-
Notifications
You must be signed in to change notification settings - Fork 9
/
Copy pathlinearregression-independent2.py
185 lines (132 loc) · 7.25 KB
/
linearregression-independent2.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
"""
Private Score: 0.16405, Public Score: 0.14499
"""
import datetime as dt
import pandas as pd
from pandas import Series, DataFrame
import numpy as np
from sklearn.linear_model import LinearRegression
from sklearn.model_selection import train_test_split
pd.options.mode.chained_assignment = None
################################################################
# Import CSV Data into Pandas DataFrames #
################################################################
training_df = pd.read_csv("data/train.csv")
# store_df = pd.read_csv("data/store.csv")
test_df = pd.read_csv("data/test.csv")
# print(training_df.head())
# print(store_df.head())
# print(test_df.head())
################################################################
# Process Data (Universal) #
################################################################
def is_nan(val):
return val != val
def less_than_ten(val):
if int(val) < 10:
return "0" + val
else:
return val
############################################
# training_df & test_df #
############################################
# Fill NaN values in test_df with Open = 1 if DayOfWeek != 7
test_df["Open"][is_nan(test_df["Open"])] = (test_df["DayOfWeek"] != 7).astype(int)
# Create "Year" & "Month" columns
# training_df["Year"] = training_df["Date"].apply(lambda x: dt.datetime.strptime(x, "%Y-%m-%d").year)
# training_df["Month"] = training_df["Date"].apply(lambda x: dt.datetime.strptime(x, "%Y-%m-%d").month)
# test_df["Year"] = test_df["Date"].apply(lambda x: dt.datetime.strptime(x, "%Y-%m-%d").year)
# test_df["Month"] = test_df["Date"].apply(lambda x: dt.datetime.strptime(x, "%Y-%m-%d").month)
# Create "YearMonth" column
# training_df["YearMonth"] = training_df["Date"].apply(lambda x: str(dt.datetime.strptime(x, "%Y-%m-%d").year) + "-" + less_than_ten(str(dt.datetime.strptime(x, "%Y-%m-%d").month)))
# test_df["YearMonth"] = test_df["Date"].apply(lambda x: str(dt.datetime.strptime(x, "%Y-%m-%d").year) + "-" + less_than_ten(str(dt.datetime.strptime(x, "%Y-%m-%d").month)))
# "StateHoliday" has values "0" & 0
training_df["StateHoliday"].loc[training_df["StateHoliday"] == 0] = "0"
test_df["StateHoliday"].loc[test_df["StateHoliday"] == 0] = "0"
# Create "StateHolidayBinary" column
# training_df["StateHolidayBinary"] = training_df["StateHoliday"].map({0: 0, "0": 0, "a": 1, "b": 1, "c": 1})
# test_df["StateHolidayBinary"] = test_df["StateHoliday"].map({0: 0, "0": 0, "a": 1, "b": 1, "c": 1})
# One-hot encoding of "DayOfWeek" & "StateHoliday" columns
training_df = pd.get_dummies(training_df, columns=["DayOfWeek"])
test_df = pd.get_dummies(test_df, columns=["DayOfWeek"])
############################################
# store_df #
############################################
# Fill NaN values in store_df for "CompetitionDistance" = 0 (since no record exists where "CD" = NaN & "COS[Y/M]" = !NaN)
# store_df["CompetitionDistance"][is_nan(store_df["CompetitionDistance"])] = 0
# Fill NaN values in store_df for "CompetitionSince[X]" with 1900-01
# store_df["CompetitionOpenSinceYear"][(store_df["CompetitionDistance"] != 0) & (is_nan(store_df["CompetitionOpenSinceYear"]))] = 1900
# store_df["CompetitionOpenSinceMonth"][(store_df["CompetitionDistance"] != 0) & (is_nan(store_df["CompetitionOpenSinceMonth"]))] = 1
# One-hot encoding of "StoreType" & "Assortment" columns
# store_df = pd.get_dummies(store_df, columns=["StoreType", "Assortment"])
################################################################
