def trainTwoWords(trainHSV=False):
print "Gathering labels..."
labels = getWordLists(method=sortAvail, checkRedundancy = False)
shuffle(labels)
training = []
targets = []
print "Converting to vectors..."
for x, y, z in labels:
training.append(convertToVector(x) + convertToVector(y))
targets.append(convertToVector(z, True))
training_data = training[:len(training) * 7 / 10]
training_targets = targets[:len(targets) * 7 / 10]
training_labels = labels[:len(labels) * 7 / 10]
print len(training_data)
print len(training_data[0])
test_data = [x for x in training if x not in training_data]
test_targets = [x for x in targets if x not in training_targets]
test_labels = [x[2] for x in labels if x not in training_labels]
print len(test_data)
print "Fitting data..."
if not trainHSV:
clf = RandomForestRegressor(n_jobs = -1)
clf.fit(training_data, training_targets)
print "Predicting distributions..."
test_data_results = clf.predict(test_data)
else: #go through everything and make a separate tree for each 'zone' in the model
clf = []
print "Fitting Initial...."
initial_clf = RandomForestRegressor(n_jobs = -1)
initial_clf.fit(training_data, [x[0:3] for x in training_targets]) #replace this with something else later
hue_clf = RandomForestRegressor(n_jobs = -1)
print "Fitting Hue..."
hue_clf.fit([x[2:14] + x[40:52] for x in training_data], [x[3:15] for x in training_targets])
saturation_clf = RandomForestRegressor(n_jobs = -1)
print "Fitting Saturation..."
saturation_clf.fit([x[14:26] + x[52:64] for x in training_data], [x[15:27] for x in training_targets])
value_clf = RandomForestRegressor(n_jobs = -1)
print "Fitting Value..."
value_clf.fit([x[26:38] + x[64:76] for x in training_data], [x[27:39] for x in training_targets])
clf.append(initial_clf)
clf.append(hue_clf)
clf.append(saturation_clf)
clf.append(value_clf)
print "Predicting distributions..."
initial_results = initial_clf.predict(test_data)
hue_results = hue_clf.predict([x[2:14] + x[40:52] for x in test_data])
saturation_results = saturation_clf.predict([x[14:26] + x[52:64] for x in test_data])
value_results = value_clf.predict([x[26:38] + x[64:76] for x in test_data])
#print initial_results
#print hue_results
#print saturation_results
#print value_results
#print len(initial_results)
test_data_results = [np.hstack((w, x, y, z)) for w, x, y, z in zip(initial_results, hue_results, saturation_results, value_results)]
#print len(test_data_results)
#print test_data_results
return test_data, test_data_results, test_labels, test_targets, clf
max_features='auto',
max_leaf_nodes=None,
min_impurity_decrease=0.0,
min_impurity_split=None,
min_samples_leaf=3,
min_samples_split=5,
min_weight_fraction_leaf=0.0,
n_estimators=400,
n_jobs=-1,
oob_score=False,
random_state=None,
verbose=0,
warm_start=False,
)
regr.fit(X_train_train, y_train_train)
base_model.append(regr)
y_pred = regr.predict(X_train_val)
df999 = pd.DataFrame({"y_val": y_train_val, "y_pred": y_pred})
df_result = pd.concat([df_result, df999], axis=0)
# validation dataによる評価指標の算出
y_val = df_result["y_val"]
y_pred = df_result["y_pred"]
mse = mean_squared_error(y_val, y_pred)
mae = mean_absolute_error(y_val, y_pred)
print(
"**** Training set score( {} ): MSE={:.3f} RMSE={:.3f} MAE={:.3f} Score={:.3f} ****"
.format(i, round(mse, 3), round(np.sqrt(mse), 3), round(mae, 3),
regr.score(X_train, y_train)))
# In[50]:
