CODEWITHSUNDEEP
pythonkeras
John Stud
John Stud

Reputation: 1779

Why is Keras training well but returning wrong predictions?

If I feed the model five Setosa flowers, I cannot get my model to predict that they are indeed Setosas.

Here is my code setup:

# Load libraries
import numpy as np
import pandas as pd
from keras import models
from keras import layers
from keras.models import Sequential
from keras.layers import Dense
from sklearn.utils import shuffle
from sklearn import preprocessing
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split, cross_val_score, GridSearchCV

# Set random seed
np.random.seed(0)

# Step 1: Load data
iris = pd.read_csv("iris.csv")

X = iris.drop('species', axis=1)
y = pd.get_dummies(iris['species']).values

# Step 2: Preprocess data
scaler = preprocessing.StandardScaler() 
X = scaler.fit_transform(X)

X, y = shuffle(X, y)

X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0)

network = models.Sequential()
network.add(layers.Dense(units=8, activation="relu", input_shape=(4,)))
network.add(layers.Dense(units=3, activation="softmax"))

# Compile neural network
network.compile(loss="categorical_crossentropy", 
                optimizer="adam", 
                metrics=["accuracy"]) 

# Train neural network
history = network.fit(X_train, # Features
                      y_train, # Target
                      epochs= 200, 
                      verbose= 1, 
                      batch_size=10, # Number of observations per batch
                      validation_data=(X_test, y_test)) # Test data

The model trained well, here is the last epoch:

Epoch 200/200
112/112 [==============================] - 0s 910us/step - loss: 0.0740 - acc: 0.9911 - val_loss: 0.1172 - val_acc: 0.9737

Now, let's pull some predictions.

new_iris = iris.iloc[0:5, 0:4] # pull out the first five Setosas from original iris dataset; 
# prediction should give me Setosa since I am feeding it Setosas

np.around(network.predict(new_iris), decimals = 2) # predicts versicolor with high probability

array([[0.  , 0.95, 0.04],
       [0.  , 0.94, 0.06],
       [0.  , 0.96, 0.04],
       [0.  , 0.91, 0.09],
       [0.  , 0.96, 0.04]], dtype=float32)\

Any ideas as to why this is the case?

Upvotes: 0

Views: 132

Answers (1)

xashru
xashru

Reputation: 3590

You need to apply the transformation learned during training at test time.

new_iris = iris.iloc[0:5, 0:4] # pull out the first five Setosas from original iris dataset; 
new_iris = scaler.transform(new_iris)
np.around(network.predict(new_iris), decimals = 2)

Outputs

array([[1.  , 0.  , 0.  ],
       [0.99, 0.01, 0.  ],
       [1.  , 0.  , 0.  ],
       [0.99, 0.01, 0.  ],
       [1.  , 0.  , 0.  ]], dtype=float32)

Upvotes: 1

Related Questions