# Process Data (Custom) #
################################################################
# Drop "Date" column
training_df.drop(["Date"], axis=1, inplace=True)
test_df.drop(["Date"], axis=1, inplace=True)
# Drop "Year", "Month" & "YearMonth" as test_df only has "Year" = 2015 & "Month" = 8 || 9
# training_df.drop(["Year", "Month", "YearMonth"], axis=1, inplace=True)
# test_df.drop(["Year", "Month", "YearMonth"], axis=1, inplace=True)
# Drop "StateHolidayBinary" as "StateHoliday_[X]" Exists
# training_df.drop(["StateHolidayBinary"], axis=1, inplace=True)
# test_df.drop(["StateHolidayBinary"], axis=1, inplace=True)
# Remove all Closed Stores ("Sales" = 0)
training_df = training_df[training_df["Open"] != 0]
# Drop Unnecessary Columns from training_df (not useful for prediction)
training_df.drop(["Open", "Customers", "StateHoliday", "SchoolHoliday"], axis=1, inplace=True)
# Save IDs of Closed Stores ("Sales" = 0 assigned later) & Remove Rows
closed_store_ids = test_df["Id"][test_df["Open"] == 0].values
test_df = test_df[test_df["Open"] != 0]
# Drop Unnecessary Columns from test_df (not useful for prediction)
test_df.drop(["Open", "StateHoliday", "SchoolHoliday"], axis=1, inplace=True)
################################################################
# RMSPE Function #
################################################################
def rmspe(y_true, y_pred):
w = np.zeros(y_pred.shape, dtype=float)
index = y_pred != 0
w[index] = 1.0 / (y_pred[index]**2)
diff = y_pred - y_true
diff_squared = diff ** 2
diff_percentage_squared = diff_squared * w
rmspe = np.sqrt(np.mean(diff_percentage_squared))
return rmspe
################################################################
# Training the Model & Predicting Sales #
################################################################
"""
Treating each store as an independent regression problem, loop through all stores training the model for the particular store and predicting its sales value.
Features: Promo, DayOfWeek (one-hot encoded)
Assumptions:
- The Year-Month has no effect on the sales as testing data is only for 2015-08 & 2015-09.
- The store's opening/closing dates does not affect the store's performance. For example, a store that was closed yesterday will not get more sales today because of that.
- The competition of each store will affect it consistently, hence, it does not matter.
- Each store's sales value is independent of the other stores and can be treated as independent regression problems.
- State and school holidays don't matter.
"""
print("Making predictions...")
training_dict = dict(list(training_df.groupby("Store")))
test_dict = dict(list(test_df.groupby("Store")))
predictions = Series()
for i in test_dict:
store = training_dict[i]
X_train = store.drop(["Sales", "Store"], axis=1)
Y_train = store["Sales"]
X_test = test_dict[i].copy()
# X_tr, X_te, Y_tr, Y_te = train_test_split(X_train, Y_train, test_size=0.4)
# lreg = LinearRegression()
# lreg.fit(X_tr, Y_tr)
# Y_pr = lreg.predict(X_te)
# print(rmspe(y_true=Y_te, y_pred=Y_pr))
store_ids = X_test["Id"]
X_test.drop(["Id", "Store"], axis=1, inplace=True)
X_test = X_test.reindex(columns=X_train.columns, fill_value=0)
lreg = LinearRegression()
lreg.fit(X_train, Y_train)
Y_pred = lreg.predict(X_test)
predictions = predictions.append(Series(Y_pred, index=store_ids))
predictions = predictions.append(Series(0, index=closed_store_ids))
submission = pd.DataFrame({"Id": predictions.index, "Sales": predictions.values})
submission.to_csv("predictions/linearregression2.csv", index=False)
print("Predictions saved.")