Breast Cancer Detection¶
Import the required libraries¶
In [1]:
Copied!
import tempfile
import tensorflow as tf
import pandas as pd
from sklearn.preprocessing import StandardScaler
import fastestimator as fe
from fastestimator.dataset.data import breast_cancer
from fastestimator.op.tensorop.loss import CrossEntropy
from fastestimator.op.tensorop.model import ModelOp, UpdateOp
from fastestimator.trace.io import BestModelSaver
from fastestimator.trace.metric import Accuracy
import tempfile
import tensorflow as tf
import pandas as pd
from sklearn.preprocessing import StandardScaler
import fastestimator as fe
from fastestimator.dataset.data import breast_cancer
from fastestimator.op.tensorop.loss import CrossEntropy
from fastestimator.op.tensorop.model import ModelOp, UpdateOp
from fastestimator.trace.io import BestModelSaver
from fastestimator.trace.metric import Accuracy
In [2]:
Copied!
#training parameters
batch_size = 4
epochs = 10
save_dir = tempfile.mkdtemp()
max_train_steps_per_epoch = None
max_eval_steps_per_epoch = None
#training parameters
batch_size = 4
epochs = 10
save_dir = tempfile.mkdtemp()
max_train_steps_per_epoch = None
max_eval_steps_per_epoch = None
Download data¶
This downloads some tabular data with different features stored in numerical format in a table. We then split the data into train, evaluation, and testing data sets.
In [3]:
Copied!
train_data, eval_data = breast_cancer.load_data()
test_data = eval_data.split(0.5)
train_data, eval_data = breast_cancer.load_data()
test_data = eval_data.split(0.5)
This is what the raw data looks like:
In [4]:
Copied!
df = pd.DataFrame.from_dict(train_data.data, orient='index')
df.head()
df = pd.DataFrame.from_dict(train_data.data, orient='index')
df.head()
Out[4]:
| x | y | |
|---|---|---|
| 0 | [9.029, 17.33, 58.79, 250.5, 0.1066, 0.1413, 0... | 1 |
| 1 | [21.09, 26.57, 142.7, 1311.0, 0.1141, 0.2832, ... | 0 |
| 2 | [9.173, 13.86, 59.2, 260.9, 0.07721, 0.08751, ... | 1 |
| 3 | [10.65, 25.22, 68.01, 347.0, 0.09657, 0.07234,... | 1 |
| 4 | [10.17, 14.88, 64.55, 311.9, 0.1134, 0.08061, ... | 1 |
In [5]:
Copied!
scaler = StandardScaler()
train_data["x"] = scaler.fit_transform(train_data["x"])
eval_data["x"] = scaler.transform(eval_data["x"])
test_data["x"] = scaler.transform(test_data["x"])
scaler = StandardScaler()
train_data["x"] = scaler.fit_transform(train_data["x"])
eval_data["x"] = scaler.transform(eval_data["x"])
test_data["x"] = scaler.transform(test_data["x"])
Building Components¶
Step 1: Create Pipeline¶
We create the Pipeline with the usual train, eval, and test data along with the batch size:
In [6]:
Copied!
pipeline = fe.Pipeline(train_data=train_data, eval_data=eval_data, test_data=test_data, batch_size=batch_size)
pipeline = fe.Pipeline(train_data=train_data, eval_data=eval_data, test_data=test_data, batch_size=batch_size)
Step 2: Create Network¶
We first define the neural network in a function that can then be passed on to the FastEstimator Network:
In [7]:
Copied!
def create_dnn():
model = tf.keras.Sequential()
model.add(tf.keras.layers.Dense(32, activation="relu", input_shape=(30, )))
model.add(tf.keras.layers.Dropout(0.5))
model.add(tf.keras.layers.Dense(16, activation="relu"))
model.add(tf.keras.layers.Dropout(0.5))
model.add(tf.keras.layers.Dense(8, activation="relu"))
model.add(tf.keras.layers.Dropout(0.5))
model.add(tf.keras.layers.Dense(1, activation="sigmoid"))
return model
def create_dnn():
model = tf.keras.Sequential()
model.add(tf.keras.layers.Dense(32, activation="relu", input_shape=(30, )))
model.add(tf.keras.layers.Dropout(0.5))
model.add(tf.keras.layers.Dense(16, activation="relu"))
model.add(tf.keras.layers.Dropout(0.5))
model.add(tf.keras.layers.Dense(8, activation="relu"))
model.add(tf.keras.layers.Dropout(0.5))
model.add(tf.keras.layers.Dense(1, activation="sigmoid"))
return model
In [8]:
Copied!
