Compare commits

..

No commits in common. "e49246ac2d51e41ed5c041f9f20d59681866d3c1" and "4c9f7d8d7d973df8ea1b896c84bbf9dbe896c82a" have entirely different histories.

11 changed files with 16 additions and 409 deletions

View File

@ -1,28 +0,0 @@
# Base Image
FROM nvidia/cuda:11.3.1-cudnn8-runtime-ubuntu20.04
# System Dependencies and Cleanup
RUN apt-get update -y && \
DEBIAN_FRONTEND=noninteractive TZ=Etc/UTC apt-get install -y tzdata && \
apt-get install -y software-properties-common ffmpeg libsm6 libxext6 libhdf5-serial-dev netcdf-bin libnetcdf-dev && \
add-apt-repository ppa:ubuntugis/ubuntugis-unstable && \
apt-get update && \
apt-get install -y curl build-essential gdal-bin libgdal-dev libpq-dev python3-gdal python3-pip apt-transport-https ca-certificates gnupg && \
apt-get clean && \
rm -rf /var/lib/apt/lists/*
# Copy only the necessary files
COPY requirements.txt /micrograd/requirements.txt
# Install Python Packages
RUN pip install --no-cache-dir -r /micrograd/requirements.txt
# Set Working Directory and Prepare App
WORKDIR /micrograd
COPY micrograd /micrograd/micrograd
COPY test /micrograd/test
COPY app.py /micrograd/app.py
RUN mkdir -p /root/.cache/torch/hub/checkpoints/
CMD ["python3", "app.py"]

View File

@ -1,16 +0,0 @@
NAME=micrograd
DOCKER=podman
PORT=8008
.PHONY=build run stop
all: stop rm build run
stop:
${DOCKER} stop ${NAME}
rm:
${DOCKER} rm ${NAME}
run:
${DOCKER} run -p ${PORT}:${PORT} -d --name ${NAME} -t ${NAME}
build:
${DOCKER} build . --tag ${NAME}

30
app.py
View File

@ -1,30 +0,0 @@
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from micrograd.train import model # Replace with your ML model library
from typing import List
# Create FastAPI app
app = FastAPI()
# Load your ML model (replace this with your model loading logic)
# model = engine.load_model("path/to/your/model")
# Define a request model
class Item(BaseModel):
data: List[float] # Change 'list' to match the input format of your model
# Endpoint for ML inference
@app.post("/predict")
async def predict(item: Item):
try:
# Perform prediction (modify this part according to your model's API)
nn = model()
prediction = nn(item.data)
return {"prediction": prediction.data}
except Exception as e:
raise HTTPException(status_code=500, detail=str(e))
# Run the API with Uvicorn
if __name__ == "__main__":
import uvicorn
uvicorn.run(app, host="0.0.0.0", port=8008)

View File

@ -1,18 +0,0 @@
{ pkgs ? import <nixpkgs> {} }:
pkgs.mkShell {
buildInputs = [
pkgs.python310
pkgs.python310Packages.numpy
pkgs.python310Packages.pytest
pkgs.python310Packages.uvicorn
pkgs.python310Packages.fastapi
];
# Set PYTHONPATH to include your Python dependencies.
# This is useful if you are using Python modules that are not installed in the
# standard location.
# shellHook = ''
# export PYTHONPATH=${pkgs.python3.sitePackages}:$PYTHONPATH
# '';
}

View File

@ -1,7 +1,4 @@
class Value: class Value:
"""
add a comment
"""
def __init__(self, data, _children=(), _op='', label=''): def __init__(self, data, _children=(), _op='', label=''):
self.data = data self.data = data
@ -33,7 +30,6 @@ class Value:
return f"Value(data={self.data})" return f"Value(data={self.data})"
def __add__(self, other): def __add__(self, other):
other = other if isinstance(other, Value) else Value(other)
out = Value(data=self.data + other.data, out = Value(data=self.data + other.data,
_children=(self, other), _op='+') _children=(self, other), _op='+')
@ -42,37 +38,13 @@ class Value:
other.grad += 1.0 * out.grad other.grad += 1.0 * out.grad
out._backward = _backward out._backward = _backward
return out return out
def __radd__(self, other):
return self + other
def __pow__(self, other):
assert isinstance(other, (int, float)), "int or float for now"
other = Value(other)
n = other.data
out = Value(data=self.data ** n,
_children=(self, other), _op=f'**{n}')
def _backward():
self.grad += n * (self.data ** (n-1)) * out.grad
out._backward = _backward
return out
def exp(self):
from math import exp
x = self.data
out = Value(data=exp(x),
_children=(self,), _op='exp')
def _backward():
self.grad += out.data * out.grad
out._backward = _backward
return out
def __sub__(self, other): def __sub__(self, other):
return self + (-1 * other) out = Value(data=self.data - other.data,
_children=(self, other), _op='-')
return out
def __mul__(self, other): def __mul__(self, other):
other = other if isinstance(other, Value) else Value(other)
out = Value(data=self.data * other.data, out = Value(data=self.data * other.data,
_children=(self, other), _op='*') _children=(self, other), _op='*')
@ -82,12 +54,6 @@ class Value:
out._backward = _backward out._backward = _backward
return out return out
def __rmul__(self, other):
return self * other
def __truediv__(self, other):
return self * other ** -1
def tanh(self): def tanh(self):
from math import exp from math import exp
t = (exp(2 * self.data) - 1) / (exp(2 * self.data) + 1) t = (exp(2 * self.data) - 1) / (exp(2 * self.data) + 1)

