162 ffd.py
View file
@@ -18,6 +18,7 @@
import itertools as it
import pprint as pp
import progressbar
import re

# standardize data frame
def read_fanduel_salaries(filename):
@@ -32,7 +33,7 @@ def read_fanduel_salaries(filename):
def read_fanduel_player_data(filename, player_info="all"):
cols = ["First Name", "Last Name"]
if player_info == "all":
player_info = ["Base Projection", "Injury Status"]
player_info = ["Base Projection", "Injury Status", "Team", "Opponent"]

# requested parameters don't always match column names
if "Base Projection" in player_info:
@@ -41,6 +42,12 @@ def read_fanduel_player_data(filename, player_info="all"):
if "Injury Status" in player_info:
cols.append("Injury Indicator")

if "Team" in player_info:
cols.append("Team")

if "Opponent" in player_info:
cols.append("Opponent")

# read .csv file
df = pd.read_csv(filename, usecols=cols)

@@ -54,6 +61,7 @@ def read_fanduel_player_data(filename, player_info="all"):
try:
df["Base Projection"] = df["FPPG"]
df.pop("FPPG")
df["Base Projection Source"] = "Fanduel"
except:
pass

@@ -86,6 +94,7 @@ def read_fantasypros_player_data(filename, player_info="all"):
try:
df["Base Projection"] = df["fpts"]
df.pop("fpts")
df["Base Projection Source"] = "FantasyPros"
except:
pass

@@ -108,15 +117,40 @@ def compute_derived_data(df):

def tune_player_projections(df, models=None):
# example model input format
# models = {
# <parameter_key>: {
# <parameter_value>: {
# 'b1': <B1>,
# 'b0': <B0>
# }
# models = [
# {
# 'cols': [
# {
# 'name': <parameter_key_0>,
# 'value': <parameter_value_0>,
# 'invert': 0
# },
# {
# 'name': <parameter_key_1>,
# 'value': <parameter_value_1>,
# 'invert': 0
# }
# ],
# 'b1': 1,
# 'b0': 0
# },
# ...
# }
# {
# 'cols': [
# {
# 'name': <parameter_key_0>,
# 'value': <parameter_value_0>,
# 'invert': 0
# },
# {
# 'name': <parameter_key_1>,
# 'value': <parameter_value_1>,
# 'invert': 0
# }
# ],
# 'b1': 1,
# 'b0': 0
# }
# ]
#
# parameter_key: column in data frame (e.g., Position)
# parameter_value: value for this player (e.g., RB)
@@ -126,19 +160,59 @@ def tune_player_projections(df, models=None):
# is first-order and independent.
df["Model Offset"] = 0

for key in models:
for val in models[key]:
print("Applying model for " + key + " = " + val)
df.loc[df[key] == val, "Model Offset"] += (df.loc[df[key] == val, "Base Projection"] * (models[key][val]["b1"] - 1) + models[key][val]["b0"])
if models is not None:
for model in models:
print("Applying model:")
pp.pprint(model)
# todo: any better way to do this?
first = 1
for col_filter in model["cols"]:
pp.pprint(col_filter)
idx = (df[col_filter["name"]] == col_filter["value"])
if col_filter["invert"] == 1:
idx = -idx
if first == 1:
first = 0
df_filter = idx
else:
df_filter &= idx
df.loc[df_filter, "Model Offset"] += \
(df.loc[df_filter, "Base Projection"] *
(model["b1"] - 1) +
model["b0"])

df["Modeled Projection"] = df["Base Projection"] + df["Model Offset"]
return df

def tune_lineup_projections(df, models=None):
# todo: define input format for models once they're actually supported
df["Model Offset"] = 0

if models is not None:
for model in models:
print("Applying model:")
pp.pprint(model)
# todo: any better way to do this?
first = 1
for col_filter in model["cols"]:
pp.pprint(col_filter)
if "." in col_filter["value"]:
idx = (df[col_filter["name"]] == df[col_filter["value"][1:]])
else:
idx = (df[col_filter["name"]] == col_filter["value"])
if col_filter["invert"] == 1:
idx = -idx
if first == 1:
first = 0
df_filter = idx
else:
df_filter &= idx
df.loc[df_filter, "Model Offset"] += \
(df.loc[df_filter, "Total Modeled Projection"] *
(model["b1"] - 1) +
model["b0"])

df["Total Modeled Projection"] += df["Model Offset"]

