ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type float)

# ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type float)

The Tensorflow "ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type float)" occurs when you try to convert a Python list or a NumPy array that doesn't contain all float values to a Tensor.

To solve the error, convert the values in the list/array to float32 numbers before converting to a tensor.

Here is an example of how the error occurs.

main.py

Copied!
import numpy as np
import pandas as pd
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential

# ⛔️ Training data has non-float values
data = [
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
]

X_train = pd.DataFrame(data=data, columns=["x1", "x2", "x3"])
y_train = pd.DataFrame(data=[1, 0, 1], columns=["y"])


print(X_train)

print(X_train.dtypes)


model = Sequential()
model.add(
    Dense(1, input_dim=X_train.shape[1], activation='sigmoid')
)
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy']
)

# ⛔️ ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type float).
model.fit(X_train.to_numpy(), y_train, epochs=3)

Running the code sample produces the following output.

shell

Copied!
    x1   x2   x3
0  0.1  0.2  0.3
1  0.1  0.2  0.3
2  0.1  0.2  0.3

x1    float64
x2    float64
x3     object # 👈️ non-float value
dtype: object


⛔️ ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type float).

value error failed to convert numpy array to tensor unsupported object type float

Notice that the data variable doesn't store all float values.

This causes an error when we try to train the model with model.fit().

# Convert all values in the list to `float32`

One way to solve the error is to convert all values in the list/array to float32 using ndarray.astype.

main.py

Copied!
import numpy as np
import pandas as pd
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential

data = [
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
]

data = np.asarray(data).astype('float32')

X_train = pd.DataFrame(data=data, columns=["x1", "x2", "x3"])
y_train = pd.DataFrame(data=[1, 0, 1], columns=["y"])


print(X_train)

print(X_train.dtypes)


model = Sequential()
model.add(
    Dense(1, input_dim=X_train.shape[1], activation='sigmoid')
)
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy']
)

model.fit(X_train.to_numpy(), y_train, epochs=3)

The code for this article is available on GitHub

Running the code sample produces the following output.

shell

Copied!
    x1   x2   x3
0  0.1  0.2  0.3
1  0.1  0.2  0.3
2  0.1  0.2  0.3

x1    float32
x2    float32
x3    float32
dtype: object

The following line converts the values in the data list to float32.

main.py

Copied!
import numpy as np

data = np.asarray(data).astype('float32')

We could've also solved the error by converting the X_train DataFrame to a NumPy array with float32 values.

main.py

Copied!
import numpy as np
import pandas as pd
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential

data = [
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
]


X_train = pd.DataFrame(data=data, columns=["x1", "x2", "x3"])

# ✅ Convert to an Array with `float32` values
X_train = np.array(X_train).astype('float32')

y_train = pd.DataFrame(data=[1, 0, 1], columns=["y"])


model = Sequential()
model.add(
    Dense(1, input_dim=X_train.shape[1], activation='sigmoid')
)
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy']
)

model.fit(X_train, y_train, epochs=3)

The code for this article is available on GitHub

The following line is important:

main.py

Copied!
X_train = np.array(X_train).astype('float32')

You can also use the following line if you don't want to hardcode the type.

main.py

Copied!
X_train = np.asarray(X_train).astype(np.float32)

Notice that we are no longer calling X_train.to_numpy() in the call to model.fit() because we already converted the DataFrame to an array.

Note that you should be using NumPy arrays as inputs and not native Python lists.

Keras/TensorFlow doesn't support using native Python lists as inputs.

You can convert a Python list to a NumPy array by using np.asarray().

main.py

Copied!
x_array = np.asarray(x_list)

If the error persists, try to also convert y_train to a NumPy array of floating-point numbers.

main.py

Copied!
import numpy as np
import pandas as pd
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential

data = [
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
]


X_train = pd.DataFrame(data=data, columns=["x1", "x2", "x3"])

# 1) Convert X_train to NumPy array of floats ✅
X_train = np.asarray(X_train).astype(np.float32)

y_train = pd.DataFrame(data=[1, 0, 1], columns=["y"])

# 1) Convert y_train to NumPy array of floats ✅
y_train = np.asarray(y_train).astype(np.float32)

model = Sequential()
model.add(
    Dense(1, input_dim=X_train.shape[1], activation='sigmoid')
)
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy']
)

model.fit(X_train, y_train, epochs=3)

The code for this article is available on GitHub

# You might also get the error if you have missing values in a DataFrame column

You might also get the error if you have missing values in a numeric or a string DataFrame column.

Here is an example.

main.py

Copied!
import numpy as np
import pandas as pd
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential

# 👇️ Has missing values
data = [
    [0.1, 0.2, None],
    [0.1, 0.2, None],
    [0.1, 0.2, None],
]


X_train = pd.DataFrame(data=data, columns=["x1", "x2", "x3"])

y_train = pd.DataFrame(data=[1, 0, 1], columns=["y"])

model = Sequential()
model.add(
    Dense(1, input_dim=X_train.shape[1], activation='sigmoid')
)
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy']
)

# ⛔️ ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type float).
model.fit(X_train, y_train, epochs=3)

Note that we have some missing values in the data variable.

This can be caused by None, np.nan values, etc.