model = fe.build(model_fn=create_dnn, optimizer_fn="adam")
network = fe.Network(ops=[
ModelOp(inputs="x", model=model, outputs="y_pred"),
CrossEntropy(inputs=("y_pred", "y"), outputs="ce"),
UpdateOp(model=model, loss_name="ce", mode="!infer")
])
model = fe.build(model_fn=create_dnn, optimizer_fn="adam")
network = fe.Network(ops=[
ModelOp(inputs="x", model=model, outputs="y_pred"),
CrossEntropy(inputs=("y_pred", "y"), outputs="ce"),
UpdateOp(model=model, loss_name="ce", mode="!infer")
])
Step 3: Create Estimator¶
In [9]:
Copied!
traces = [
Accuracy(true_key="y", pred_key="y_pred"),
BestModelSaver(model=model, save_dir=save_dir, metric="accuracy", save_best_mode="max")
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=epochs,
log_steps=10,
traces=traces,
max_train_steps_per_epoch=max_train_steps_per_epoch,
max_eval_steps_per_epoch=max_eval_steps_per_epoch)
traces = [
Accuracy(true_key="y", pred_key="y_pred"),
BestModelSaver(model=model, save_dir=save_dir, metric="accuracy", save_best_mode="max")
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=epochs,
log_steps=10,
traces=traces,
max_train_steps_per_epoch=max_train_steps_per_epoch,
max_eval_steps_per_epoch=max_eval_steps_per_epoch)
Training¶
In [10]:
Copied!
estimator.fit()
estimator.fit()
______ __ ______ __ _ __
/ ____/___ ______/ /_/ ____/____/ /_(_)___ ___ ____ _/ /_____ _____
/ /_ / __ `/ ___/ __/ __/ / ___/ __/ / __ `__ \/ __ `/ __/ __ \/ ___/
/ __/ / /_/ (__ ) /_/ /___(__ ) /_/ / / / / / / /_/ / /_/ /_/ / /
/_/ \__,_/____/\__/_____/____/\__/_/_/ /_/ /_/\__,_/\__/\____/_/
FastEstimator-Start: step: 1; model_lr: 0.001;
FastEstimator-Train: step: 1; ce: 0.58930933;
FastEstimator-Train: step: 10; ce: 1.2191963; steps/sec: 342.02;
FastEstimator-Train: step: 20; ce: 0.6330318; steps/sec: 422.95;
FastEstimator-Train: step: 30; ce: 0.68403095; steps/sec: 400.86;
FastEstimator-Train: step: 40; ce: 0.70622563; steps/sec: 277.93;
FastEstimator-Train: step: 50; ce: 0.7649698; steps/sec: 443.68;
FastEstimator-Train: step: 60; ce: 0.70189; steps/sec: 455.18;
FastEstimator-Train: step: 70; ce: 0.6120157; steps/sec: 486.19;
FastEstimator-Train: step: 80; ce: 0.6461396; steps/sec: 495.01;
FastEstimator-Train: step: 90; ce: 0.709924; steps/sec: 397.38;
FastEstimator-Train: step: 100; ce: 0.69695604; steps/sec: 545.18;
FastEstimator-Train: step: 110; ce: 0.5406225; steps/sec: 587.24;
FastEstimator-Train: step: 114; epoch: 1; epoch_time: 3.3 sec;
FastEstimator-ModelSaver: saved model to ./model_best_accuracy.h5
FastEstimator-Eval: step: 114; epoch: 1; ce: 0.49621966; min_ce: 0.49621966; since_best: 0; accuracy: 0.9824561403508771;
FastEstimator-Train: step: 120; ce: 0.55340487; steps/sec: 7.94;
FastEstimator-Train: step: 130; ce: 0.31839967; steps/sec: 308.44;
FastEstimator-Train: step: 140; ce: 0.16889682; steps/sec: 482.01;
FastEstimator-Train: step: 150; ce: 0.5158031; steps/sec: 407.94;