# still in development. passthrough for now.
df["Total Modeled Projection"] = df["Total Base Projection"]
return df

def anal_lineups(data_df, lineup_df, fom, anal_info):
@@ -186,8 +260,6 @@ def report_data(data_df, anal_res, fom, anal_info):
# todo: don't hardcode positions
positions = ["QB", "RB", "WR", "TE", "K", "D"]

print(stats_df)

print("\n\nRelative frequencies (based on top " + str(anal_info["max_lineups"]) + " lineups):")
stats_df.sort("Frequency", ascending=False, inplace=True)
for pos in positions:
@@ -198,8 +270,24 @@ def report_data(data_df, anal_res, fom, anal_info):
str(player_info["Frequency"]))

trimmed_df = trimmed_df.head(anal_info["top_lineups"])

report_cols = (
"QB_1",
"RB_1",
"RB_2",
"WR_1",
"WR_2",
"WR_3",
"TE_1",
"K_1",
"D_1",
"Total Salary",
"Total Base Projection",
"Total Modeled Projection"
)

print("\n\nTop " + str(anal_info["top_lineups"]) + " lineups:")
print(trimmed_df)
print(trimmed_df.loc[:,report_cols])

# filter data
# todo: there's a better way to do this. but fuck it
@@ -224,6 +312,8 @@ def filter_df(df, filters):
pos_filter &= (df["Injury Status"] != "O")
if filters[pos]["exclude_injury"]["injured_reserve"] == 1:
pos_filter &= (df["Injury Status"] != "IR")
if filters[pos]["exclude_injury"]["not_active"] == 1:
pos_filter &= (df["Injury Status"] != "NA")

# filtered_idx is the OR of all filters
if first == 1:
@@ -254,21 +344,41 @@ def verify_lineup(player_dict, rules, lineup):
"K_1": lineup[4][0],
"D_1": lineup[5][0],
"Total Salary": 0,
"Total Base Projection": 0
"Total Base Projection": 0,
"Total Modeled Projection": 0
}

# also include per-player info for lineup models
# using first entry in dictionary - all players have the same parameters
# todo: only need to do this once. determine list of column names outside
# of loop
for pos in lineup_dict.keys():
for param in player_dict[player_dict.keys()[1]]:
col_name = pos + " " + param
lineup_dict[col_name] = None

valid_lineup = 1
for pos in lineup_dict:
if pos == "Total Salary" or pos == "Total Base Projection":
m = re.search("_(\d)$", pos)
if m is None:
continue
depth = m.group(1)

name = lineup_dict[pos]
lineup_dict["Total Salary"] += player_dict[name]["Salary"]
lineup_dict["Total Base Projection"] += player_dict[name]["Base Projection"]
lineup_dict["Total Modeled Projection"] += player_dict[name]["Modeled Projection"]

# todo: break immediately if over salary cap
if lineup_dict["Total Salary"] > rules["cap"]:
valid_lineup = 0
break

# add per-player info for lineup models
for param in player_dict[name]:
col_name = pos + " " + param
lineup_dict[col_name] = player_dict[name][param]

if valid_lineup == 1:
# create data frame with dummy index (ignored by concat)
result = lineup_dict
@@ -399,6 +509,10 @@ def gen_lineup_df(df, rules):
# really, this is just used to trim down the dataset to be more manageable.
data_df = filter_df(data_df, config_info["filters"])

# save data frame
data_df.to_csv(config_info["output"]["filenames"]["players"], index=False,
index_label="Name")

# build permutations, generate list of "valid" lineups
# todo: currently, a "valid" lineup is only defined as a team that is under
# the salary cap.
@@ -410,13 +524,13 @@ def gen_lineup_df(df, rules):

# anal-eyes top lineups
res = anal_lineups(data_df, lineup_df, "Total Modeled Projection",
config_info["anal"])
config_info["output"])

# report anal-isis
report_data(data_df, res, "Total Modeled Projection", config_info["anal"])
report_data(data_df, res, "Total Modeled Projection", config_info["output"])

# we should really just report the sorted and trimmed lineup list
lineup_df = res["trimmed_df"]
lineup_df.to_csv("/tmp/poop.csv", index=False)
lineup_df.to_csv(config_info["output"]["filenames"]["lineups"], index=False)