-
Notifications
You must be signed in to change notification settings - Fork 0
/
trainZIR_230726.py
170 lines (145 loc) · 7.37 KB
/
trainZIR_230726.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
# UPDATES:
# 1. New tile drainage with more counties included
# 2. Using mean soil properties for the top 30cm
# 3. Random state = 4 because it gave best performing model in the previous ensemble
# imports
import pandas as pd
import numpy as np
import datetime
import pytz
import pickle
import os
from sklearn import metrics
from sklearn.svm import SVC
from sklearn.linear_model import LinearRegression
from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
from sklearn.model_selection import cross_val_score, RandomizedSearchCV, train_test_split
from sklego.meta import ZeroInflatedRegressor
from scipy.stats import uniform as sp_randFloat
from scipy.stats import randint as sp_randInt
random_state = 4
df = pd.read_csv('/project2/moyer/ag_data/prevented-planting/traindata-corn-excessmoist.csv')
df['fips'] = df.fips.astype(str).str.zfill(5)
df['region'] = df.region.astype(str).str.zfill(5)
# Set values above 1.0 to 1.0
df.loc[df["ppfrac"] > 1.0, "ppfrac"] = 1.0
state_exc_100lon = ['HI','AK','WA','OR','CA','ID','NV','AZ','MT','WY','UT','CO','NM']
df = df[~df.state.isin(state_exc_100lon)]
# Split data into labels & features -- and convert to numpy arrays
# CUSTOM VARIABLES
labels = np.array(df['ppfrac'])
months_incl = np.array([1,2,3,4,5,6])
months_excl = np.array([month for month in np.arange(1,13) if month not in months_incl])
weather_vars = ['evaptrans_','runsurf_','runsub_','rain_','tempair_','watersoil_','tempsoil_']
weather_vars = [var+str(month).zfill(2) for var in weather_vars for month in months_incl]
cst_vars = [
'frac_tile_drained', 'lat', 'lon', #'fips'
'awc_mean',#'awc_mean_0_5', 'awc_mean_5_15', 'awc_mean_15_30', 'awc_mean_30_60', 'awc_mean_60_100',
'om_mean',#'om_mean_0_5', 'om_mean_5_15', 'om_mean_15_30', 'om_mean_30_60', 'om_mean_60_100',
'clay_mean',#'clay_mean_0_5', 'clay_mean_5_15', 'clay_mean_15_30', 'clay_mean_30_60', 'clay_mean_60_100',
'ksat_mean',#'ksat_mean_0_5', 'ksat_mean_5_15', 'ksat_mean_15_30', 'ksat_mean_30_60', 'ksat_mean_60_100',
]
df_features = df[cst_vars+weather_vars]
print(df_features.columns)
feature_list=list(df_features.columns)
features=np.array(df_features)
# # TAKE SOME YEARS OUT OF TRAINING
# excl_years = [2019]
# features, labels = features[~df.year.isin(excl_years)], labels[~df.year.isin(excl_years)]
# RANDOM SPLIT
train_features, test_features, train_labels, test_labels = train_test_split(features,
labels,
test_size=0.20,
random_state = random_state,
shuffle = True)
# # YEAR SPLIT
# test_years = [2019]
# train_features, test_features, train_labels, test_labels = (features[~df.year.isin(test_years)], features[df.year.isin(test_years)],
# labels[~df.year.isin(test_years)], labels[df.year.isin(test_years)])
# Create a reference model to be tuned.
classifier_criterion='gini'
regressor_criterion='squared_error'
zir = ZeroInflatedRegressor(
classifier=RandomForestClassifier(random_state=random_state, criterion=classifier_criterion),
regressor=RandomForestRegressor(random_state=random_state, criterion=regressor_criterion)
)
def getTunedModel( baseModel, random_state ):
random_grid = {
'regressor__n_estimators': sp_randInt(10, 801),
'regressor__min_samples_leaf': sp_randInt(1, 51),
'regressor__max_depth': sp_randInt(10, 41),
'regressor__max_samples': sp_randFloat(),
'classifier__n_estimators': sp_randInt(10, 801),
'classifier__min_samples_leaf': sp_randInt(1, 51),
'classifier__max_depth': sp_randInt(10, 41),
'classifier__max_samples': sp_randFloat()
}
print(random_grid)
model_tuned = RandomizedSearchCV(cv=5, estimator = baseModel, param_distributions = random_grid, n_iter = 25, verbose=1, random_state=random_state , n_jobs = -1)
return model_tuned
zir_tuned = getTunedModel( zir, random_state )
# Run tuning to find optimal hyperparameters
zir_tuned.fit(train_features,train_labels)
## Select best parameters and fit model
result = pd.DataFrame.from_dict(zir_tuned.cv_results_)