FastEstimator-Train: step: 160; ce: 0.6378304; steps/sec: 386.75;
FastEstimator-Train: step: 170; ce: 1.1647241; steps/sec: 447.16;
FastEstimator-Train: step: 180; ce: 0.5274984; steps/sec: 500.64;
FastEstimator-Train: step: 190; ce: 0.68258667; steps/sec: 481.22;
FastEstimator-Train: step: 200; ce: 0.35559005; steps/sec: 476.14;
FastEstimator-Train: step: 210; ce: 0.46034247; steps/sec: 508.85;
FastEstimator-Train: step: 220; ce: 0.95580393; steps/sec: 548.84;
FastEstimator-Train: step: 228; epoch: 2; epoch_time: 1.39 sec;
FastEstimator-Eval: step: 228; epoch: 2; ce: 0.3193086; min_ce: 0.3193086; since_best: 0; accuracy: 0.9824561403508771;
FastEstimator-Train: step: 230; ce: 0.33260575; steps/sec: 8.49;
FastEstimator-Train: step: 240; ce: 0.2510308; steps/sec: 232.08;
FastEstimator-Train: step: 250; ce: 0.2878321; steps/sec: 666.82;
FastEstimator-Train: step: 260; ce: 0.1154226; steps/sec: 368.11;
FastEstimator-Train: step: 270; ce: 0.26300237; steps/sec: 414.99;
FastEstimator-Train: step: 280; ce: 0.5653368; steps/sec: 421.39;
FastEstimator-Train: step: 290; ce: 0.5872185; steps/sec: 402.68;
FastEstimator-Train: step: 300; ce: 0.27621573; steps/sec: 440.24;
FastEstimator-Train: step: 310; ce: 0.5477217; steps/sec: 481.69;
FastEstimator-Train: step: 320; ce: 0.4602429; steps/sec: 398.85;
FastEstimator-Train: step: 330; ce: 0.38244748; steps/sec: 546.57;
FastEstimator-Train: step: 340; ce: 0.5337428; steps/sec: 571.02;
FastEstimator-Train: step: 342; epoch: 3; epoch_time: 1.42 sec;
FastEstimator-Eval: step: 342; epoch: 3; ce: 0.18308732; min_ce: 0.18308732; since_best: 0; accuracy: 0.9824561403508771;
FastEstimator-Train: step: 350; ce: 0.13466343; steps/sec: 8.53;
FastEstimator-Train: step: 360; ce: 0.22628057; steps/sec: 368.34;
FastEstimator-Train: step: 370; ce: 0.5836228; steps/sec: 485.06;
FastEstimator-Train: step: 380; ce: 0.37300625; steps/sec: 409.87;
FastEstimator-Train: step: 390; ce: 0.2717349; steps/sec: 413.59;
FastEstimator-Train: step: 400; ce: 0.07554119; steps/sec: 433.84;
FastEstimator-Train: step: 410; ce: 0.20552614; steps/sec: 439.36;
FastEstimator-Train: step: 420; ce: 0.28509304; steps/sec: 448.96;
FastEstimator-Train: step: 430; ce: 0.32158756; steps/sec: 492.58;
FastEstimator-Train: step: 440; ce: 1.1102628; steps/sec: 525.49;
FastEstimator-Train: step: 450; ce: 0.31964102; steps/sec: 548.06;
FastEstimator-Train: step: 456; epoch: 4; epoch_time: 1.4 sec;
FastEstimator-ModelSaver: saved model to ./model_best_accuracy.h5
FastEstimator-Eval: step: 456; epoch: 4; ce: 0.105911165; min_ce: 0.105911165; since_best: 0; accuracy: 1.0;
FastEstimator-Train: step: 460; ce: 0.4391592; steps/sec: 8.37;
FastEstimator-Train: step: 470; ce: 0.29870045; steps/sec: 297.42;
FastEstimator-Train: step: 480; ce: 0.03247342; steps/sec: 597.74;
FastEstimator-Train: step: 490; ce: 0.13323224; steps/sec: 393.92;
