This commit is contained in:
Anna Nowak 2021-05-16 23:41:55 +02:00
parent ba28ee1234
commit fb163e5653
5 changed files with 26 additions and 5 deletions

1
.gitignore vendored
View File

@ -65,3 +65,4 @@ dev.csv
.venv/ .venv/
model.h5 model.h5
evaluation.png evaluation.png
mlruns/*

11
MLProject Normal file
View File

@ -0,0 +1,11 @@
name: Fifa Players
docker_env:
image: docker.io/adnovac/ium_s434760:2.0
entry_points:
train:
parameters:
batch_size: {type: int, default: 15}
epochs: {type: int, default: 16}
command: "python train.py {batch_size} {epochs}"
evaluate:
command: "python evaluate.py"

View File

@ -2,8 +2,8 @@ import pandas as pd
import numpy as np import numpy as np
from os import path from os import path
from tensorflow import keras from tensorflow import keras
import sys
import matplotlib.pyplot as plt import matplotlib.pyplot as plt
import mlflow
model_name = "model.h5" model_name = "model.h5"
@ -16,6 +16,8 @@ Y_test=test_data[["Overall"]].to_numpy()
#MeanSquaredError #MeanSquaredError
results_test = model.evaluate(X_test, Y_test, batch_size=128) results_test = model.evaluate(X_test, Y_test, batch_size=128)
mlflow.log_metric("rmse", results_test)
with open('results.txt', 'a+', encoding="UTF-8") as f: with open('results.txt', 'a+', encoding="UTF-8") as f:
f.write(str(results_test) +"\n") f.write(str(results_test) +"\n")

View File

@ -5,3 +5,4 @@ sklearn
tensorflow==2.4.1 tensorflow==2.4.1
jinja2==2.11.3 jinja2==2.11.3
matplotlib matplotlib
mlflow==1.17.0

View File

@ -2,6 +2,7 @@ import pandas as pd
from os import path from os import path
from tensorflow import keras from tensorflow import keras
from tensorflow.keras import layers from tensorflow.keras import layers
import mlflow
import sys import sys
model_name = "model.h5" model_name = "model.h5"
@ -25,11 +26,16 @@ model.compile(
loss=keras.losses.MeanSquaredError(), loss=keras.losses.MeanSquaredError(),
) )
batch_size = int(sys.argv[1])
epochs = int(sys.argv[2])
mlflow.log_param("batch_size", batch_size)
mlflow.log_param("epochs", epochs)
history = model.fit( history = model.fit(
X, X,
Y, Y,
batch_size=int(sys.argv[1]), batch_size=batch_size,
epochs=int(sys.argv[2]), epochs=epochs,
) )
model.save(model_name) model.save(model_name)