# Choose the best hyperparameters from the random CV search.
best = result[result.rank_test_score == 1]
key = list(best.params.keys())[0]
best_params = dict(best.params)[key]
best_params
# Create a new model instance with the optimal hyperparameters.
zir_opt = ZeroInflatedRegressor(
classifier=RandomForestClassifier(
random_state=random_state, criterion=classifier_criterion,
n_estimators = best_params['classifier__n_estimators'],
max_samples = best_params['classifier__max_samples'],
min_samples_leaf = best_params['classifier__min_samples_leaf'],
max_depth = best_params['classifier__max_depth'],
),
regressor=RandomForestRegressor(
random_state=random_state, criterion=regressor_criterion,
n_estimators = best_params['regressor__n_estimators'],
max_samples = best_params['regressor__max_samples'],
min_samples_leaf = best_params['regressor__min_samples_leaf'],
max_depth = best_params['regressor__max_depth']
)
)
zir_opt.fit(features, labels)
y_pred = zir_opt.predict(features)
# Add performance metrics to the blurb output.
blurb = 'ZIR model (split train-test): 25-iter CV'
blurb = blurb + '\nRandom State: {0}'.format(random_state)
blurb = blurb + '\nGoodness of Fit (R2): {0}'.format(metrics.r2_score(labels, y_pred))
blurb = blurb + '\nMean Absolute Error (MAE): {0}'.format(metrics.mean_absolute_error(labels, y_pred))
blurb = blurb + '\nMean Squared Error (MSE): {0}'.format(metrics.mean_squared_error(labels, y_pred))
blurb = blurb + '\nRoot Mean Squared Error (RMSE): {0}'.format(np.sqrt(metrics.mean_squared_error(labels, y_pred)))
mape = np.mean(np.abs((labels - y_pred) / np.abs(labels+0.001)))
blurb = blurb + '\nMean Absolute Percentage Error (MAPE): {0}'.format(round(mape * 100, 2))
blurb = blurb + '\nAccuracy: {0}'.format(round(100*(1 - mape), 2))
print(blurb)
# Create unique filename for model run.
def get_file_id():
now = datetime.datetime.now(pytz.timezone('US/Pacific'))
fileid = '{0}-{1}-{2}-{3}-{4}-{5}'.format(str(now.year).zfill(4), str(now.month).zfill(2), str(now.day).zfill(2),
str(now.hour).zfill(2), str(now.minute).zfill(2), str(now.second).zfill(2))
return fileid
filename = 'ZIR-' + get_file_id()
savedir = '/project2/moyer/ag_data/prevented-planting/Models/ZIR/'+filename
# Create new directory for model run and save blurb.
os.mkdir(savedir)
with open(savedir+'/run_notes.txt', 'w') as f:
f.write(blurb)
f.write('\n\n')
for item in feature_list:
f.write("{0}\n".format(item))
with open(savedir+'/feature_list.pkl', 'wb') as f:
pickle.dump(feature_list, f)
# Save fitted model.
filepath = savedir+'/model.pkl'
pickle.dump(zir_opt, open(filepath, 'wb'))
# Predict historical data and add to dataset.
df['pred'] = zir_opt.predict(np.array(df[feature_list]))
df['pred_cl'] = zir_opt.classifier_.predict(np.array(df[feature_list])).astype(int)
df['pred_re'] = zir_opt.regressor_.predict(np.array(df[feature_list]))
# Save new dataframe with saved model.
df.to_csv(savedir+'/predictions-fldas.csv',index=False)