Go to Qiita Advent Calendar Top
LoginSignup
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 1 year has passed since last update.

@sarara

ニューラルネットワークの基本コード

Posted at 
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
import datetime, os, random
import seaborn as sns

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.utils import Sequence
%load_ext tensorboard

from google.colab import auth
from oauth2client.client import GoogleCredentials
import gspread

データセット読み込み

from google.colab import auth
auth.authenticate_user()

import gspread
from google.auth import default
creds, _ = default()

gc = gspread.authorize(creds)

filename = "命令コード20から1"
ss = gc.open(filename)

np.random.seed(0)
# tf.keras.set_random_seed(0)

worksheet = ss.worksheet("norm")
# すべての値を多次元配列datasetへ
dataset = pd.DataFrame(worksheet.get_all_values())
# 0行目を列名にする
# dataset.columns = list(dataset.loc[0, :])
dataset.columns = list(dataset.loc[0, :])
dataset.drop(0, inplace=True)
dataset.reset_index(inplace=True)
dataset.drop('name', axis=1, inplace=True)
dataset.l_0 = dataset.l_0.astype(float)
dataset.l_1 = dataset.l_1.astype(float)
dataset.l_2 = dataset.l_2.astype(float)
dataset.l_3 = dataset.l_3.astype(float)
dataset.l_4 = dataset.l_4.astype(float)
dataset.l_5 = dataset.l_5.astype(float)
dataset.l_6 = dataset.l_6.astype(float)
dataset.l_7 = dataset.l_7.astype(float)
dataset.l_8 = dataset.l_8.astype(float)
dataset.l_9 = dataset.l_9.astype(float)
dataset.l_10 = dataset.l_10.astype(float)
dataset.l_11 = dataset.l_11.astype(float)
dataset.l_12 = dataset.l_12.astype(float)
dataset.l_13 = dataset.l_13.astype(float)
dataset.l_14 = dataset.l_14.astype(float)
dataset.l_15 = dataset.l_15.astype(float)
dataset.l_16 = dataset.l_16.astype(float)
dataset.l_17 = dataset.l_17.astype(float)
dataset.l_18 = dataset.l_18.astype(float)
dataset.l_19 = dataset.l_19.astype(float)


dataset.cycle = dataset.cycle.astype(float)

dataset.sample(5)

train_dataset = dataset.sample(frac=0.8,random_state=0)
test_dataset = dataset.drop(train_dataset.index)

# 訓練データを8割指定
# train_dataset = dataset.head(173)
# index列を消す
# test_dataset = dataset.tail(17)
sns.pairplot(train_dataset[["l_0"]], diag_kind="kde")

train_dataset = dataset.sample(frac=0.8,random_state=0)
test_dataset = dataset.drop(train_dataset.index)

# 訓練データを8割指定
# train_dataset = dataset.head(173)
# index列を消す
# test_dataset = dataset.tail(17)
sns.pairplot(train_dataset[["l_0"]], diag_kind="kde")

train_labels = pd.DataFrame.from_dict({
    'cycle': train_dataset.pop('cycle'),
})
train_dataset.pop("index")

test_labels = pd.DataFrame.from_dict({
    'cycle': test_dataset.pop('cycle'),

})
test_dataset.pop("index")

test_labels.sample(4)

学習用データを正規化

train_stats = train_dataset.describe()
train_stats = train_stats.transpose()
train_stats

def norm(x):
    return x
    # return (x - train_stats['mean']) / train_stats['std']

normed_train_data = norm(train_dataset)
normed_test_data = norm(test_dataset)
normed_train_data.sample(5)

モデル

def build_model():
    lambd = 0.1
    inputs = layers.Input(shape=len(train_dataset.keys()),)
    # x = layers.Dense(8, activation='relu', kernel_regularizer=keras.regularizers.L1(float(lambd)))(inputs)
    x = layers.Dense(20, activation='relu')(inputs)
    x = layers.Dense(20, activation='relu')(x)
    for i in range(1,10):
        x = layers.Dense(20, activation='relu')(x)
        #x = layers.Dropout(0.2)(x)
    a_load = layers.Dense(1, name='cycle')(x)
    model = keras.Model(inputs=inputs, outputs=a_load, name='LLVM')

    model.compile(
        optimizer=tf.keras.optimizers.RMSprop(0.001),
        loss_weights=[0.8],
        loss = 'mse',
        metrics = ['mae'])

    return model

model = build_model()
model.summary()

