diff options
Diffstat (limited to 'train.py')
| -rw-r--r-- | train.py | 12 |
1 files changed, 7 insertions, 5 deletions
@@ -8,7 +8,7 @@ THETAS_FILE = "thetas.csv" def normalize(data): min_val = min(data) max_val = max(data) - return [(x - min_val) / (max_val - min_val) for x in data], min_val, max_val + return [(x - min_val) / (max_val - min_val) for x in data] def load_data(): @@ -28,7 +28,7 @@ def estimate_price(mileage, theta0, theta1): # DV: dependant variable, IV: independant variable -def train_once(learning_rate, DV, IV, theta0, theta1): +def compute_gradients(learning_rate, DV, IV, theta0, theta1): tmp0 = ( learning_rate * (1.0 / len(DV)) @@ -52,12 +52,14 @@ def denormalize_thetas(t0, t1, km_min, km_max, price_min, price_max): def train(learning_rate, iterations): kms, prices = load_data() - kms_norm, km_min, km_max = normalize(kms) - prices_norm, price_min, price_max = normalize(prices) + km_min, km_max = min(kms), max(kms) + price_min, price_max = min(prices), max(prices) + kms_norm = normalize(kms) + prices_norm = normalize(prices) t0 = 0.0 t1 = 0.0 for _ in range(iterations): - grad0, grad1 = train_once(learning_rate, prices_norm, kms_norm, t0, t1) + grad0, grad1 = compute_gradients(learning_rate, prices_norm, kms_norm, t0, t1) t0 -= grad0 t1 -= grad1 return denormalize_thetas(t0, t1, km_min, km_max, price_min, price_max) |