(9, 12), fontsize=12)
if save:
eu.save_fig('essential_thresh_MAE_table', dir=fdir)
else:
plt.show()
non_ess.plot_color_table(figsize=(9, 12), fontsize=12, prec=3)
if save:
eu.save_fig('nonessential_MAE_table', dir=fdir)
else:
plt.show()
# %%
# Make train/cv/test table and print to LaTeX.
with open('../data/.models/weekly/optimal_models.json', 'r') as fid:
opt_models = json.load(fid)
def get_set_mae_list(subset_comparison, models=opt_models):
df = subset_comparison.mae_df
mae_list = []
for _, row in df.iterrows():
mae_list.append(row[models[row['Position']][row['Stat']]])
return mae_list
df = ess_train.mae_df
fmt_methods = {
'MEAN': 'Mean',
'MEDIAN': 'Median',
'FLOOR': 'Min',
'CEIL': 'Max',
}
algos = []
for _, row in df.iterrows():
algo = models[row['Position']][row['Stat']]
algos.append(fmt_methods.get(algo, algo))
pos = []
for p in df['Position'].values:
pos.append('{}' if p in pos else p)
table = pd.DataFrame({
'Position': pos,
'Stat': df['Stat'],
'Projection Function': algos,
'Train': get_set_mae_list(ess_train),
'CV': get_set_mae_list(ess_cv),
'Test': get_set_mae_list(ess_test),
})
cap = 'MAE values for training, cross-validation, and test sets with ' \
'selected optimal projection function for each stat.'
eu.latex_print(table, hide_index=True, prec=2, col_fmt='lllcrrr',
caption=cap)
def plot_missing_data(fdir='../figures', save=True):
"""Plot # of sources and % data missing for each essential stat."""
stats = {
'QB': ['Pass Yds', 'Pass TD', 'Pass Int', 'Rush Yds', 'Rush TD'],
'RB': ['Rush Yds', 'Rush TD', 'Receptions', 'Rec Yds', 'Rec TD'],
'WR': ['Rush Yds', 'Rush TD', 'Receptions', 'Rec Yds', 'Rec TD'],
'TE': ['Receptions', 'Rec Yds', 'Rec TD'],
'DST': ['PA', 'YdA', 'TD', 'Sack', 'Int', 'Fum Rec'],
}
n_stats = sum(len(s) for s in stats.values())
n_sources = np.zeros(n_stats)
pct_nan = np.zeros(n_stats)
x_ticks = []
i = 0
for pos in stats.keys():
proj = TrainProjections(pos)
proj.load_data(weeks=range(1,18), impute_method=False)
for stat in stats[pos]:
df = proj._stats_df[stat]
# Subset DataFrame to only include only projection columns.
ignored_cols = ['Player', 'Team', 'Pos', 'Week', 'STATS']
keep_cols = [c for c in list(df) if c not in ignored_cols]
proj_df = df[keep_cols].copy()
# Find percentage of NaNs in for the stat and # of sources.
n, m = proj_df.shape
pct_nan[i] = np.sum(np.sum(proj_df.isnull())) / (n * m)
n_sources[i] = m
x_ticks.append(f'{pos} - {stat}')
i += 1
# Creat figure.
fig, ax = plt.subplots(1, 1, figsize=[12, 6])
x = np.arange(n_stats)
ax.bar(x, n_sources, color='steelblue', alpha=0.7)
ax.bar(x, n_sources*pct_nan, color='firebrick', alpha=0.9)
for xt, yt, s in zip(x, n_sources*pct_nan, pct_nan):
ax.text(xt, yt+0.05, f'{s:.1%}', color='firebrick', ha='center',
fontsize=7)
ax.set_ylabel('Number of Sources')
ax.set_xlabel('Projected Stat')
plt.xticks(x, x_ticks, rotation=45, ha='right')
ax.xaxis.grid(False)
# Save figure.
if save:
fid = 'missing_data'
eu.save_fig(fid, dir=fdir)
cap = 'Number of sources collected for each essential stat (blue). Red '
cap += 'indicates percentage of missing data for each respective stat.'