You can use the DataFrame.fillna() method to solve the error in this case.

main.py

Copied!
import numpy as np
import pandas as pd
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential

# 👇️ Has missing values
data = [
    [0.1, 0.2, None],
    [0.1, 0.2, None],
    [0.1, 0.2, None],
]


X_train = pd.DataFrame(data=data, columns=["x1", "x2", "x3"])

# ✅ Fill missing values using fillna()
X_train.fillna(value=0, inplace=True)

y_train = pd.DataFrame(data=[1, 0, 1], columns=["y"])

model = Sequential()
model.add(
    Dense(1, input_dim=X_train.shape[1], activation='sigmoid')
)
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy']
)

model.fit(X_train, y_train, epochs=3)

The code for this article is available on GitHub

The code sample uses the DataFrame.fillna() method to fill the missing values.

main.py

Copied!
# ✅ Fill missing values using fillna()
X_train.fillna(value=0, inplace=True)

If you have a string column that has missing values, you would likely want to set the value argument to an empty string.

main.py

Copied!
# ✅ Fill missing values using fillna()
df.fillna(value='', inplace=True)

You can also call the fillna() method on the specific column.

main.py

Copied!
# for a string column
df['YourColumn'].fillna(value="", inplace=True)

# or for a numeric column
df['YourColumn'].fillna(value=0, inplace=True)

If you want to call fillna() on all string columns in the DataFrame, use a for loop.

main.py

Copied!
cols = df.select_dtypes(include=['object'])

for column in cols.columns.values:
    # for string values
    df[column] = df[column].fillna('')

    # or for numeric values
    # df[column] = df[column].fillna(0)

# Using `tf.convert_to_tensor()` to solve the error

You can also use the tensorflow.convert_to_tensor() method to solve the error.

main.py

Copied!
import numpy as np
import pandas as pd
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential
import tensorflow as tf

data = [
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
]
X_train = pd.DataFrame(data=data, columns=["x1", "x2", "x3"])
y_train = pd.DataFrame(data=[1, 0, 1], columns=["y"])


model = Sequential()
model.add(
    Dense(1, input_dim=X_train.shape[1], activation='sigmoid')
)
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy']
)

# 👇️ Using tf.convert_to_tensor
model.fit(
    tf.convert_to_tensor(X_train.to_numpy(), dtype=tf.float32),
    y_train,
    epochs=3
)

The code for this article is available on GitHub

The tf.convert_to-tensor() method converts a given value to a Tensor.

The method takes a dtype argument that can be set to tf.float32.

The dtype argument specifies the element type for the returned tensor.

If you don't explicitly set the dtype argument, the type is inferred from the type of the supplied value.

You could also convert the X_train and y_train variables to tensors using convert_to_tensor and pass them directly to model.fit().

main.py

Copied!
import numpy as np
import pandas as pd
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential
import tensorflow as tf

data = [
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
    [0.1, 0.2, '0.3'],
]
X_train = pd.DataFrame(data=data, columns=["x1", "x2", "x3"])

# ✅ Convert X_train to tensor
X_train = tf.convert_to_tensor(X_train, dtype=tf.float32)

y_train = pd.DataFrame(data=[1, 0, 1], columns=["y"])

# ✅ Convert y_train to tensor
y_train = tf.convert_to_tensor(y_train, dtype=tf.float32)


model = Sequential()
model.add(
    Dense(1, input_dim=X_train.shape[1], activation='sigmoid')
)
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy']
)
model.fit(
    X_train,
    y_train,
    epochs=3
)

# ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type numpy.ndarray)

If you get the error "ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type numpy.ndarray)", you likely passed an array of arrays to model.fit().

Here is an example of how the error occurs.

main.py

Copied!
import numpy as np
import pandas as pd
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential
import tensorflow as tf

data = [
    [np.asarray([0.1]), np.asarray([0.2]), np.asarray([0.3])],
    [np.asarray([0.1]), np.asarray([0.2]), np.asarray([0.3])],
    [np.asarray([0.1]), np.asarray([0.2]), np.asarray([0.3])],
]
X_train = pd.DataFrame(data=data, columns=["x1", "x2", "x3"])

y_train = pd.DataFrame(data=[1, 0, 1], columns=["y"])

print(X_train)

print(X_train.dtypes)


model = Sequential()
model.add(
    Dense(1, input_dim=X_train.shape[1], activation='sigmoid')
)
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy']
)

# ⛔️ ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type numpy.ndarray).
model.fit(
    X_train.to_numpy(),
    y_train,
    epochs=3
)

Running the code sample produces the following output.

shell

Copied!
      x1     x2     x3
0  [0.1]  [0.2]  [0.3]
1  [0.1]  [0.2]  [0.3]
2  [0.1]  [0.2]  [0.3]

x1    object
x2    object
x3    object
dtype: object

ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type numpy.ndarray).

One way to solve the error is to flatten each array in the DataFrame.

main.py

Copied!
import numpy as np
import pandas as pd
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential
import tensorflow as tf

data = [
    [np.asarray([0.1]), np.asarray([0.2]), np.asarray([0.3])],
    [np.asarray([0.1]), np.asarray([0.2]), np.asarray([0.3])],
    [np.asarray([0.1]), np.asarray([0.2]), np.asarray([0.3])],
]
X_train = pd.DataFrame(data=data, columns=["x1", "x2", "x3"])

# ✅ Flatten each array in the DataFrame
X_train = X_train.apply(
    lambda x: pd.Series(pd.core.common.flatten(x))
)

y_train = pd.DataFrame(data=[1, 0, 1], columns=["y"])

print(X_train)

print(X_train.dtypes)


model = Sequential()
model.add(
    Dense(1, input_dim=X_train.shape[1], activation='sigmoid')
)
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy']
)

model.fit(
    X_train.to_numpy(),
    y_train,
    epochs=3
)

The code for this article is available on GitHub

# Additional Resources

You can learn more about the related topics by checking out the following tutorials:

I wrote a book in which I share everything I know about how to become a better, more efficient programmer.

You can use the search field on my Home Page to filter through all of my articles.

ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type float)