Intro

In [ ]:

import torch
import torchcrop
from torchcrop.utils.io import make_constant_weather

weather = make_constant_weather(batch_size=2, n_days=150)
model = torchcrop.Lintul5Model()
output = model(weather, start_doy=60)

print(output.yield_)  # [B] final storage-organ biomass (g m-2)
print(output.lai.shape)  # [B, T+1] LAI trajectory
print(output.dvs.shape)  # [B, T+1] development stage trajectory

import torch import torchcrop from torchcrop.utils.io import make_constant_weather weather = make_constant_weather(batch_size=2, n_days=150) model = torchcrop.Lintul5Model() output = model(weather, start_doy=60) print(output.yield_) # [B] final storage-organ biomass (g m-2) print(output.lai.shape) # [B, T+1] LAI trajectory print(output.dvs.shape) # [B, T+1] development stage trajectory

In [ ]:

import torch
import torch.nn as nn
from torchcrop import Lintul5Model, CropParameters

crop = CropParameters().to(dtype=torch.float64)
crop.tsum1 = nn.Parameter(torch.tensor(900, dtype=torch.float64))

model = Lintul5Model(crop_params=crop).double()
optimizer = torch.optim.Adam([crop.rue], lr=1e-1)

observed_yield = 1200

# Add gradient inspection
for i in range(50):
    optimizer.zero_grad()
    out = model(weather.to(torch.float64), start_doy=60)
    loss = ((out.yield_ - observed_yield) ** 2).mean()
    loss.backward()

    # Check gradient magnitude
    grad = crop.tsum1.grad
    print(f"Iter {i}: loss={loss:.6f}, yield={out.yield_.mean():.2f}, grad={grad:.6e}")

    optimizer.step()

import torch import torch.nn as nn from torchcrop import Lintul5Model, CropParameters crop = CropParameters().to(dtype=torch.float64) crop.tsum1 = nn.Parameter(torch.tensor(900, dtype=torch.float64)) model = Lintul5Model(crop_params=crop).double() optimizer = torch.optim.Adam([crop.rue], lr=1e-1) observed_yield = 1200 # Add gradient inspection for i in range(50): optimizer.zero_grad() out = model(weather.to(torch.float64), start_doy=60) loss = ((out.yield_ - observed_yield) ** 2).mean() loss.backward() # Check gradient magnitude grad = crop.tsum1.grad print(f"Iter {i}: loss={loss:.6f}, yield={out.yield_.mean():.2f}, grad={grad:.6e}") optimizer.step()