eu.latex_figure(fid, dir=fdir, width=0.95, caption=cap)
# Nonessential stats.
stats = {
'QB': ['Receptions', 'Rec Yds', 'Rec TD', '2PT'],
'RB': ['Pass Yds', 'Pass TD', 'Pass Int', '2PT'],
'WR': ['Pass Yds', 'Pass TD', 'Pass Int', '2PT'],
'TE': ['Pass Yds', 'Pass TD', 'Pass Int', 'Rush Yds', 'Rush TD', '2PT'],
'DST': ['Saf', 'Blk'],
}
n_stats = sum(len(s) for s in stats.values())
n_sources = np.zeros(n_stats)
pct_nan = np.zeros(n_stats)
x_ticks = []
i = 0
for pos in stats.keys():
proj = TrainProjections(pos)
proj.load_data(weeks=range(1,18), impute_method=False)
for stat in stats[pos]:
df = proj._stats_df[stat]
# Subset DataFrame to only include only projection columns.
ignored_cols = ['Player', 'Team', 'Pos', 'Week', 'STATS']
keep_cols = [c for c in list(df) if c not in ignored_cols]
proj_df = df[keep_cols].copy()
# Find percentage of NaNs in for the stat and # of sources.
n, m = proj_df.shape
pct_nan[i] = np.sum(np.sum(proj_df.isnull())) / (n * m)
n_sources[i] = m
x_ticks.append(f'{pos} - {stat}')
i += 1
# Creat figure.
fig, ax = plt.subplots(1, 1, figsize=[12, 6])
x = np.arange(n_stats)
ax.bar(x, n_sources, color='steelblue', alpha=0.7)
ax.bar(x, n_sources*pct_nan, color='firebrick', alpha=0.9)
for xt, yt, s in zip(x, n_sources*pct_nan, pct_nan):
ax.text(xt, yt+0.05, f'{s:.1%}', color='firebrick', ha='center',
fontsize=7)
ax.set_ylabel('Number of Sources')
ax.set_xlabel('Projected Stat')
plt.yticks(np.arange(max(n_sources)+1))
plt.xticks(x, x_ticks, rotation=45, ha='right')
ax.xaxis.grid(False)
# Save figure.
if save:
fid = 'nonessential_missing_data'
eu.save_fig(fid, dir=fdir)
cap = 'Number of sources collected for each nonessential stat (blue). '
cap += 'Red indicates percentage of missing data for each '
cap += 'respective stat.'
eu.latex_figure(fid, dir=fdir, width=0.95, caption=cap)
def plot_stat_hists(fdir='../figures', save=True):
"""
Plot all essential stat histograms for the appendix and
an example histograms figure for the main body of the paper.
"""
# Load data.
positions = 'QB RB WR TE DST'.split()
stats = {pos: TrainProjections(pos) for pos in positions}
for pos in positions:
stats[pos].load_data(weeks=range(1, 18))
# Plot all no threshold histograms for appendix.
no_thresh_fids = {pos: f'no_threshold_hist_{pos}' for pos in positions}
for pos in positions:
n = len(stats[pos].essential_stats)
fig, axes = plt.subplots(n, 1, figsize=[6, 2.5*n])
for stat, ax in zip(stats[pos].essential_stats, axes.flat):
stats[pos].plot_projection_hist(
stat, bins=None, threshold=False, ax=ax)
plt.tight_layout()
if save:
eu.save_fig(no_thresh_fids[pos], dir=fdir)
else:
plt.show()
# Plot all threshold histograms for appendix.
thresh_fids = {pos: f'threshold_hist_{pos}' for pos in positions}
for pos in positions:
n = len(stats[pos].essential_stats)
fig, axes = plt.subplots(n, 1, figsize=[6, 2.5*n])
for stat, ax in zip(stats[pos].essential_stats, axes.flat):
stats[pos].plot_projection_hist(
stat, bins=None, threshold=True, ax=ax)
plt.tight_layout()
if save:
eu.save_fig(thresh_fids[pos], dir=fdir)
else:
plt.show()
# Create LaTeX code to plot figures.
if save:
for pos in positions:
cap = 'Essential stat raw histograms and thresholded histograms '
cap += f'for {pos}.'
eu.latex_figure(
fids=[no_thresh_fids[pos], thresh_fids[pos]],
dir=fdir,
subcaptions=['Raw histogram with threshold (red).',
'Histogram above threshold.'],
caption=cap,
width=0.9,
)
print()
print('\pagebreak')
# Plot raw histograms for example stats.
fids = ['no_theshold_example_hists', 'no_theshold_example_hists_RB']
fig, axes = plt.subplots(2, 1, figsize=[6, 9])
for pos, stat, ax in zip(['QB', 'RB'], ['Pass Yds', 'Rush Yds'], axes.flat):
stats[pos].plot_projection_hist(
stat, bins=None, threshold=False, ax=ax)
plt.tight_layout()
if save:
eu.save_fig(fids[0], dir=fdir)
else:
plt.show()
# Plot thresholded histograms for example stats.
fids = ['no_theshold_example_hists', 'no_theshold_example_hists_RB']
fig, axes = plt.subplots(2, 1, figsize=[6, 9])
for pos, stat, ax in zip(['QB', 'RB'], ['Pass Yds', 'Rush Yds'], axes.flat):
stats[pos].plot_projection_hist(
stat, bins=None, threshold=True, ax=ax)
plt.tight_layout()
if save:
eu.save_fig(fids[1], dir=fdir)
else:
plt.show()
# Create LaTeX code to plot example stats figure.
if save:
print('\n\n\n')
cap = 'Example raw and thresholded histograms for QB passing yards '
cap += f'and RB rushing yards.'