FastEstimator-Train: step: 500; ce: 0.58429027; steps/sec: 405.0;
FastEstimator-Train: step: 510; ce: 0.2376658; steps/sec: 455.97;
FastEstimator-Train: step: 520; ce: 0.4150503; steps/sec: 424.88;
FastEstimator-Train: step: 530; ce: 0.22695109; steps/sec: 451.62;
FastEstimator-Train: step: 540; ce: 0.42051294; steps/sec: 402.12;
FastEstimator-Train: step: 550; ce: 0.17364319; steps/sec: 389.83;
FastEstimator-Train: step: 560; ce: 0.06320181; steps/sec: 466.97;
FastEstimator-Train: step: 570; ce: 0.13996354; steps/sec: 518.8;
FastEstimator-Train: step: 570; epoch: 5; epoch_time: 1.46 sec;
FastEstimator-Eval: step: 570; epoch: 5; ce: 0.066059396; min_ce: 0.066059396; since_best: 0; accuracy: 1.0;
FastEstimator-Train: step: 580; ce: 0.12985338; steps/sec: 8.17;
FastEstimator-Train: step: 590; ce: 0.6419388; steps/sec: 373.15;
FastEstimator-Train: step: 600; ce: 0.2857446; steps/sec: 404.92;
FastEstimator-Train: step: 610; ce: 0.21400735; steps/sec: 381.65;
FastEstimator-Train: step: 620; ce: 0.27899668; steps/sec: 394.87;
FastEstimator-Train: step: 630; ce: 0.31599885; steps/sec: 472.31;
FastEstimator-Train: step: 640; ce: 0.036415085; steps/sec: 457.09;
FastEstimator-Train: step: 650; ce: 0.10052729; steps/sec: 461.82;
FastEstimator-Train: step: 660; ce: 0.40688303; steps/sec: 474.46;
FastEstimator-Train: step: 670; ce: 0.40816957; steps/sec: 517.75;
FastEstimator-Train: step: 680; ce: 0.40120217; steps/sec: 555.53;
FastEstimator-Train: step: 684; epoch: 6; epoch_time: 1.44 sec;
FastEstimator-Eval: step: 684; epoch: 6; ce: 0.04396173; min_ce: 0.04396173; since_best: 0; accuracy: 1.0;
FastEstimator-Train: step: 690; ce: 0.20741543; steps/sec: 8.33;
FastEstimator-Train: step: 700; ce: 0.12485474; steps/sec: 324.64;
FastEstimator-Train: step: 710; ce: 2.8970864e-05; steps/sec: 534.71;
FastEstimator-Train: step: 720; ce: 0.110491954; steps/sec: 402.98;
FastEstimator-Train: step: 730; ce: 0.10486858; steps/sec: 432.34;
FastEstimator-Train: step: 740; ce: 0.2951797; steps/sec: 421.45;
FastEstimator-Train: step: 750; ce: 0.65293443; steps/sec: 433.71;
FastEstimator-Train: step: 760; ce: 0.32570755; steps/sec: 461.43;
FastEstimator-Train: step: 770; ce: 0.35400242; steps/sec: 433.46;
FastEstimator-Train: step: 780; ce: 0.023054674; steps/sec: 483.0;
FastEstimator-Train: step: 790; ce: 0.16433364; steps/sec: 540.17;
FastEstimator-Train: step: 798; epoch: 7; epoch_time: 1.43 sec;
FastEstimator-Eval: step: 798; epoch: 7; ce: 0.040205613; min_ce: 0.040205613; since_best: 0; accuracy: 1.0;
FastEstimator-Train: step: 800; ce: 0.42427045; steps/sec: 8.52;
FastEstimator-Train: step: 810; ce: 0.39827985; steps/sec: 266.34;
FastEstimator-Train: step: 820; ce: 0.43165076; steps/sec: 775.07;
FastEstimator-Train: step: 830; ce: 0.06976031; steps/sec: 412.9;
FastEstimator-Train: step: 840; ce: 0.37039524; steps/sec: 441.05;
FastEstimator-Train: step: 850; ce: 0.10960688; steps/sec: 418.01;