View File

@ -1,66 +0,0 @@
import random
from .engine import Value
from typing import List, Union
class Module:
def zero_grad(self):
for p in self.parameters():
p.grad = 0.0
def parameters(self):
return []
class Neuron(Module):
def __init__(self, n_in: int):
self.w = [Value(random.uniform(-1, 1)) for _ in range(n_in)]
self.b = Value(random.uniform(-1, 1))
def __call__(self, x) -> Value:
# w * x + b
from functools import reduce
from operator import add
if len(x) != len(self.w):
raise ValueError(f"mismatch dimension: x: {len(x)}, w: {len(self.w)}")
act = reduce(add, [w_i * x_i for w_i, x_i in zip(self.w, x)]) + self.b
return act.tanh()
def parameters(self):
return self.w + [self.b]
class Layer(Module):
def __init__(self, n_in: int, n_out: int):
self.neurons = [Neuron(n_in) for _ in range(n_out)]
def __call__(self, x) -> Union[List[Value], Value]:
out = [n(x) for n in self.neurons]
return out[0] if len(out) == 1 else out
def parameters(self):
out = []
for neuron in self.neurons:
for p in neuron.parameters():
out.append(p)
return out
class MLP(Module):
def __init__(self, n_in: int, n_outs: List[int]):
sizes = [n_in] + n_outs
self.layers = []
for i in range(len(n_outs)):
self.layers.append(Layer(sizes[i], sizes[i+1]))
def __call__(self, x):
for layer in self.layers:
x = layer(x)
return x
def parameters(self):
out = []
for layer in self.layers:
for l in layer.parameters():
out.append(l)
return out

View File

@ -1,35 +0,0 @@
from micrograd.nn import MLP
def model():
n = MLP(3, [4, 4, 1])
xs = [
[2.0, 3.0, -1.0],
[3.0, -1.0, 0.5],
[0.5, 1.0, 1.0],
[1.0, 1.0, 1.0],
]
y_true = [1.0, -1.0, -1.0, 1.0]
epochs = 20
lr = 0.08
for epoch in range(epochs):
# forward
y_pred = [n(x) for x in xs]
loss = sum([(y_p - y_t)**2 for y_p, y_t in zip(y_pred, y_true)])
# backward
n.zero_grad()
loss.backward()
# update (sgd)
for p in n.parameters():
p.data += -lr * p.grad
print(f"{epoch=:02d} | loss: {loss.data:4f}")
print(f"{[y.data for y in y_pred]}")
print(f"{[y for y in y_true]}")
return n
if __name__ == "__main__":
model()

View File

@ -1,5 +1,3 @@
pytest pytest
notebook notebook
graphviz graphviz
fastapi
uvicorn

View File

@ -2,6 +2,16 @@ import pytest
from micrograd.engine import Value from micrograd.engine import Value
def test_backward_tanh():
# inputs
x = Value(0.8814)
y = x.tanh()
y.grad = 1.0
y._backward()
assert pytest.approx(x.grad, 0.1) == 0.5
def test_large_backprop(): def test_large_backprop():
# inputs # inputs
x1 = Value(2.0, label='x1') x1 = Value(2.0, label='x1')
@ -97,80 +107,3 @@ def test_accumulation():
b = a + a b = a + a
b.backward() b.backward()
assert a.grad == 2.0 assert a.grad == 2.0
def test_backward_tanh():
# inputs
x = Value(0.8814)
y = x.tanh()
y.backward()
assert pytest.approx(x.grad, 0.1) == 0.5
def test_backward_exp():
# inputs
x = Value(1.0)
y = x.exp()
y.backward()
assert pytest.approx(x.grad, 0.1) == 2.7
def test_backward_pow():
# inputs
x = Value(1.0)
y = x ** 2
y.backward()
assert x.grad == 2.0
def test_backward_div():
a = Value(4.0)
b = Value(2.0)
c = a / b
c.backward()
assert a.grad == 0.5
def test_auto_diff_replace_tan_with_exp():
# inputs
x1 = Value(2.0, label='x1')
x2 = Value(0.0, label='x2')
# weights
w1 = Value(-3.0, label='w1')
w2 = Value(1.0, label='w2')
# bias
b = Value(6.8813735870195432, label='b')
h1 = x1 * w1
h1.label = 'h1'
h2 = x2 * w2
h2.label = 'h2'
h = h1 + h2
h.label = 'h'
n = h + b
n.label = 'n'
e = (2*n).exp()
y = (e - 1) / (e + 1)
y.label = 'y'
y.backward()
assert pytest.approx(n.grad, 0.001) == 0.5
assert pytest.approx(b.grad, 0.001) == 0.5
assert pytest.approx(h.grad, 0.001) == 0.5
assert pytest.approx(h1.grad, 0.001) == 0.5
assert pytest.approx(h2.grad, 0.001) == 0.5
assert pytest.approx(x1.grad, 0.001) == -1.5
assert pytest.approx(w1.grad, 0.001) == 1.0
assert pytest.approx(x2.grad, 0.001) == 0.5
assert pytest.approx(w2.grad, 0.001) == 0.0