eu.latex_figure(
fids=fids,
dir=fdir,
subcaptions=['Raw histogram with threshold (red).',
'Histogram above threshold.'],
caption=cap,
width=0.9,
)
ax.set_xlabel('Projected Points')
plt.ylim(n_players+2, -2)
plt.tight_layout()
# %%
for i in range(1, 6):
plot_weekly_pos_projections('TE', i)
plt.show()
# %%
pos = 'TE'
week = 17
plot_weekly_pos_projections(pos, week, figsize=(12, 8))
# plt.show()
eu.save_fig(f'{pos}_{week}', dir=fdir)
# %%
# %%
class CompareModels:
"""
Load data for model comparison and compare MAE values across all
models for given stat.
Parameters
----------
stats: {'essential', 'essential_below_thresh', 'non_essential'}
Stats to compare.
period: {'weekly', 'season'}
Season or weekly models.
subset: {'train', 'cv', 'test'}
def main():
n_sims = 50
essential_stats = False
impute_methods = [
# 'BiScaler',
'IterativeImpute',
# 'IterativeSVD',
'KNN',
# 'MatrixFactorization',
'Mean',
'Median',
# 'NuclearNorm',
'SoftImpute',
]
if essential_stats:
impute_stats = {
'QB': ['Pass Yds', 'Pass TD', 'Pass Int', 'Rush Yds', 'Rush TD'],
'RB': ['Rush Yds', 'Rush TD', 'Receptions', 'Rec Yds', 'Rec TD'],
'WR': ['Rush Yds', 'Rush TD', 'Receptions', 'Rec Yds', 'Rec TD'],
'TE': ['Receptions', 'Rec Yds', 'Rec TD'],
'DST': ['PA', 'YdA', 'TD', 'Sack', 'Int', 'Fum Rec'],
}
fid_MAE = '../figures/impute_MAE'
fid_RMSE = '../figures/impute_RMSE'
else:
impute_stats = {
'QB': ['Receptions', 'Rec Yds', 'Rec TD', '2PT'],
'RB': ['Pass Yds', 'Pass TD', 'Pass Int', '2PT'],
'WR': ['Pass Yds', 'Pass TD', 'Pass Int', '2PT'],
'TE':
['Pass Yds', 'Pass TD', 'Pass Int', 'Rush Yds', 'Rush TD', '2PT'],
'DST': ['Saf', 'Blk'],
}
fid_MAE = '../figures/nonessential_impute_MAE'
fid_RMSE = '../figures/nonessential_impute_RMSE'
j = 0
x_ticks = []
n_stats = sum(len(stats) for stats in impute_stats.values())
mae = np.zeros([len(impute_methods), n_stats])
rmse = np.zeros([len(impute_methods), n_stats])
with tqdm(total=n_stats) as pbar:
for pos in impute_stats.keys():
proj = TrainProjections(pos)
proj.load_data(weeks=range(1, 18), impute_method=False)
for stat in impute_stats[pos]:
df = proj._stats_df[stat]
# Subset DataFrame to only include only projection columns.
ignored_cols = ['Player', 'Team', 'Pos', 'Week', 'STATS']
impute_cols = [c for c in list(df) if c not in ignored_cols]
proj_df = df[impute_cols].copy()
# Find percentage of NaNs in for the stat.
n, m = proj_df.shape
nan_pct = np.sum(np.sum(proj_df.isnull())) / (n * m)
# Remove rows with NaNs.
proj_df.dropna(how='any', axis=0, inplace=True)
print(f'\n{pos} - {stat}\n')
# Get average RMSE and MAE for each imputation method.
for i, method in enumerate(impute_methods):
try:
rmse[i, j], mae[i, j] = simulate_imputing(
proj_df.copy(), method, n_sims, nan_pct)
except ValueError:
pass
x_ticks.append(f'{pos} - {stat}')
j += 1
pbar.update(1)
# %%
mae_df = pd.DataFrame(mae, columns=x_ticks, index=impute_methods)
mae_df = (mae_df - mae_df.mean()) / mae_df.std()
plot_heatmap(mae_df, cbar_label='Normalized MAE')
eu.save_fig(fid_MAE)
plt.show()
# %%
rmse_df = pd.DataFrame(rmse, columns=x_ticks, index=impute_methods)
rmse_df = (rmse_df - rmse_df.mean()) / rmse_df.std()
plot_heatmap(rmse_df, cbar_label='Normalized RMSE')
eu.save_fig(fid_RMSE)
plt.show()