Stacking models

Advanced Deep Learning with Keras

Zach Deane Mayer

Data Scientist

Stacking models requires 2 datasets

from pandas import read_csv
games_season = read_csv('datasets/games_season.csv')
games_season.head()

   team_1  team_2  home  score_diff
0    3745    6664     0          17
1     126    7493     1           7
2     288    3593     1           7
3    1846    9881     1          16
4    2675   10298     1          12

games_tourney = read_csv('datasets/games_tourney.csv')
games_tourney.head()

   team_1  team_2  home  seed_diff  score_diff
0     288      73     0         -3          -9
1    5929      73     0          4           6
2    9884      73     0          5          -4
3      73     288     0          3           9
4    3920     410     0          1          -9

Enrich the tournament data

in_data_1 = games_tourney['team_1']
in_data_2 = games_tourney['team_2']
in_data_3 = games_tourney['home']
pred = regular_season_model.predict([in_data_1, in_data_2, in_data_3])

games_tourney['pred'] = pred
games_tourney.head()

   team_1  team_2  home  seed_diff      pred  score_diff
0     288      73     0         -3  0.582556          -9
1    5929      73     0          4  0.707279           6
2    9884      73     0          5  1.364844          -4
3      73     288     0          3  0.699145           9
4    3920     410     0          1  0.833066          -9

3 input model with pure numeric data

games_tourney[['home','seed_diff','pred']].head()

   home  seed_diff      pred
0     0         -3  0.582556
1     0          4  0.707279
2     0          5  1.364844
3     0          3  0.699145
4     0          1  0.833066

3 input model with pure numeric data

from tensorflow.keras.layers import Input, Dense
in_tensor = Input(shape=(3,))
out_tensor = Dense(1)(in_tensor)

from tensorflow.keras.models import Model
model = Model(in_tensor, out_tensor)
model.compile(optimizer='adam', loss='mae')
train_X = train_data[['home','seed_diff','pred']]
train_y = train_data['score_diff']
model.fit(train_X,train_y, epochs=10, validation_split=.10)

test_X = test_data[['home','seed_diff','pred']]
test_y = test_data['score_diff']
model.evaluate(test_X, test_y)
1066/1066 [==============================] - 0s 14us/step
9.11321775461451

Let's practice!

Advanced Deep Learning with Keras