View File

@ -1,66 +0,0 @@
from micrograd.nn import Neuron, Layer, MLP
from micrograd.engine import Value
import pytest
def test_init_neuron():
inputs = 2
x = [1.0, 0.0]
n = Neuron(inputs)
assert len(n.w) == inputs
y = n(x)
assert isinstance(y, Value)
def test_mismatch_number():
with pytest.raises(ValueError):
x = [1.0, 0.0]
n = Neuron(7)
n(x)
def test_large_n_in():
n_in = 100
x = [1.0] * n_in
n = Neuron(n_in)
y = n(x)
assert isinstance(y, Value)
def test_well_known_weights():
x = [1.0, 0.0]
w = [Value(0.0), Value(0.0)]
b = Value(0.0)
n = Neuron(2)
n.w = w
n.b = b
y = n(x)
assert y.data == sum([x[0] * w[0], x[1] * w[1], b]).tanh().data
def test_mlp():
x = [2.0, 3.0, -1.0]
n = MLP(3, [4, 4, 4])
y = n(x)
assert len(y) == 4
def test_mlp_single_out():
x = [2.0, 3.0, -1.0]
n = MLP(3, [4, 4, 1])
y = n(x)
assert isinstance(y, Value)
def test_sample_mlp():
n = MLP(3, [4, 4, 1])
xs = [
[2.0, 3.0, -1.0],
[3.0, -1.0, 0.5],
[0.5, 1.0, 1.0],
[1.0, 1.0, 1.0],
]
y_true = [1.0, -1.0, -1.0, 1.0]
y_pred = [n(x) for x in xs]
mse = sum([(y_p - y_t)**2 for y_p, y_t in zip(y_pred, y_true)])
def test_mlp_parameters():
n = MLP(3, [4, 4, 1])
assert len(n.parameters()) == 41

View File

@ -1,4 +1,3 @@
import pytest
from micrograd.engine import Value from micrograd.engine import Value
@ -12,21 +11,21 @@ def test_value_repr():
assert "Value(data=2.0)" == repr(v) assert "Value(data=2.0)" == repr(v)
def test_value_add_opt(): def test_value_add():
v1 = Value(2.0) v1 = Value(2.0)
v2 = Value(4.0) v2 = Value(4.0)
assert (v1 + v2).data == 6.0 assert (v1 + v2).data == 6.0
assert "Value(data=6.0)" == repr(v1 + v2) assert "Value(data=6.0)" == repr(v1 + v2)
def test_value_sub_opt(): def test_value_sub():
v1 = Value(2.0) v1 = Value(2.0)
v2 = Value(4.0) v2 = Value(4.0)
assert (v1 - v2).data == -2.0 assert (v1 - v2).data == -2.0
assert "Value(data=-2.0)" == repr(v1 - v2) assert "Value(data=-2.0)" == repr(v1 - v2)
def test_value_mul_opt(): def test_value_mul():
v1 = Value(2.0) v1 = Value(2.0)
v2 = Value(4.0) v2 = Value(4.0)
v3 = Value(-1.0) v3 = Value(-1.0)
@ -34,36 +33,6 @@ def test_value_mul_opt():
assert (v1 * v3).data == -2.0 assert (v1 * v3).data == -2.0
def test_value_rmul_opt():
a = Value(2.0)
b = 2 * a
assert b.data == 4.0
def test_value_pow_opt():
a = Value(2.0)
b = a ** 2
assert b.data == 4.0
def test_value_exp_opt():
a = Value(1.0)
b = a.exp()
assert pytest.approx(b.data, 0.1) == 2.7
def test_value_int_opt():
a = Value(2.0)
b = a - 1
assert b.data == 1.0
def test_value_div_opt():
a = Value(2.0)
b = a / 2
assert b.data == 1.0
def test_value_mul_add(): def test_value_mul_add():
v1 = Value(2.0) v1 = Value(2.0)
v2 = Value(4.0) v2 = Value(4.0)