# import standard scientific libraries import os import math import numpy as np import pandas as pd # import ML models from scikit-learn from sklearn.linear_model import LinearRegression from sklearn.kernel_ridge import KernelRidge from sklearn.gaussian_process import GaussianProcessRegressor from sklearn import svm from sklearn.preprocessing import PolynomialFeatures from sklearn.metrics import mean_absolute_error from sklearn.preprocessing import StandardScaler from sklearn.model_selection import train_test_split from tensorflow.keras.layers.experimental import preprocessing import matplotlib.pyplot as plt

def plot_loss(history): plt.plot(history.history['loss'], label='loss') plt.legend() plt.grid(True)

RANDOM_SEED = 4 np.random.seed(RANDOM_SEED) pd.set_option('max_columns', None) pd.set_option("display.precision", 8) dataset = "../dataset/"

train = pd.read_csv("train.csv")#[:66000] train.shape train

train.info()

train = train.replace([np.inf, -np.inf], np.nan) #train = train[train['heat_adsorption_CO2_P0.15bar_T298K [kcal/mol]'].notna()] train = train[train['functional_groups'].notna()]

train.info()

# train = pd.get_dummies(train, columns=["functional_groups"]) # train = pd.get_dummies(train, columns=["topology"]) # train col = ["functional_groups", "topology"] for i in col: train[i] = train[i].astype("category").cat.codes

train = train.drop(['MOFname'],axis=1) train= train[train['void_fraction'] > 0] train= train[train['void_volume [cm^3/g]'] > 0] #train train[train['CO2/N2_selectivity'] > 0] train = train[train['heat_adsorption_CO2_P0.15bar_T298K [kcal/mol]'].notna()] train= train[train['surface_area [m^2/g]'] > 0]

train

train.info()

# find rows having NaN train.isnull().any(axis=0) #train.fillna(method='pad', inplace=True) #train.groupby('functional_groups')['functional_groups'].count()

count = 0 for i in train['functional_groups']: if i == 240: count = count+1 print(count)

# find row having inf np.isinf(train).any(axis=0)

train

x = train.drop(['CO2_working_capacity [mL/g]'],axis=1) y = train['CO2_working_capacity [mL/g]']

x_train, x_test, y_train, y_true = train_test_split(x, y, test_size=0.2,random_state=RANDOM_SEED)

scaler = StandardScaler() import tensorflow as tf x_train = pd.DataFrame(scaler.fit_transform(x_train), columns = x_train.columns) x_test = pd.DataFrame(scaler.fit_transform(x_test), columns = x_test.columns) normalizer = preprocessing.Normalization(axis=-1) normalizer.adapt(np.array(x_train)) print(normalizer.mean.numpy())

from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense

from tensorflow.keras import datasets, layers, models import tensorflow as tf from tensorflow.keras import regularizers from tensorflow.keras.layers import BatchNormalization initializer = tf.keras.initializers.VarianceScaling( scale=0.1, mode='fan_in', distribution='uniform',seed = RANDOM_SEED) tf.random.set_seed(RANDOM_SEED) model = Sequential(normalizer) model.add(Dense(416,kernel_initializer=initializer, kernel_regularizer=regularizers.l2(0.0001),input_dim=(x_train.shape[1]), activation='relu')) # input model.add(layers.Dropout(0.2)) model.add(Dense(416,kernel_initializer=initializer, kernel_regularizer=regularizers.l2(0.0001),activation='relu')) # hidden 1 model.add(layers.Dropout(0.2)) model.add(Dense(416,kernel_initializer=initializer, kernel_regularizer=regularizers.l2(0.0001),activation='relu')) # hidden 2 model.add(layers.Dropout(0.2)) model.add(Dense(416,kernel_initializer=initializer, kernel_regularizer=regularizers.l2(0.0001),activation='relu')) # hidden 3 model.add(layers.Dropout(0.2)) model.add(Dense(1))# output

model.summary()

import tensorflow as tf lr_schedule = tf.keras.optimizers.schedules.InverseTimeDecay( 0.001, decay_steps=x_train.shape[0]*1000, decay_rate=0.1, staircase=False) opt = tf.keras.optimizers.Adamax(lr_schedule) # stop_early = tf.keras.callbacks.EarlyStopping(monitor='loss', patience=5) model.compile(loss='mean_absolute_error',optimizer=opt) history = model.fit(x_train, y_train, epochs=98, batch_size=128)

plot_loss(history)

val_acc_per_epoch = history.history['loss'] best_epoch = val_acc_per_epoch.index(min(val_acc_per_epoch)) + 1 print('Best epoch: %d' % (best_epoch,))

y_pred = model.predict(x_train)

log_mae = np.log(mean_absolute_error(y_pred, y_train)) log_mae

y_pred = model.predict(x_test)

log_mae = np.log(mean_absolute_error(y_pred, y_true)) log_mae

pretest = pd.read_csv("test.csv") pretest.shape

pretest.info()

pretest['functional_groups'] = pretest['functional_groups'].replace({np.nan:0}) # train['void_fraction'] = train['void_fraction'].replace({'0':np.nan, 0:np.nan}) # train['void_volume [cm^3/g]'] = train['void_volume [cm^3/g]'].replace({'0':np. nan, 0:np.nan}) #train['functional_groups'] = train['functional_groups'].fillna(train.groupby('functional_groups')['functional_groups'].transform('mean')) # train['void_fraction'] = train['void_fraction'].fillna(train.groupby('functional_groups')['void_fraction'].transform('mean')) # train['void_volume [cm^3/g]'] = train['void_volume [cm^3/g]'].fillna(train.groupby('functional_groups')['void_volume [cm^3/g]'].transform('mean'))

pretest.info()

col = ["functional_groups", "topology"] for i in col: pretest[i] = pretest[i].astype("category").cat.codes

pretest

pretest = pretest.drop(['MOFname'],axis=1)

scaler = StandardScaler() pretest = pd.DataFrame(scaler.fit_transform(pretest), columns = pretest.columns) pretest = scaler.inverse_transform(pretest)

pretest

pretest_pred = model.predict(pretest) pretest_pred

submission = pd.DataFrame({ "id": [str(i) for i in range(68614,85614)], "CO2_working_capacity [mL/g]": pretest_pred.T[0] }) submission.to_csv("submission.csv", index=False)

!ls

%%capture !sudo apt-get update !sudo apt-get install zip

!zip submission_NN.zip submission.csv