FastEstimator-Train: step: 860; ce: 0.0070317476; steps/sec: 450.1;
FastEstimator-Train: step: 870; ce: 0.020452987; steps/sec: 434.06;
FastEstimator-Train: step: 880; ce: 0.12914097; steps/sec: 476.49;
FastEstimator-Train: step: 890; ce: 0.25528443; steps/sec: 466.42;
FastEstimator-Train: step: 900; ce: 0.18017673; steps/sec: 549.95;
FastEstimator-Train: step: 910; ce: 0.31777602; steps/sec: 583.67;
FastEstimator-Train: step: 912; epoch: 8; epoch_time: 1.41 sec;
FastEstimator-Eval: step: 912; epoch: 8; ce: 0.028554583; min_ce: 0.028554583; since_best: 0; accuracy: 1.0;
FastEstimator-Train: step: 920; ce: 0.24684253; steps/sec: 8.42;
FastEstimator-Train: step: 930; ce: 0.19438684; steps/sec: 365.05;
FastEstimator-Train: step: 940; ce: 0.1568121; steps/sec: 477.85;
FastEstimator-Train: step: 950; ce: 0.3368427; steps/sec: 371.39;
FastEstimator-Train: step: 960; ce: 0.20518681; steps/sec: 411.72;
FastEstimator-Train: step: 970; ce: 0.13320616; steps/sec: 401.91;
FastEstimator-Train: step: 980; ce: 0.1800138; steps/sec: 470.79;
FastEstimator-Train: step: 990; ce: 0.10868286; steps/sec: 421.41;
FastEstimator-Train: step: 1000; ce: 0.040300086; steps/sec: 467.66;
FastEstimator-Train: step: 1010; ce: 0.42622733; steps/sec: 505.31;
FastEstimator-Train: step: 1020; ce: 0.06701453; steps/sec: 530.27;
FastEstimator-Train: step: 1026; epoch: 9; epoch_time: 1.42 sec;
FastEstimator-Eval: step: 1026; epoch: 9; ce: 0.019402837; min_ce: 0.019402837; since_best: 0; accuracy: 1.0;
FastEstimator-Train: step: 1030; ce: 0.27714887; steps/sec: 8.63;
FastEstimator-Train: step: 1040; ce: 0.074241355; steps/sec: 302.11;
FastEstimator-Train: step: 1050; ce: 0.025415465; steps/sec: 640.07;
FastEstimator-Train: step: 1060; ce: 0.21693969; steps/sec: 447.73;
FastEstimator-Train: step: 1070; ce: 0.120441705; steps/sec: 432.52;
FastEstimator-Train: step: 1080; ce: 0.25360084; steps/sec: 459.08;
FastEstimator-Train: step: 1090; ce: 0.22401881; steps/sec: 493.29;
FastEstimator-Train: step: 1100; ce: 0.112028226; steps/sec: 485.33;
FastEstimator-Train: step: 1110; ce: 2.4293017; steps/sec: 446.65;
FastEstimator-Train: step: 1120; ce: 0.19810674; steps/sec: 514.16;
FastEstimator-Train: step: 1130; ce: 0.12599353; steps/sec: 529.87;
FastEstimator-Train: step: 1140; ce: 0.23468983; steps/sec: 580.97;
FastEstimator-Train: step: 1140; epoch: 10; epoch_time: 1.38 sec;
FastEstimator-Eval: step: 1140; epoch: 10; ce: 0.017586827; min_ce: 0.017586827; since_best: 0; accuracy: 1.0;
FastEstimator-Finish: step: 1140; total_time: 28.21 sec; model_lr: 0.001;
Model testing¶
Estimator.test triggers model testing with the test dataset that was specified in our Pipeline. We can use this to evaluate our model's accuracy on previously unseen data:
In [11]:
Copied!
estimator.test()
estimator.test()
FastEstimator-Test: epoch: 10; accuracy: 0.9649122807017544;