入出力の正しさ検証

example_batch = normed_train_data[:100]
example_result = model.predict(example_batch)
example_result

訓練

# エポックが終わるごとにドットを一つ出力することで進捗を表示
class PrintDot(keras.callbacks.Callback):
  def on_epoch_end(self, epoch, logs):
    if epoch % 100 == 0: print('')
    print('.', end='')

logdir = os.path.join("logs", datetime.datetime.now().strftime("%Y%m%d-%H%M%S"))
tensorboard_callback = tf.keras.callbacks.TensorBoard(logdir, histogram_freq=1)

EPOCHS =300

model.fit(
  normed_train_data,
  train_labels.cycle,
  epochs=EPOCHS,
  validation_split=0.1,
  verbose=0,
  callbacks=[tensorboard_callback, PrintDot()])

%tensorboard --logdir logs

テストデータで予測

test_predictions = model.predict(normed_test_data)
result = normed_test_data.copy()
result['cycle'] = dataset['cycle']

result['predict_cycle'] = test_predictions
result['diff_cycle'] = result['predict_cycle'] - result['cycle']
result['error_cycle'] = round((result['predict_cycle'] - result['cycle']) / result['cycle'] * 100)

result.reindex(columns=['cycle', 'predict_cycle', 'diff_cycle', 'error_cycle']).sort_values('error_cycle')

worksheet = ss.worksheet("eval")
# すべての値を多次元配列datasetへ
evaldata = pd.DataFrame(worksheet.get_all_values())
# 0行目を列名にする
# evaldata.columns = list(evaldata.loc[0, :])
evaldata.columns = list(evaldata.loc[0, :])
evaldata.drop(0, inplace=True)
evaldata.reset_index(inplace=True)
evaldata.drop('name', axis=1, inplace=True)

evaldata.l_0 = evaldata.l_0.astype(float)
evaldata.l_1 = evaldata.l_1.astype(float)
evaldata.l_2 = evaldata.l_2.astype(float)
evaldata.l_3 = evaldata.l_3.astype(float)
evaldata.l_4 = evaldata.l_4.astype(float)
evaldata.l_5 = evaldata.l_5.astype(float)
evaldata.l_6 = evaldata.l_6.astype(float)
evaldata.l_7 = evaldata.l_7.astype(float)
evaldata.l_8 = evaldata.l_8.astype(float)
evaldata.l_9 = evaldata.l_9.astype(float)
evaldata.l_10 = evaldata.l_10.astype(float)
evaldata.l_11 = evaldata.l_11.astype(float)
evaldata.l_12 = evaldata.l_12.astype(float)
evaldata.l_13 = evaldata.l_13.astype(float)
evaldata.l_14 = evaldata.l_14.astype(float)
evaldata.l_15 = evaldata.l_15.astype(float)
evaldata.l_16 = evaldata.l_16.astype(float)
evaldata.l_17 = evaldata.l_17.astype(float)
evaldata.l_18 = evaldata.l_18.astype(float)
evaldata.l_19 = evaldata.l_19.astype(float)


eval_labels = pd.DataFrame.from_dict({
    'cycle': evaldata.pop('cycle')})
evaldata.pop("index")

eval_dataset = norm(evaldata)

eval_dataset

test_predictions = model.predict(eval_dataset)

result = eval_dataset.copy()
result['cycle'] = eval_labels['cycle'].astype(float)
result['predict_cycle'] = test_predictions

result['cycle'].dtype #predictはfloat #cycle(正解)はobject

result['diff_cycle'] = result['predict_cycle'] - result['cycle']
result['error_cycle'] = round((result['predict_cycle'] - result['cycle']) / result['cycle'] * 100)

result.reindex(columns=['cycle', 'predict_cycle', 'diff_cycle', 'error_cycle']).sort_values('error_cycle')

for i in range(1):
  filename = str(i) + "層" + ".csv"

    # 結果をCSVファイルに保存
  result.to_csv(filename, index=False)
0
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
Sign upLogin
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?